Map Happenings

Well, I’ve been a little bit busy recently.

Yours truly just made a giant update to the F1 maps I created for my article back in August 2024.

It’s now completely revamped to be a fully interactive 3D mapping experience complete with maps of all the F1 tracks, maps of all the F1 team factories and tons and tons of stats.

You can check it out at maphappenings.com/f1 or by selecting “F1 Map” from the Map Happenings site menu bar.

Here’s a sneak peak:

The map was built using the mapping platform Elemap, powered by Elebase. Elemap provides a super easy way to publish interactive maps and data without any need to code or be a mapping geek. It’s perfect for non technical folks to create some really absorbing content. You can think of it like Squarespace, only for interactive maps!

My particular thanks to John Frandsen and his team for making Elemap available to me for this very fun project!

Tips:

• On a Mac the map provides the most 3D detail on Chrome (sorry to Safari aficionados)
• Click/tap on “List” (top left) to get to the list of all the races and teams
• Click/tap any item in the list to zoom in. 
• Click/tap any corner of a race circuit to get the turn number (and name if available).
• Don’t forget to try out the 3D button … it’s so cool (check out Round 22: Las Vegas)
• To spin the map on an iPad or tablet: use your fingers.
• To spin the map on a Mac or PC: hold down the control key while using the keypad/mouse.

You can send comments about the map directly to me at contact@maphappenings.com.

Enjoy!

Ok, it’s that time of year isn’t it? When people in the US celebrate something called “Thanksgiving” while everyone else celebrates… well, let’s not get into that.

And immediately afterwards of course it’s all about commerce. Money has to be spent. Wallets have to be emptied. And credit cards have to be maxed out.

How could Map Happenings possibly not do its part?

So without further ado, and in no particular order, please gorge on this rather delicious list of fourteen ‘mappy’ books that you, your friends or your family might slurp up.

These mapping gems range from amusing, to epic, to colourful, to insightful, to quaint, to just pure ‘must have’.

I didn’t bother with any affiliate links, so don’t worry, no enrichment of yours truly will take place if you happen to buy something.

Terrible Maps

Maps on Vinyl

This Way Up: When Maps Go Wrong

Longitude

The Atlas of Unusual Borders

Our Dumb World by The Onion

At Atlas of Extinct Countries

A History of the World in 12 Maps

How to Lie with Maps

Transit Maps of the World

Utterly British Maps

Great City Maps

Brilliant Maps for Curious Minds

The Rand McNally Road Atlas!

As many of you may know, my career started back in 1985 at a wee company called Etak. This startup, seed funded by Nolan Bushnell, was most famous for pioneering in-vehicle navigation systems. It was about 20 years ahead of its time.

But it didn’t stop there. Navigation systems require digital maps. The problem was that there weren’t any available at that time, so Etak had to invent a system to manufacture them at scale. And yours truly was privileged to be part of the team that designed and built the production line.

It wasn’t straightforward. Using a VAX minicomputer with 2MB RAM the team designed a system that was used to scan photographic images of topographic maps. These were then manually digitised on cobbled together PC-clones that used very expensive state-of-the-art graphics cards.

Despite the herculean efforts of everybody involved there was a little problem. The accuracy of the maps depended on the source material. In our case we relied on topographic maps published by national mapping agencies. Alas, these maps were often years out of date.

Don’t forget this was 1985. Aerial imagery and satellite imagery just wasn’t readily available back then. We could (and did) contact local agencies for more up-to-date maps of critical intersections, but this was really hard to do at scale. It was just too time consuming and too expensive to track all the material down.

The original Etak Navigator didn’t provide turn-by-turn directions. Instead it guided you to your location by a flashing star on the map. You zoomed the map as you drove and had to use your own noggin to figure out which roads to take to get there. As a result the requirements for the digital map were incredibly light: we didn’t need to collect information on one way systems or turn restrictions. We just collected the streets, the street names and the addresses and we made sure we got the road topology right.

It was only later when the enterprise customers like Bosch and GM wanted real-time guidance that we had to collect more information like one ways and turn restrictions. And it was so painful!

Fast forward a decade to 1995 and satellite imagery and low altitude aerial photography became more readily available, so the job of collecting the basic street network got easier. But the road attributes were still hard. Digital mapping companies had to send out crews in cars, with one person driving and another person collecting data using pen, paper and a clipboard. Ouch!

It was natural to start looking at technology to collect this information automatically. Indeed in the 1990s mapping companies started mounting cameras on vehicles. But it wasn’t until 2007 when Google launched Street View that it became common enough that the general public started to notice.

People often got excited to see a Google Maps car on their street — as well cars from rival fleets such as TomTom, HERE and later Apple Maps.

These fleets, each in their hundreds, drive around collecting all manner of data, not just the photographic images, but also a ton of information deduced from the images: for example street names, one ways, turn-restrictions, lane information, stop signs, traffic lights, speed limits to name just a few.

But there’s a problem.

Running a fleet of 100+ dedicated vehicles solely for map data collection is expensive.

And then there’s an even bigger problem: that 100+ vehicle fleet simply can’t collect everything, everywhere, all at once.

Even Google Maps can’t do this. It’s evident from a typical Google Street View image: if you look closely you can see the image capture date. Take the example image below: in this case it’s from October 2024, so now it’s over a year old¹:

This issue of trying to keep volatile map data current became abundantly clear to me in 2019 when Apple Maps became the first consumer mapping product to display stop signs and traffic lights on a map.

I remember thinking: “Do you realise what you’ve just done? Do you realise what it’s going to take to maintain all that data? Do you realise just how often this data changes?!”

And it didn’t stop there. Speed limits were added to the map too. And that data is even more volatile…

But it gets worse.

Let’s move on to a product whose mapping needs go way beyond that of a consumer navigation app: autonomous vehicles.

With very few exceptions these vehicles require a map an order of magnitude more detailed. And the mapping industry has given this very detailed map a name: they call it an “HD Map”. It has not only a ton more data, but centimetre level accuracy too.

Here’s what a typical HD map looks like:

As you can see it’s capturing gobs of detail that you don’t typically see in your everyday Google Map or Apple Map.

The companies developing autonomous vehicles rely on HD Maps to varying degrees. Waymo and GM/Cruise are the biggest users. If you’re interested, you can nerd out on the details in this table (click/tap to embiggen):

So what on earth do you do?

First you could try collecting all the necessary data yourself and try to keep it up to date. This is what Waymo does. It may go a long way to explain the astronomical cost for their service:

Alternatively you could take another approach. And that’s where this week’s “Mapping Amazing” company, Bee Maps, comes in:

Now I should make it clear from the start that Bee Maps isn’t trying to do everything, but they are doing a hell of a lot. And it’s not just the technology that’s interesting. It’s their unique business model for accomplishing it.

So let’s dive in…

Founded by Ariel Seidman², this 45 person company based in San Francisco got its start 10 years ago in 2015. Their first product was a platform called Hivemapper. Rather bravely it required the development of hardware as well as software.³

Why hardware?

Because you can’t just write a map data capture app for a smartphone and expect it to work in the real world. In vehicles phones overheat — quickly and easily. Also, they don’t have accurate enough GPS. Besides which in order to capture map data continuously you need a dedicated device, so it would never be practical to use a personal phone.

Bee’s hardware is now on its third generation⁴. The device is about the size of a small hardcover book. It takes about five minutes to install and get running. After that you never have to do anything. No app to start up. No button to press. No firmware to update. It’s completely “fire and forget”.

The device has a main camera, stereo depth cameras and, crucially, a high precision GPS. And it has built-in LTE for uploading data anywhere around the globe:

The original goal of Bee Maps was to gamify the collection of map data. The incentive? As you drive and collect data you earn currency: in this case a cryptocurrency called HONEY. And the Hivemapper network is built in such a way that it can be used to incentivise users to collect data in particular locations. They call these HONEY Bursts.

The idea was that an individual who was driving all hours for their job (e.g. an Uber driver) would opt to install a Bee Maps device in their vehicle, and use it to earn some extra dosh on top of the money they were earning for their main job.

And the customers for all this data the Bee Maps devices collect? The traditional map makers of course! And it must make them salivate when they see the ever growing list of data being sucked up:

• Signs for one way, speed limit, stop, yield, turn restriction, highway exit
• Signs for parking restriction, height restriction, rail crossing
• Traffic lights
• Road widths and lanes
• Road construction
• Fire hydrants

In the US they’ve even started to collect speed cameras, gas prices and toll prices.

Bee Maps’ business model has pivoted a little since they were founded. It’s now a subscription model: starting at US$19 per month you get access to a Bee device. And yes, you still earn HONEY cryptocurrency by driving around.

With a Bee device you get all the benefits of a traditional dashboard camera: continuous video to be used in case of accident reports and telemetry to be used for car insurance discounts.

Commercial fleets get even more: monitoring of all trips and events, including any exciting driving. Here’s a snapshot of the Bee Maps dashboard proving yours truly is a great supporter of exciting driving himself:

In effect Bee Maps is now competing with the likes of Garmin for consumer dash cams and the likes of Samsara and Lytx for fleet dash cams.

But they have a huge edge.

Unlike their competitors, Bee Maps is earning an ever growing revenue stream from the map data they collect. And they’re already licensing their map data to some very significant customers. For example: TomTom, HERE, Mapbox, Lyft, Trimble, MAXAR, and more recently, VW.

All this has attracted new investors: Bee Maps closed a US$32M Series A last month.

It’s not only the wealth of data that Bee Maps is capturing that’s impressive, it’s their coverage too:

• 22M unique kilometres mapped (over 13M miles) ~= 3X road network in US.
• 36% of all roads around the whole globe.
• 665M km mapped total (meaning each road has been captured multiple times)

One statistic I find particularly cool: there are tens of thousands of active weekly Bee devices. Compare that to a traditional map maker who has a few hundred dedicated mapping vehicles at best (which are certainly not all active all of the time). In other words Bee Maps’ fleet is at least two orders of magnitude bigger than a traditional mapping company. Nice.

You can start to appreciate the coverage by looking at Bee’s Strava-like heat maps. See below (click/tap to embiggen) or go here for the full interactive coverage map.

So I know what some of you nerdy types might be thinking: surely companies like Tesla are capturing enough data from all the cameras on their cars that they far surpass what Bee Maps is doing?

Well no.

Firstly: if you approach Elon and ask him nicely if you can license his data, he’s simply going to tell you to fuck off. Secondly: Teslas are only sold to well-to-do people, ipso facto they only generally collect data in well-to-do places. Even Toyotas don’t go everywhere!

Yeah, I know, perhaps one day the likes of the NVIDIA Drive Platform might get embedded in enough vehicles. Or perhaps we’ll all end up wearing some data-hoovering douche bag glasses from some dubious social media company. Or maybe Jony Ive will wow us all with a must-have OpenAI facehugger. But that’s all speculation.

In the meantime, get real, do your part, and get Bee Mapping!

---

Footnotes

1. And yes, it’s even worse for Apple Maps. The Apple Maps “Look Around” image of the house where I used to live in Oakland, California is five years out of date. 😱 ↩︎
2. Ariel is pronounced “R-e-ull”, not like Ariel in the Little Mermaid. ↩︎
3. I say “rather bravely” as so many hardware startups have failed miserably or limp along at best. Juicero, Pebble, Rabbit or Humane AI Pin anybody? ↩︎
4. So, Donald Trump, since I know you’re reading, I have four words for you: built in the USA. ↩︎

We’ll start today’s post with a little segue.

My son is a “gear head”. He got his start working for a Mercedes dealership but quickly got bored. He quit and went back to trade school and learned all about machining: the use of mills, lathes, drill presses, CNC machines and industrial size 3D printers. This turned out to be the perfect set of skills — knowing how to work on cars and knowing how to craft and fabricate complex custom parts.

He’s now a founding partner of a small start up in town. They build custom cars for very rich clients who typically spend $200,000 – $500,000 on their vehicles. Many of them are what arecalled “resto-mods” which is an abbreviation for restoration-modifications. In other words, taking an old car and completely restoring it, but replacing all the old components — suspension, brakes, engine, transmission — with modern equivalents. The finished vehicle often looks like the original but drives like a brand new modern one, right off a production line.

Here’s one of the finished products (click to play video with the sound turned up. It’s worth it!):

The art of building these custom vehicles requires not only an extensive set of standard tools but also a large number of specialised tools, and often some very custom built tools too. For example, here’s a totally custom tool specially developed by the shop to enable cutting exhaust pipe at precise angles:

The end result of all this extensive work? A beautiful and unique vehicle that precisely fits the requirements of its customer in every single detail, perfect for its job.

Now, why am I bringing all this up? Well, funnily enough it all relates back to mapping in a very roundabout way.

For roughly the last 30 years people have been starting to use maps in their lives. It started slowly and then grew exponentially, especially for consumers, once the internet took off. And in parallel with that the use of maps in business has gradually grown. Even now it’s not pervasive in the enterprise, but at least maps have slowly made their way into business analytics products like Tableau and Power BI. It’s now even possible to add a basic map to a spreadsheet in Microsoft Excel 365, at least in a rudimentary way.

But there is a fundamental problem: all of these mapping products, including ones from the enterprise mapping technology behemoths like Esri, are general purpose tools. Translating back to the automotive world for a moment, it’s like having a very basic set of spanners and wrenches. To use these generic products for any very specific business problem is actually very difficult, and even if you do use them, they don’t really provide what you need and they really aren’t fit for purpose.

What you need in these cases then is a very specific product. Rather like that custom built resto-mod vehicle — precisely fitting the needs of the customer in every single way. But, unlike that custom vehicle, it can’t break the bank! It has to be affordable and provide a significantly better return on investment.

So that brings us to this week’s topic: the second in the series of “Mapping Amazing” companies. Today we’re going to focus on a small start up based in Munich, called Mily Tech.

Mily Tech got its start just a four years ago in 2021, when its founder, Miloš Zlatković, realised that there was a significant problem in the pervasive and growing industry of last mile delivery.

The problem revolves around something called “Out of Home delivery” or “OOH delivery” whereby a parcel or package is delivered to a smart locker or a pickup and delivery counter. OOH delivery is perhaps slightly less common in the US than in Europe and Asia, but the need for it everywhere remains strong and it’s growing like crazy…

Think about it.

The need for OOH delivery is obvious:

• Parcels get stolen from front door steps.
• Deliveries get missed because the customer isn’t home.
• Deliveries can’t be made because there is a locked or key coded gate.

And there’s another significant incentive for delivery companies: a smart locker is a single delivery point for dozens of parcels, so use of them significantly increases delivery efficiency and reduces carbon footprint.

In the US this OOH delivery is dominated by Amazon Lockers. But in other markets many delivery companies national mail carriers are providing OOH delivery, mostly in the form of smart lockers.

At first glance the problem would seem to be a classic site selection challenge. Where do you put the lockers? You can’t just carpet bomb a whole city. It costs too much money and you can’t just set up a smart locker anywhere you want. And just like opening up a retail outlet in the incorrect location, it’s a very costly mistake to pick the wrong location.

And this site selection problem is different in every market. Tokyo is very different from Sydney which is very different from NYC.

But it doesn’t stop at putting the lockers in the correct location — it’s not fire and forget problem.

Once you’ve established your locker network you to have to make sure the utilisation of them is continuously monitored, optimised and adjusted. Why? Because the environment around each locker constantly changes. The number of parcels being delivered changes constantly. Sometimes a particular set of smart lockers is under utilised and sometimes it’s over subscribed. Competitive locations may pop up or disappear. And the population and demographics are always evolving.

Like all good companies, when Mily Tech initially looked into solving this problem, they started by talking to a ton of potential customers. They wanted to understand the precise issues, the root causes and, just as importantly, the set of tools the companies were currently using in an attempt to manage the challenge.

It was during this discovery process that the light bulbs quickly started to go off. It was like these big delivery companies and national mail carriers were trying to build a custom car using a set of basic wrenches and spanners.

Not only that, the set of tools they were using fell well short of addressing all the problems. Even if they came close, they had to accomplish it using multiple sets of tools across several disciplines. You can probably imagine: spreadsheets in one department, BI tools in another. And if you were lucky —very lucky — maybe a geographically insidious system (GIS) tool in a third.

Basically it didn’t work. Last mile delivery companies were desperately in need for the equivalent of that perfect, custom resto-mod car!

So that’s what Mily Tech set out to do. Build the perfect custom software platform to solve this very tricky problem. It’s a platform that transcends several disciplines. It’s not a spreadsheet. It’s not a BI tool. And it’s definitely not a general purpose GIS.

Using Mily Tech customers are able to build an optimal network of OOH delivery points, manage them for maximum efficiency and then easily scale them to expanded or entirely new markets.

And they’ve already had tremendous success. They’ve won business from DHL in Europe as well as a number of national mail carriers, for example in Italy and Slovenia. And now potential customers are starting to ask the internal question: “Why aren’t we using MilyTech?”

You’d think the market for last mile delivery is actually pretty small. There’s only so many delivery organisations after all. But they deliver hundreds of millions of parcels every day and every tiny gain in efficiency gain makes a huge difference to the bottom line.

But delivery isn’t the only problem to solve. There’s actually an even bigger problem. And it’s about optimising returns. Why? Because returns are a pain and expensive. They are the bane of not only every delivery company, but also every e-commerce company too. And there are orders of magnitude more e-commerce companies than delivery companies.

Mily Tech contrasts this problem with last mile delivery and they’ve given it a name. They call it “First Mile Returns”. And they’re coming after the problem to solve it. Stay tuned!

In the meantime, if you’d like to learn more check out milytech.com.

Until the next time…

---

The Mily Tech Product in Action:

Welcome to the first in a new series of articles on Map Happenings that I am calling “Mapping Amazing”.

Each post in the series is going to feature a small company or organization that, through use of geospatial technology, is doing innovative / rockin’ / lit / vibing / happening / mapping amazing work (pick your adjective!) to make a small dent in the universe.

And today we’re going to feature a little known startup in London called POSTX. It’s a little different, but it’s using geospatial excellence at its core.

I got to know the founder of POSTX, Max Glaisher, as a result of business dealings I had with him during his career at DPD, one of the largest delivery companies in Europe. While at DPD Max was one of the leaders on their innovation team. I was always impressed with his energy, spark and drive to “think different”.

Sure enough, he did. In 2023 Max founded POSTX with the aim of changing the way the world moves goods with zero emissions.

A big hairy audacious idea. But how can a tiny startup with fewer than 10 full time employees possibly begin to go head to head with the likes of FedEx, UPS, DHL, DPD and Amazon Logistics?

Well the answer is two fold.

First — they don’t. They’re not aiming to replace any of these behemoths. Instead they’re aiming to transform the old sausage making plants the big guys use and bring them firmly into the 21st AI century. So, for the boomers among you, POSTX’s MO is more akin to the famous BASF ad from the 1980s: “We don’t make a lot of the products you buy. We make a lot of the products you buy better.”

Second — through innovation. To deliver all goods with zero emissions is about SO much more than buying a fleet of electric trucks and using them in your delivery network. And actually, it’s not about buying any trucks at all.

But before we get into the details, let’s start with a little history lesson on the last time the whole world of global delivery got disrupted …

Wind the clock back 54 years to 1971. This was the year that Intel released its first commercially available microprocessor (the Intel 4004). It was also the year that electronic calculators and digital watches started to appear on the market.

But something else happened in 1971. While at Yale University, a chap called Frederick W. Smith came up with a devastatingly disruptive idea for a new kind of logistics company that would revolutionise time-sensitive shipments. This was around the era that “just-in-time” manufacturing was becoming vogue, and Fred saw a critical need for fast, reliable delivery.

The revolutionary idea behind this new company was to use a hub-and-spoke model, where by all goods were first shipped to, and then sorted at a centralized hub in Memphis, Tennessee. This model allowed the company to guarantee next day delivery, something no other carrier could offer reliably at the time.

But this wasn’t the only secret ingredient. The company also used a system of barcode tracking and real-time package updates that set a whole new standard for logistics transparency. By pioneering overnight delivery and package tracking, this new upstart fundamentally changed customer expectations and forced competitors like UPS and the U.S. Postal Service to adapt. It didn’t just deliver parcels — it redefined speed and reliability in global commerce.

The company? Well, it’s FedEx of course.

So what’s happened in the intervening half century? Well most logistics companies have copied FedEx, not only the hub-and-spoke model for air shipments, but in using the same hub-and-spoke model for ground shipments too. This has all meant the need to build massive warehouses so parcels can be collected, sorted and then shipped back out to their final destination. In other words: a huge capital expense.

So has anything innovative happened since the dawn of FedEx? Well yes, inside the warehouse there’s now a ton of expensive computer controlled conveyer belts and sorting systems, as well as increased use of robots (although robots are more applicable to organizations that have to pick products as well as sort them —think Amazon or Walmart). And for route planning obviously everyone is using digital maps and mapping software to help drivers find their way.

But there’s some fundamental underpinnings of all logistics companies that are still stuck in the mid 20th century. What am I talking about? The answer: postcodes — or to you Yanks — ZIP codes.¹

For example, in the US, ZIP codes were invented in 1963 and in the UK, postcodes were invented in 1959. And yes, both were invented to help streamline and automate sorting. But 60+ years on they are all getting rather long in the tooth. Coupled with that, the postcode system is different in every country and by no means do all countries have them. For example, Ireland only released their “Eircode” system in 2015 and populous countries such as the UAE, Saudi Arabia and Hong Kong have yet to create or fully develop a system.

So what’s wrong with postcodes?

Well first they don’t tell you how close things really are to each other. Then they don’t indicate how far one postcode is from another postcode. And they certainly don’t convey how routes should be grouped to be as efficient as possible based on distance, time and cost.

Take a look at the postcodes for London as an example and you see the issue. The denser the population the smaller the postcodes are (look at the inset for EC, SW1, W1, WC1 for example). So for logistics purposes the postcode actually tells you little about geography or geographical distance:

This issue creates a range of planning issues when designing service areas for depots and routes.

So what do legacy logistics companies do to try and get around the issue? They build “origin-destination” matrices of pre-computed distances between postcodes. But even then they don’t create these matrices on the full postcode — there are simply far too many of them². So instead they fall back to creating matrices just based on general postcode areas.

For example, in the UK they only use the first few letters of the postcode, known as the “outcode”:

Ultimately, these precomputed matrices don’t reflect real travel distances, times or costs — so it’s not actually very useful for route planning.

To work around that, companies do the following:

1. Pay a third party to validate postcodes
2. Pay a third party to geocode each address into coordinates
3. Run a distance matrix to calculate travel time and distance between every delivery point for a particular day
4. Use a solver to figure out the best stop sequence based on business rules

It’s expensive, slow and relies on multiple external systems — just to get inefficient routes.

Even Amazon, with all their sophistication, still uses these matrices as a fundamental part of their mechanism for sorting out where packages should go.

Couldn’t there be a better way?

POSTX not only thinks so. They know so. And it became one of the primary pillars of their new venture.

Pillar 1: Nexus Grid + the Atom Sorting Index

One of POSTX’s first tasks was to generate an entirely new sorting index designed specifically for logistics, not for just for delivering letters. And it’s not national. It’s global.

They call this new sorting index “Atom”, and Atom is part of an overall platform that they are building called “Nexus Grid” which will also provide predictive routes and sequencing.

And how is Atom different? Well it replaces postcode matrices with a smart, machine-readable grid — so every location is already sorted by how it connects to others at local, sub regional, regional and national levels.

This lets them:

• Sort vertically (national → regional → local)
• Sort horizontally (between any two places nearby)
• Automatically group routes based on real travel cost and time, not mailing zones
• Do it all without 3rd party maps or postcode logic

Unlike postcode boundaries that are drawn by people, Atom was built using a machine learning model to map and organise the world into hierarchical 6-unit geospatial zones. This pre-indexed structure embeds real-world distance, time, and cost into every location, allowing routes to be clustered and optimised almost instantly — without relying on postcodes, third-party maps, or heavy sorting infrastructure.

This video compares traditional postcodes to the hierarchical nature of Atom:

Another groundbreaking change …

Nexus Grid enables something POSTX calls Virtual Sortation.

Virtual sortation eliminates the traditional “collect → central sort → deliver” model.

Instead of loading everything onto a large truck and driving it to a central sorting center for unloading and belt sorting, goods are pre-sorted directly at the origin warehouse into modular numbered trailer PODs (individually tracked by Bluetooth) — each one already grouped by its delivery destination or route.

Those PODs can be:

• Pre-loaded at ground level for speed in or outside warehouses efficiently
• Swapped between vehicles without unloading
• Sent directly to their destination depot or delivery zone
• Used as mobile depots for local delivery drivers, skipping the long commute into industrial estates to collect their parcels

This removes several-handling steps, eliminates the need to expand large sorting hubs, cuts loading/unloading times dramatically, and lowers transport costs by reducing unnecessary miles.

Now actually POSTX’s method for using trailer PODs isn’t altogether new. In fact the concept of separating the goods that need to be delivered from the driver all goes back to the horse and carriage days, and this was later replaced by the “mechanical horse”. While separating the goods and the driver has remained with us for long haul routes (e.g. 18-wheelers and 53′ trailers) it’s been lost for middle distance and local deliveries. POSTX is bringing that concept back to life.

Let’s look at one of POSTX’s mailing labels. They are actually bi-directional and work for returns. And they eliminate the need for manual re-labelling for in-network transfer of goods or transfer across networks:

Note Einstein is embedded in the QR code! As Max from POSTX told me:

Einstein gave us the tools to understand the chaotic dance of atoms. POSTX uses the Atom sorting index to decode the dance of shipments, reordering how we plan routes using space, time, and cost for logistics. I thought it made sense!

“But come on now!” I hear you say, “How could this tiny company with fewer than ten people get any business just by inventing a geospatial index for logistics?”

Well that’s where the other pillars of POSTX come in. Let’s tackle them one by one.

Pillar 2: Zero Emissions

If you visit the POSTX website, you’ll see that zero emissions is front and center in their message. And it’s certainly a key selling point that their customers are keen to leverage. The ability to claim “100% Emission Free Delivery” is a value that resonates strongly with the majority of consumers.

So every single one of the deliveries made within the POSTX network is done with electric vehicles. But they’re not doing it like Amazon and paying Rivian hundreds of millions of dollars to develop custom electric delivery vehicles. And nor are they doing it by buying a fleet of Ford E-Transits or Mercedes eSprinters. And they are not buying any electric trucks.

Nope, in fact they’re not using vans or trucks at all. They’re using electric cars, SUVs to be precise, all sourced from (cue “hiss”) … Tesla. But even then they’re not buying them. They’re renting them. And part of the secret sauce is they are using Teslas not on their own, but with trailers. It’s the modern day Scammell Scarab:

Here’s a short promo video explaining how it all works. If you pick one video to watch in this article then choose this one:

Phew, well that’s a bit different, eh? But that’s not all. Let’s move on…

Pillar 3: No CapEx

So, no purchasing of any electric vehicles. And no purchasing of any trailers either (they’re rented too). So no CapEx in fleet costs. In fact eventually POSTX envisages using a network of drivers who happen to own an electric vehicle that is capable of towing (for example, a Tesla Model Y).

The owners get to use their car as their daily driver but then they also get to make money from it by making deliveries. The folks at Tesla love this because POSTX is short circuiting Elon’s master plan for people to make money from their Tesla vehicles…

But it doesn’t stop with the fleet.

If you operate a logistics company — remember the story about FedEx? — one of your biggest costs is setting up a sorting warehouse. But due to the magic of the 21st-century geospatial logistics software behind POSTX, guess what?

You don’t need any warehouses with expensive conveyor belts dedicated to sorting goods and packages.

Goods are pre-sorted at the beginning of the fulfilment process, tracked with POSTX’s Atom destination code and return numbered labels, and then loaded directly into modular trailer PODs with the assigned numbers.

Any further sorting happens virtually — it appears in the driver’s route as a scheduled stop. At that stop, drivers either drop-swap & go (exchanging trailers moving in opposite directions) or perform a shipment exchange (transferring goods from one trailer to another).


Think of it like internet data packets being sent from router-to-router, but one step further — instead of the routers being at a fixed locations they’re actually mobile.

These exchanges remove regional sorting overheads, significantly reduce sorting time, and can be completed while vehicles charge at designated “stop points” — eliminating the delays and costs of traditional hub-based processing.

Bam. There’s a humongous cost eliminated.

Pillar 4: Get Anywhere: Faster and MUCH Cheaper

There are a number the advantages of using a car with a trailer rather than a traditional panel truck or van.

First, unlike a truck, they are small. They can get virtually anywhere — for example, up someone’s narrow gravel driveway.

Second they are permitted virtually anywhere — there are generally no weight or size restrictions unlike there are for trucks.

Third, trailers are much faster to load and unload than trucks with tail lifts. Also, they are also a lot safer to load and unload. POSTX has shown that their trailers can be loaded 14X faster than a truck with a lift gate:

Fourth, unlike vans the trailers can be dropped off anywhere, sans driver. So they can be dropped off at a manufacturer’s facility the night before, loaded, and picked up later. They have the advantages of a full size 18-wheeler rig and trailer, but the size and weight advantages of a van.

But then let’s talk about cost. Not theoretically, but as proven in real world operations:

POSTX is showing their costs are 50% less than using a diesel truck, not including the driver. If you take into account the full cost of the driver, this cost savings rises to 70% or more, not because POSTX pays their drivers less, but because of the efficiency brought to bear by the geospatial excellence of their system. As a result they need significantly fewer driver hours to make the same deliveries.

Pillar 5: Logistics-as-a-Service

So actually POSTX isn’t looking to compete with the big boys at all. You see they don’t see themselves as a logistics company. They see themselves as a Logistics-as-Service platform that helps existing logistics companies significantly reduce their costs. They do this by automating planning, routing and sortation, all powered by a network of driver owned EVs and rented trailers.

As such POSTX is aiming to provide a service to logistics companies for 20-40% of their routes with a much more flexible and lower cost solution.

By optimising sortation and routes using their modern Nexus Grid platform they can increase speed, efficiency and reliability all while significantly reducing costs.

Pillar 6: Focus!

For any startup, focus is critical. And for POSTX it’s no different. They are focused on just one aspect of the logistics market. It’s called the “middle mile”. As opposed to last mile, the middle mile is involved with bulk shipments that commonly involve moving pallets of goods over longer distances, often with fewer stops.

In the UK, existing middle mile fleets typically spend £300+ ($400+) per route and cost £60/hour ($80) to run. POSTX has proven they can cut these costs in half.

And so POSTX’s is the end-to-end platform for logistic companies to make their middle mile deliveries.

And there’s another network effect. As more customers use the POSTX platform, it continuously indexes detailed movement patterns — rather like Google indexing the web — enabling increasingly accurate predictions on how to move goods with the optimal balance of time, cost, and service.

So How is POSTX doing?

Incredibly well.

Growing 29% MoM. A solid base of middle mile logistics customers across the UK. Providing a capital light / no CapEx model with simple pricing. And already over 300,000 miles of tailpipe emissions avoided.

One of their key customers is a group called Segen, who is a top distributor of solar systems, energy storage and EV charging solutions across the UK. Faced with high damage claims, delays, failed deliveries and rising middle mile costs they turned to POSTX to help them with deliveries.

The outcomes?

• 100% on-time, in-full deliveries with 0% damage
• 50% cost savings
• Multi-site expansion after 12 week pilot

And POSTX’s success is showing up in other areas too:

• In October last year they won a National Sustainability Award together with their customers Segen and E.ON
• This year they are being recognised as building the UK’s safest courier business.

Mapping Amazing: POSTX

So there you have it. The first in a series on “Mapping Amazing” companies as highlighted by Map Happenings.

POSTX, providing a next-generation Logistic-as-a-Service platform with geospatial excellence at its core.

Thanks for reading.

---

Disclosure:

I am on the advisory board of POSTX. Cynics will say: “That’s why you promoted them.” Non-cynics will say: “If they really are good, why wouldn’t you join their advisory board?”. I’m a cynic. So I get it.

Further Viewing:

• AirTags Expose Dodgy Postal Industry (DHL Responds)
• “The Postcode Connection”, A 1970s documentary about how the Post Office deliver post

Acknowledgements:

1. POSTX
2. Vox media
3. FridayAdventureClub on YouTube for explaining Amazon shipping labels
4. IdealPostcodes for their explanation of UK postcodes

Footnotes:

1. Gold star if you know what ZIP stands for. ↩︎
2. In the UK there are about 1.8 million full postcodes (e.g. “SW1A 2AA”). So an origin-destination (OD) matrix for every postcode would have over 3 trillion entries. And that just for one variable, e.g. distance. That’s a massive matrix — definitely not something you’d want to hold in memory. This contrasts with the number of outcodes in UK postcodes (e.g. “SW1A”) of which there are only about 3,000. So an OD matrix just based on outcodes for one variable only has 10 million entries. ↩︎

Just over three years ago, having been lucky enough to be immersed in mapping technology for many decades I thought it was time to give back. Thus the creation of what you are reading now, Map Happenings.

In the first post, on 27 July 2022, I wrote:

So after 37 years in the mapping industry I thought it might be fun to inflict my view of happenings in the geospatial technology world on a few of you. 

I’d like to throw you the occasional tasty tidbit of various ‘goings on’ and, from time-to-time, do a deeper dive into some relevant topic or company. Oh, and I might try to invoke a reaction from one or two of you through a particular musing or postulation. 

While I’m sure much of the focus will be on what’s happening now, it’s always fun to make guesses about the future. And, of course, let’s not forget the past. As Churchill said: “Those that fail to learn from history are doomed to repeat it.”

I will aim for quality, not quantity. Hopefully I’ll meet your expectations.

Included in the many posts was an attempt at a fun and irreverent series on the history of maps and mapping technology called “The 12 Map Happenings the Rocked Our World“.

Despite my poking fun at the French, the series still got a very favourable reception from none other than the New York Times Literary Review ¹

Given this history of maps and mapping series is now finally complete, (you can let out big sigh of relief), I thought I might transition to something new, with the emphasis on the word ‘new’ rather than ‘history’. So I have decided to embark on a series that I am calling “Mapping Amazing”.

Each article in the series is going to focus on a small company that, through use of geospatial technology, is doing innovative / rockin’ / lit / vibing / happening work (pick your adjective) in an attempt to make a small dent in the universe.

To get a flavour of what the articles will be like, take a look at the Map Happenings article from May 2024: “PLACE: Fuelling Maps for the Rest of the World“. This featured the work of a small organisation that is doing an awesome job to provide the “data fuel” for the creation of detailed maps for the developing world.

So, stay tuned for the new series, “Mapping Amazing”. I think you’re going to love it.

---

1. Gold star if you can spot the AI. ↩︎

The Whole World in Your Pocket.

Something happened on November 10, 2001.

Suddenly, you could buy a mobile device that would store about 100 albums of music and allow you to listen to it — anywhere. At the time, this capability was considered astounding and was unmatched by any other product.

The device that enabled all this? iPod, of course — the first dedicated music device from Apple in the renaissance years of Jobs 2.0.

The tag line for the product — “1,000 Songs in your pocket.” — immediately resonated and people quickly ‘got’ what the product was for.¹

It didn’t take long before the whole iPod product line took off. Within a few years nearly everybody seemed to be carrying one of these handy devices in their pocket, pinned to their shirt or strapped to their arm.

But of course it didn’t stop there. Six years later, by the time Apple had produced its 100 millionth iPod, Apple launched another earth shaking product: iPhone.  

Announced in Steve Job’s now much lauded keynote speech at Macworld on January 9, 2007, it generated shock and awe around the world.

As I am sure many of you will remember, iPhone was portrayed as three amazing products in one: a widescreen iPod with touch controls, a revolutionary phone and a breakthrough internet communications device. When Steve demo’d it, the crowd went wild over how it made previously extremely difficult (or impossible) tasks super easy and fun.

Much of the demo time was dedicated to showing how iPhone could be used for the basic stuff — listening to music, watching videos, making calls, sending email and texts and browsing the web.

But there were a few other things that Steve showed off: a weather and stocks ‘widget’, which we would now call an ‘app’, and then something else. Something that he spent several minutes demoing … something he called “the best version of Google Maps on the planet”.

This particular part of the keynote caused roars of applause, especially when he used the maps app to find a nearby Starbucks — which he promptly called, live on stage, to order 4,000 lattes “to go” for the audience.

And that wasn’t the only maps feature to get a roar — the audience was particularly amazed when he zoomed into the Washington Monument, the Eiffel Tower and the Colosseum in Rome using satellite view. Watch the excerpt:

Amazing though it was there is a whole story behind this part of the demo. Steve Jobs talked about it later that year when he and Bill Gates were interviewed by Walt Mossberg and Kara Swisher at the D5 conference on May 7, 2007 (video is well worth a watch BTW! Also here is the transcript):

I love Google Maps, use it on my computer, you know, in a browser. But when we were doing the iPhone, we thought, wouldn’t it be great to have maps on the iPhone? And so we called up Google and they’d done a few client apps in Java on some phones and they had an API that we worked with them a little on. And we ended up writing a client app for those APIs. They would provide the back-end service. And the app we were able to write, since we’re pretty reasonable at writing apps, blows away any Google Maps client. Just blows it away. Same set of data coming off the server, but the experience you have using it is unbelievable. It’s way better than the computer. And just in a completely different league than what they’d put on phones before.

But maps weren’t even part of the original plan for iPhone until just a few weeks before the unveiling on January 9, 2007. Nick Wingfield and Brian Chen recount their 2012 story in their article on Apple Maps in the New York Times:

Including a maps app on the first iPhone was not even part of the company’s original plan as the phone’s unveiling approached in January 2007. Just weeks before the event, Mr. Jobs ordered a mapping app to show off the capabilities of the touch-screen device.

Two engineers put together a maps app for the presentation in three weeks, said a former Apple engineer who worked on iPhone software, and who declined to be named because he did not want to speak publicly about his previous employer. The company hastily cut a deal with Google to use its map data.

Not only did maps get featured in Jobs’ launch keynote it also became a mainstay feature of the original iPhone ads:

And, 18 years later, where are we now?

Well today there are estimated to be 7.2 billion smartphones in use around the planet. They are used by 4.7 billion people.² That’s about 58% of the world’s population.

And every one of those devices has access to a maps app. And every maps app has a map of the planet.

So, if you boil it all down: at least one out of every two people on the planet now has a detailed map of the whole world in their pocket.

If that isn’t a Map Happening, I don’t know what is!

Epilogue

Well you did it. You suffered through all 12 episodes of “Map Happenings that Rocked Our World”. Hopefully many of you will have enjoyed at least a few of the posts.

If you missed some of the earlier articles and want to subjugate yourself to them, you can bookmark the whole list at maphappenings.com/history or simply browse them here:

12 Map Happenings That Rocked Our World

Part 1 — The First Map
The story of how, many thousands of years ago, the first map came to be.

Part 2 — The Birth of Coordinates
Geographic coordinates haven’t been around for ever. But they have been around for a while. Read the fascinating story of who invented them.

Part 3 — Road Maps!
Think road maps came out when people started driving cars? Wrong! Read on to find out the real story.

Part 4 — The Epic Quest for Longitude
Best selling books have been written and films have been made of this amazing story from the 1700s. It truly is an epic tale.

Part 5 — The Dawn of Tube Maps
The London ‘Tube’ (or subway) was the first of its kind to invent an entirely new kind of map. Now everyone’s copied it. Learn the history of how it came to be. 

Part 6 — The Advent of Computer Based Mapping
Computers started to emerge after the Second World War. Learn about the brilliant minds that first started using them for maps and mapping.

Part 7 — Those Views from Above…
Today we’re all accustomed to looking at satellite and aerial imagery of cities and places. This is the story of how it all came about.

Part 8 — Oh Brother, Where Art Thou?
Determining your current location is easy. You just open your phone and look at a map. But it wasn’t always that easy. The full story is in this post.

Part 9 — A Curious Phenomenon Called ‘Etak’
This is the story of the world’s first practical navigation system for cars. And it’s an amazing one that is hard to believe!

Part 10 — A Relentless Quest for Maps
Before Google Maps launched in 2005 there was another website that dominated maps on the web. This is the insider story of that site: MapQuest

Part 11 — Maps, Earth and Views
The fascinating background story about how Google Maps, Google Earth and Street View came to be.

Part 12 — The Whole World in Your Pocket.
This article!

Thanks for reading.

---

1. The tagline was invented by the advertising agency Chiat Day, not Steve Jobs. ↩︎
2. Source: Genius: How Many People Own Smartphones In The World? (2025-2032) ↩︎

Maps, Earth and Views

So tomorrow, Friday February 7, 2025 at 7pm Pacific time, marks an anniversary.

That precise time will be the 20th anniversary of the launch of Google Maps. And what a journey it has been. Not only for all the hard working folks at Google, but also for the billions of human beings on this planet that have benefited from this ubiquitous geographic platform.

Only someone at Google could delve into this, but I’d have to guess that the number of journeys that Google Maps has aided must number in the many trillions.

Many of you mapping nerds out there might already know the full history, but perhaps many of you don’t. Or perhaps you need a refresher…

Let’s wind back the clock to 1999. At that time if you wanted online mapping and directions — at least in the USA — there was only one place to go: MapQuest.com. For the history of that groundbreaking company see “12 Map Happenings that Rocked our World: Part 10: A Relentless Quest for Maps“.

But that same year another tiny, tiny seed was being been planted. And the people doing the planting were a couple of chaps called John Hanke and Brian McClendon. John and Brian were both tracking the big bang explosion of the internet and the web that was consuming everyone at the time. And they had a new stealthy startup. But it turned out it wasn’t anything to do with the internet — their secret new product all ran on a huge Dell PowerEdge server.

And on one warm spring day in Austin they demo’d the product to Bill Kilday, who had been John’s friend for 15 years.

Before I get to reveal the meat of the demo, let’s quickly wind the clock forward to the present day: as a result of that demo, John, Brian and Bill started a journey together that lasted for the next 24 years. It begat a geographic product that nearly everyone on the planet knows: Google Maps. How about that for a Map Happening That Rocked Our World?!

In 2018 Bill wrote a book about their travails. It is the definitive, insider story of Google Maps. The book is called “Never Lost Again” and I can’t recommend it enough. Among other places it’s available on Amazon and Apple Books:

In writing this article I’ve referenced the book liberally and, speaking frankly, there is absolutely no way this article would have the detail it does without having had “Never Lost Again” as my guide. So @Bill — thank you, thank you and thank you.

I recently had the privilege of chatting with Bill and I told him about my plans for this post. We both agreed that the timing should coincide with the 20th anniversary of Google Maps.

So, with that in mind, here we go!

Let’s go back to that demo that John and Brian gave Bill in the spring of 1999. Bill recounts the demo:

On the screen floated a perfectly detailed rendition of Planet Earth. Oh, I thought, there’s that photo. You know the one. It was on the cover of every junior high science book. Known simply as the Blue Marble, this classic image was captured on December 7, 1972, by the crew of the Apollo 17 spacecraft five hours into a lunar mission. This one spectacular photo is widely credited with jump-starting the modern-day environmental movement.

The Earth seemed to be rotating. Moving somehow. “What’s your address again?” John asked. “It’s 465 Joe Sayers, Austin, Texas,” I said as he typed.

That’s curious, I thought. The photo on the screen. The Earth. It is moving! Maybe it’s a QuickTime movie animation of some sort? But QuickTime movies aren’t interactive in any way, they are linear. Why would John need my address if this were a movie and . . . Wait, how is this thing zooming in? What the hell is happening here!?

These thoughts raced through my head during the fifteen seconds that we zoomed from outer space down to what was clearly my house. I recognized my standing seam metal roof, the neighbor’s trampoline in their backyard, my red Ford Explorer parked out front, and the Arroyo Seco running behind the house.

“Holy shit!” I exclaimed.

What Bill was seeing was the beginning of something that would take the world by a storm. But it wasn’t Google Maps. What he was seeing was a product that would come from a new stealthy startup, called Keyhole ¹.

Keyhole was born out of the work of another Silicon Valley startup, Intrinsic Systems, who had pioneered a super efficient way to stream large image databases to a screen. It was this technology that made the demo possible. Initially the thought of putting it on the web was deemed a risky strategy. To quote John Hanke at the time:

If we put it in a browser, it will be just like MapQuest and they already own that market.

By 2001, the Dot com bubble had burst and Keyhole missed their initial opportunity for a funding round. They scampered to find a business model that could be built around their incredibly sexy demo. And their attention turned to enterprise mapping — at the time still very much dominated by Esri. Bill recounts the history:

Creating maps using Esri software required sophisticated data analysis and training. You could get an accredited college degree in GIS, spending four years learning how to use it. More realistically, a team of Esri-trained specialists (or Esri’s own consulting arm) was needed to configure a custom solution for a client.

Esri was essentially a blank sheet of paper. In order to begin using it, you had to go out and find the data you needed, and download and import it. In the meantime, you had to hope that the data was the right format and map projection so that the multiple layers of data worked together. Esri specialists were needed to acquire and integrate mapping data in order to make the application work.

Lastly, it was slow. Viewing an Esri-generated map online involved a click followed by a long, painful wait for the map to refresh. If you wanted the base map for your Esri map to be an aerial image, the software bogged down even more. Esri was glacially slow, almost unusable, especially if you didn’t have the aerial image loaded on your local hard drive.

By comparison, Keyhole was simple to use, came bundled with access to terabytes of data, and was lightning fast.

Bill went on to say, half joking:

Maybe Keyhole could be a simpler GIS aimed at non-GIS experts? Maybe we could be the GIS solution for people who can’t spell GIS?

It was with this idea that Keyhole’s first product was born on June 14, 2001: the Keyhole EarthViewer. It was aimed at one just vertical: commercial real estate. But scaling sales proved hard. By 2002 Keyhole was running on fumes. Despite cash injections from Nvidia and another Japanese company their burn rate was way too high. There were other deals on the horizon, one with CNN and more intriguingly an opportunity with the US government in the aftermath of September 11. But by January 2003 the board was at the point of laying people off or shutting down the company. Management was given three months.

On March 20, 2003 the US invaded Iraq. And a week later Keyhole’s deal with CNN was closed. Within a few days CNN was using EarthViewer — CNN called it their Map Viewer — on live TV to zoom interactively to military action on the ground.

If there was one single event that saved Keyhole’s you-know-what, it was a key clause in their contract with CNN: John had got them to agree to display Keyhole’s URL on the screen anytime they used EarthViewer on air.

Within a few days a headline in the USA Today newspaper told the whole story: “Tiny Tech Company Awes Viewers”.

CNN’s use caught the eye of the US government. It wasn’t long before In-Q-Tel, essentially the US Federal Government VC arm, was heavily engaged. And Keyhole still only employed 20 people.

Keyhole started to blossom — and not just in financial ways. They managed to snag John Rohlf who is still considered among the best 3D graphics programmers in the world. It was John who invented something groudbreaking and it was called: KML.

KML stood for Keyhole Markup Language and has gone on to become a phenomenon that enables one of the most prolific ways to exchange geographic information. Much like you and I can easily exchange and read PDF documents, KML allows people to exchange and read geographic information, which could then be displayed on top of Keyhole’s interactive digital globe. This changed Keyhole EarthViewer from a single user app into a hugely collaborative tool.

Now organizations could collaborate and share geographic data, viewing it in Keyhole’s super easy and super fast tool. This caught the attention of public safety and law enforcement departments across the country. Fire and police departments started to use Keyhole in place of their legacy enterprise “Geographic Information Systems” (GIS) software made by the 800 pound gorilla, Esri.

I was working at Esri at the time. I can tell you that it sent their CEO and founder, Jack Dangermond, into an apoplectic panic. In fact the shivers were felt down the spine of the entire organization. How could this tiny company be having such a disruptive effect?

But for Esri at least, the worst was yet to come.

By early 2004 Keyhole reached a classic inflection point. It was time to raise a “Series B”. Menlo Ventures was the lead. On April 21, 2004 they provided a term sheet and it was time to make a decision. But instead of signing John and his cohorts delayed and holed themselves up in a conference room.

After a few hours the news finally broke to the other “need-to-know” employees.

Google wanted to buy them.

Series “G”

By October 27, 2004 the deal was finally made public. Keyhole was now wholly owned by the Google. Google had only just gone public a few months before. They had become super bold and were mushrooming fast.

Strategic meetings with Sergey Brin and Larry Page ensued. Keyhole was almost immediately challenged. Bill Kilday, by then Product Marketing Manager for Keyhole, was anxious to hear about Google’s plans for the company. He wanted to ask Larry and Sergey:

A year from now, thinking about what success looks like, would it be better if Keyhole generated ten million dollars in revenue or had ten million users?

To Bill these were audacious goals.

But Larry and Sergey had other questions, such as “How big would their database need to be to cover the entire planet at one meter resolution?”

Bill finally got the chance to ask his question. Larry and Sergey thought for a moment: “I think you guys should be thinking much bigger than that.”

It took quite a while for their answer to sink in.

Where2Tech

It turned out Google had separately been experimenting with mapping as a way to add location information to search results. But it was it was about to get another lift, and that lift came from Where2Tech.

Where2Tech was founded by two Danish brothers, Jens and Lars Rasmussen. In the latter part of 2003 they found themselves unemployed from a company called Digital Fountain in Sydney, Australia.

Jens had been tinkering with the idea of adding mapping capabilities to a CD-ROM based business search app (or “Yellow Pages” app). This was to be an app that you installed — it was not a website. And, like Keyhole, Where2Tech stayed clear of browser based mapping because of the dominance of MapQuest.

With both of them unemployed they pitched their idea to anyone that would listen. One of Lars’ contacts made a fortuitous introduction to Larry Page and a meeting was finally scheduled on June 4, 2004. Legend has it that with holes in his socks and only $13 in his bank account, Jens booked a flight to California on his credit card.

Larry immediately challenged the brothers to make their app work in a browser. The Where2Tech app was certainly better than what MapQuest provided: it was fast — much faster than MapQuest. Unlike MapQuest, who generated each map image on the fly, Where2Tech relied on a vast array of pre-generated map “tiles”.

Where2Tech’s approach allowed not only fast maps, but it also allowed big maps — and not only big maps, but big “slippy” maps — maps that you could simply click and drag to pan rather than having to click on a ‘Pan’ button and wait for a new image to be generated.

Three weeks later Jen and Lars came back with another demo. It was enough to get Larry to offer them employment contracts to work at Google.

By October 2004 the Where2Tech and Keyhole teams were combined. In time they were collaborating to make each of their fledgling products better. By November 2004 the goal was set: they were to launch Google Maps in three months.

Data, Data, Data

The planned Google Maps product was not without its dependencies. It relied on satellite imagery and digital street map data. Furious negotiations ensued with imagery and street map data providers.

For imagery data Google primarily worked with Digital Globe. A deal was negotiated for imagery of the 200 most populated cities on the planet for $3 million. In Keyhole terms it seemed like a staggering price, but with Larry Page’s challenge to think bigger, the proposed deal was sent up the chain.

When Larry Page reviewed the deal his question was simple: “Is this all their content?” Larry didn’t want imagery for the top 200 cities. He wanted imagery for the whole planet. The maps team was flummoxed. Had they lost their collective mind?!

The team’s focus wasn’t just on maps and imagery. It was also on business locations.

At the time MapQuest, Google and others relied on business listings from the likes of a company called InfoUSA. InfoUSA in turn relied on teams of telemarketers to call businesses once a year. Their data was therefore always painfully out of date and, as a result, users of MapQuest were often directed to businesses that had moved or had closed.

Google started taking a different approach. They set up a system to crawl the now pervasive web pages that every business had, making the assumption that businesses had the incentive to keep their information much more up-to-date than any electronic “Yellow Page” directory. This effort fit perfectly with Google’s mantra to “organize the world’s information”. Now the mantra was simply being applied to “organize the world’s geographic information”.

Google Maps Launches

With fast, large, “slippy” maps and vastly superior list of businesses, Google Maps was launched, for initially just the US, at 7pm Pacific time, Monday February 7, 2005.

One hour later traffic started to pick up.

By 7.45am the next morning the site had been “slashdotted” — meaning that someone had started a thread about it on the influential tech website, slashdot.org.

By 9am the site exceeded the forecast (and server allotment) for the entire first 24 hours.

A phenomenon was birthed. Within days people in countries outside of the US were clamoring for the same UI. It didn’t take too long. In June 2005 Google Maps was launched in Japan. Then came the UK, then Ireland and then France.

For licensing reasons satellite imagery was held back as part of the initial launch. But it arrived on April 4, 2005, causing the servers to once again become overwhelmed.

Then, in June 2005 Google launched the Google Maps API. It was the equivalent of Apple’s App Store moment for iPhone. It unleashed whole new industries: Yelp, Trulia, Zillow and Strava to name just a few.

For a while things went swimmingly, but there were some snafus. For internal political reasons the Google Local team (focused on businesses) and the Google Maps team were later merged. And the name of the newly merged product? Google Local.

Google Local? WTF was that?

It didn’t take too long before the name was changed back!

But Whithersoever Google Earth?

You maybe thinking … didn’t Google Earth come out before Google Maps? The answer is no. While Keyhole EarthViewer continued to be nurtured under Google’s ownership, the priority was put on the launch of Google Maps rather than the relaunch of Keyhole’s product.

The relaunch was going to have access to 10 times the satellite imagery of Keyhole EarthViewer. It was also going to be much faster due to the vast infrastructure that was allocated to it.

As work continued on the relaunch an internal debate ensued: should it be “Google Earth” or “Google Globe”? The answer was decided through a simple test: say “Google Earth” ten times fast. Then try the same test by saying “Google Globe”. The decision was made. Google Earth it was!

There was also debate about the price. Keyhole EarthViewer was still $29.95 per year for personal use and up to $400 per year for GIS professionals. The answer from up high came down.

It was going to be free.²

Larry and Sergey preferred to do something great for users rather than make a few extra tens of millions of dollars in revenue. I guess also they also weren’t too concerned about cost!

Then there was a debate about registration. Given it was free, would you at least be required to register? Bill Kilday succeeded in pushing for removing even that barrier.

Google Earth was set to launch on 27 June 2005. A last minute internal scuffle with Marissa Mayer, then the final approver of everything that went live on Google, almost tanked the launch at the last moment. But finally everything was approved and the launch went ahead as planned.

At 28 MBytes per install (large for those days) the initial load on Google servers took its toll.

On the first day of its release Google Earth was downloaded 450,000 times. And within 28 hours the team met their download goal for the first two years.

Those Views of Streets

In 2005 Google went from having zero mapping products to having two ultra viral ones within just a few months. By the end of 2006, Google Maps had launched in 47 countries, and Google Earth had been downloaded 120 million times.

But like all good product managers the Google geo team didn’t rest on their laurels.

It just so happened Google had hired a guy called Luc Vincent in 2004. He was given the task of scanning the contents of libraries throughout the world. Ethical and moral questions aside, this was a monumental undertaking.

But soon after joining Luc was summoned to Larry Page’s office. Larry told Luc about another project, one that had started with a computer science professor at Stanford named Marc Levoy.

The project involved capturing videos of cities at the street level, collating them into a strip and then making them searchable by typing in an address. Larry was focused on something entirely new: making the physical world searchable from an entirely different point of view. Thus the idea of Google Street View came to be.

To cut a long story short, John Hanke introduced Google Street View to the world at the Where 2.0 mapping conference in San Jose on May 29, 2007. I was attending the conference at the time and I remember it well. There was a palpable gasp from the room, but I also remember people commenting “Yes, but can they scale?”

And I think you all know the answer to that question.

But Did Google “Win” — and What of the Future?

When pop stars go on a “World Tour” it doesn’t mean they go on a world tour. It generally means they tour in the USA, Canada, Japan, Europe and — if you’re lucky —Sydney.

With that in mind does Google dominate the consumer mapping world? Well the answer is no. Not quite. There are still a few key areas of the world where people don’t flock to their product:

• In China it’s Baidu Maps, Amap or Tencent Maps
• In Russia it’s Yandex Maps
• In India it’s sometimes MapMyIndia
• In South Korea it’s Naver Maps or KakaoMap
• In the rest of Southeast Asia Grab Maps is gaining ground

Then there’s Waze — but that’s a Google owned property.

Finally there is my alma mater — Apple Maps, still valiantly trying to usurp its rival. For owners of iPhones they’ve had a lot of success in grabbing mindshare in the USA. And it’s actually quite fascinating: many users don’t even realize they’re using Apple Maps, and remain adamant they’re using Google!

What of the future? Will some 20 person disruptive team tear up the mapping world with some artificially intelligent geographic agent? It’s going to be fascinating to watch.

But looking back on the last 20 years you have to admire the grit and perseverance of the whole Google Maps team.

It was truly a Map Happening That Rocked Our World.

Final Thanks

My heartfelt thanks and acknowledgment to Bill Kilday, author of “Never Lost Again“, without whose insider view of Google Maps my story would forever be lost!

So, as a final thought: go buy Bill’s book. ³

---

Footnotes

1. The name “Keyhole” was derived from the KH-11 (or “Keyhole-11”) satellite, a spy satellite launched by the US National Reconnaissance Office in 1976. ↩︎
2. Although a professional version, Google Earth Pro, would continue to be sold to enterprises. ↩︎
3. No, it’s not an affiliate link. ↩︎

Avid fans of Map Happenings may recall an article titled “ChatGPT (et al) for Geospatial. You Ain’t Seen Nothing… Yet.” That was 18 months ago, which is essentially an epoch in AI terms.

Given it’s been so long I decided it was high time to do a little digging and provide an update on where things stand. Has any progress been made?

If you read the article mentioned above you might remember that back in that infantile period ChatGPT not only failed miserably at basic directions, it couldn’t even do basic maths: ¹

Today, for mathematics at least, ChatGPT looks like it’s started to pay a little more attention at school:

But what about geography? Has ChatGPT — or any other so called ‘AI’ for that matter — had any schooling?

Lets Start with the Basics

For the following examples I used ChatGPT. I also tried using Claude: it produced similar results.

My first attempt to poke and prod was to determine if ChatGPT could figure out basic information about addresses and coordinates.

Ok, let’s plot that coordinate on a map:

Oh dear, ChatGPT seems to have the location in the middle of the freeway, about 3.3 miles away from the actual address:

If you read my last missive on the topic you may recall the attempt to use ChatGPT for directions: the result was a hilarious hallucination, resulting in the need for catapults. Has it got any better? Well you won’t need catapults anymore, but you’ll still get hopelessly lost:

The Issue with LLMs

Those who have a smidgen of perspicacity will by now have realized that there is a basic flaw in my poking and prodding LLMs. It is like poking Michelin-star chefs and expecting them to produce the winning goal at the World Cup final. Yes, dear reader, we should focus on the meaning of the acronym ‘LLM’ and, in particular, the second ‘L’ . As I think most of you will know, it stands for ‘language’. LLMs are power-hungry beasts that are designed principally to spit out words — and they’re barely capable of doing anything else.

In order for any geographic intelligence to become apparent in your favorite chatbot it has to get plugged into something that understands concepts beyond human language. ChatGPT started down this path in April 2023 with the very concept of ‘plugins’. Plugins didn’t take off, and instead were quickly replaced with the concept of custom GPTs. One of the first companies to jump on the custom GPT bandwagon was Wolfram. Wolfram is best known for developing Mathematica, a powerful computational software, and perhaps more well known for Wolfram|Alpha, a computational knowledge engine. While signed into ChatGPT you can use Wolfram’s GPT to ask natural language questions regarding just about anything that Wolfram|Alpha can answer.

And it does a better job on geographic basics than vanilla ChatGPT. For example, here is Wolfram GPT’s answer to the question, “what is at 51.50104° N, 0.14207° W?” — it correctly deduces that the coordinate is at Buckingham Palace, whereas ChatGPT thinks the coordinate is about a mile away at Wellington Arch:

Ask Wolfram GPT a question even remotely more complex and it quickly falls over, but at least it doesn’t hallucinate:

Perhaps this is unfair to the primarily mathematical Wolfram GPT. Perhaps instead I should use another custom GPT that is more capable of understanding geography and geospatial functions?

Good luck finding one.

The problem with the search function in ChatGPT’s “GPT Store” is that the search results are singularly execrable — to put it mildly: they are an unfilterable quagmire of useless slop.

If you search for geospatial GPTs in ChatGPT’s GPT store it will come back with a seemingly never ending list, but you’ll quickly find that the vast majority of them are nothing more that textual guides to try to help you navigate Geographically Insidious Systems (‘GIS’). It’s impossible to find the nuggets. And asking ChatGPT directly isn’t much help either.

But we shouldn’t overlook another key issue with trying to use LLMs for geospatial …

They fail to provide a map-centric UI.

But then conversely, if you use traditional consumer or professional geospatial apps, you will find that they lack the deliciousness of a natural language interface that LLMs provide.

You can see this with a simple query to Google Maps where I ask for “grocery stores within 2 miles of 311 E Congress St Tucson, AZ 85701 United States that are open after 8pm”:

Google Maps fails to search on “grocery stores within 2 miles” and “that are open after 8pm”. It can only find the address:

I should also note that if you try the same query on Apple Maps you will find it fairs no better.

But Are There Signs of Progress?

It turns out there are a number of areas where AI is beginning to seep in to the world of geospatial, and it’s just starting to get interesting.

First let’s go back to the Google query and use it to poke once again at ChatGPT:

At first glance: the results look amazing — especially compared to the non-answers from Google Maps and Apple Maps. But dig a little deeper and you’ll find the results aren’t so rosy:

• The first result is acceptable, it is roughly within two miles.
• The second result just does not exist: it’s an empty lot.
• The third result may well be open 24 hours, but it’s over three miles away.
• More importantly perhaps: the results missed an amazing local store called Time Market, which is just one mile away and is open until 10pm.

So lovely, friendly answers, but entirely untrustworthy. For the query above ChatGPT claimed to talk to Bing as well as a bunch of yellow page and grocery store sites. Yellow page sites? Are you surprised it didn’t work?!

Now my tinkering is just that. I am not even scratching the surface. However, it turns out there is a highly respected institution that’s been doing some much more serious digging into the question of geospatial AI. That organization is the Alan Turing Institute who was commissioned to write an independent report on Geospatial AI for Land Use, specifically to “to explore how land use decision making can be supported by geospatial data, including earth observation data, and artificial intelligence”.

In the report they include a section on the value of LLMs:

The point therefore is to make the data more accessible, enabling a non-geospatial analyst to get the answers they’re looking for by simply asking natural language questions . The Turing Institute explored how a map-centric UI with a built in specialized LLM chatbot could help broaden the audience, in their case with an app called ‘Demoland’:

Here is typical conversation with the chatbot in Demoland:

Question: In areas of high deprivation has access to jobs improved? 

Answer: In approximately 15.4% of the areas with high levels of deprivation, job accessibility has improved. This suggests that while some progress has been made, there is still a significant proportion of high deprivation areas where job accessibility has not improved. It’s important to continue efforts to improve job accessibility in these areas, as this can play a key role in reducing deprivation. 

Question: How has air quality changed around schools? 

Answer: Sure, let’s take a look at the changes in air quality around some schools: 

1. St Joseph’s Roman Catholic Aided Primary School: The air pollution index has increased by 0.1074 units. This school is located in Gateshead. 
2. Newcastle Bridges School: The air pollution index has increased slightly by 0.0479 units. 
3. Alternative Education Service – The Beacon Centre: There has been no change in the air pollution index at this location in South Tyneside. 
4. Yamaha Music School: Similarly, there has been no change in the air pollution index at this location in North Tyneside. 
5. New Silksworth Academy Junior and Infant: Both these schools, located in Sunderland, have also seen no change in the air pollution index. 

Please note that these changes are relative to the current air pollution index at each location. 

This certainly seems to be very promising. One of the key recommendations coming out of the report is to develop a set of Geospatial AI Toolkits for LLMs:

Clearly this is precisely what we’re missing from geospatial apps — it’s evidenced not only by my tinkering but by the institutes performing much more serious study. But when those geospatial AI toolkits are integrated into LLMs the results are very promising.

I’m sure there are many organizations, large and small, focused on such integration. One startup I’ve come across that is doing precisely this is Aino. Their natural language interface is used for fairly atomic queries as this point, but it still significantly lowers the barrier to entry relative to a traditional Geographically Insidious System.

Here’s a demo of Aino’s platform being used for site selection (click to play):

But what about the larger organizations? What are they doing?

One notable and recent example of employing AI and natural language interfaces to query geospatial data has been the collaboration between Microsoft and NASA for something Microsoft is calling “Earth Copilot”.

If you follow Microsoft then you’ll know that “Copilot” is the brand they use to indicate integration of AI into their products. You can think of it like their version of “Apple Intelligence”. You might think, therefore, that Earth Copilot is Microsoft’s attempt to turn the whole planet into a computer to answer the Ultimate Question of Life, The Universe, and Everything.

Alas, it’s not quite that bold. But it is designed to help non-geospatial analysts quickly answer questions about the planet from the petabytes of earth observation data that NASA has collected. Here’s Microsoft’s CEO, Satya Nadella, announcing Earth Copilot at Microsoft Ignite 2024 just a few days ago:

Now we all know Microsoft has a tendency to pre-announce something as a product well before it’s generally available, and Earth Copilot is no doubt one of those cases. Nonetheless, the announcement paints a clear vision of what’s (hopefully) coming to all of us.

An American might summarize all this by saying: “We want to democratize access to geospatial data and geospatial systems”.

However, a Brit might simply say: “Let’s make stodgy and dodgy mapping technology work for everyone!”²

Beyond Natural Language Interfaces

When contemplating using AI for geography we shouldn’t be myopic and limit our thinking to the use of LLMs and natural language interfaces. Indeed there are two other important use cases for AI that are worth exploring:

• Use of AI for classification of objects on the ground
• Use of AI for positioning — i.e. locating oneself

Let’s dig into this a little further…

Object Classification

The concept of using AI for object classification is not new. Indeed Google Photos started giving users the ability to search for objects inside photos back in 2015. That’s almost ten years ago. So back then you could attempt to search your photo library for “cats” or “cars” and the app would miraculously list the photos with cats or cars. In 2024 this has now advanced significantly and with the Ask Photos feature you can ask questions like “show me the best photo from each national park I visited”.

Now this approach also works in the geospatial world for identifying objects on the ground. In fact organizations like NASA and the US Department of Defense were experimenting with rudimentary approaches to automatically identify like forests and urban areas back in the 1980s. However, it wasn’t until recently, with the advent of machine learning, neural networks and sheer GPU horsepower that the capability has really taken off.

Obviously there’s a great deal of work going on in this space that we don’t know about. For example, it has tremendous applicability to military applications. So no doubt the likes of Palantir are focused on this problem full bore.

For the rest of us though there’s another company that enabled some enormous progress. And that company is …

Meta.

Meta’s research labs have developed and released something they call “SAM”, otherwise known as the “Segment Anything Model”. By ‘segmenting’ Meta means classifying things or objects that are present in a photographic image. And Meta’s lofty goal is to live up to its tag line: they want to be able to recognize any object or thing in any image.

You can learn a bit more about SAM by visiting the Segment Anything website. Here’s a demo from the site (you may need to tap/click to play):

Why is SAM relevant to geospatial? Well it’s been used by geospatial organizations to interpret aerial and satellite imagery very effectively. One such organization that is successfully employing SAM is that age-old enterprise mapping technology juggernaut: Esri.

Esri’s R&D arm has published something they call “Text SAM” that combines the power of Meta’s SAM with another open source model called “Grounding DINO” which can find objects given a text prompt.³ The results are pretty impressive:

You could argue that Esri has not actually invented anything here. They are merely integrating capabilities developed by other organizations into an Esri workflow. Indeed this may be true, but their work still illustrates great potential.

Imagine if you were to combine this object recognition capability with the power of a natural language interface:

• “Produce a map of all the buildings under construction in the suburbs of Chicago together with an accompanying chart that shows total area under construction by ZIP Code”
• “Compare the tree cover for the years 2016 to 2024 and produce a chart showing changes by county over the years”

Positioning

The challenge of figuring out your location is not new. It’s also a problem that has by no means been solved. In fact with GPS jamming it’s getting worse. BTW: if you’re interested in the history of positioning technology then I recommend you read “12 Map Happenings that Rocked our World: Part 8 — Oh Brother, Where Art Thou?“

I’m sure you’ve all experienced the frustration of trying to locate yourself.

One of the most confounding issues facing anybody who has used a navigation app is “which way do I go first?”. For example, this is particularly perplexing when exiting a subway station: should I start by going left or by going right?

In an effort to solve this problem organizations like Google and Apple have started to provide augmented reality guided solutions to help people get off on the right foot. Basically you are told to hold up your phone and wave it around the buildings in front of you. It will then magically recognize exactly where you are and promptly encourage you to go in a specific direction.

Apple calls this “Visual Localization”, but more recently the industry has been calling this “VPS” which stands for “Visual Positioning System”.

Now it turns out providing VPS requires collecting a tremendous amount of imagery data, and so this feature is generally only available in a few cities around the world.

But there is an organization that is about to change all that.

Their name is Niantic, and they’re probably most famous for building the incredibly popular game: Pokémon GO.

Now what you might not know is that the super smart guys who formed Niantic also invented Google Earth and later made Google Maps what it is today. Niantic started as a separate organization within the greater Google empire back in 2010 and was later spun off as an independent company in 2015.

In 2016 Niantic launched Pokéman GO. This game quickly became the most popular — and most profitable — augmented reality app of all time.

But just a few weeks ago, on November 12, Niantic announced they were pioneering something well beyond a popular game:

They announced that they were pioneering something they call a “Large Geospatial Model”.

In other words Niantic has set out to replace the ‘L’ (for ‘Language’) in LLMs with a ‘G’ (for ‘Geospatial’) to create some called an ‘LGM’.

But what, exactly, does this all mean?

Well to quote from their blog:

At Niantic, we are pioneering the concept of a Large Geospatial Model that will use large-scale machine learning to understand a scene and connect it to millions of other scenes globally.

When you look at a familiar type of structure – whether it’s a church, a statue, or a town square – it’s fairly easy to imagine what it might look like from other angles, even if you haven’t seen it from all sides. As humans, we have “spatial understanding” that means we can fill in these details based on countless similar scenes we’ve encountered before. But for machines, this task is extraordinarily difficult. Even the most advanced AI models today struggle to visualize and infer missing parts of a scene, or to imagine a place from a new angle. This is about to change: Spatial intelligence is the next frontier of AI models.

Even with all the scans that Niantic collects — they have 10 million scanned locations around the world and are receiving about 1 million fresh scans each week, each containing hundreds of discrete images — it’s a drop in the bucket. There is simply not enough data to make VPS pervasive, even with Niantic’s scale.

But that’s where Niantic’s LGM will come in: it will essentially fill in the blanks:

Imagine yourself standing behind a church. Let us assume the closest local model has seen only the front entrance of that church, and thus, it will not be able to tell you where you are. The model has never seen the back of that building. But on a global scale, we have seen a lot of churches, thousands of them, all captured by their respective local models at other places worldwide. No church is the same, but many share common characteristics. An LGM is a way to access that distributed knowledge.

An LGM distills common information in a global large-scale model that enables communication and data sharing across local models. An LGM would be able to internalize the concept of a church, and, furthermore, how these buildings are commonly structured. Even if, for a specific location, we have only mapped the entrance of a church, an LGM would be able to make an intelligent guess about what the back of the building looks like, based on thousands of churches it has seen before. Therefore, the LGM allows for unprecedented robustness in positioning, even from viewpoints and angles that the VPS has never seen.

The global model implements a centralized understanding of the world, entirely derived from geospatial and visual data.

With Niantic LGM the goal is to be able to locate you to within a few centimeters — just about anywhere on the planet.

Essentially it will be your ultimate hack for GeoGuessr … 😱

So What Does The Future Hold?

As is usual for all things AI, things are moving fast. Here are just a few predictions as to where we’ll end up…

Converse With Any Geospatial App

Gone will be the days of brain dead search boxes. You will now expect any mapping app, consumer or professional, to be not only capable of understanding what you type into a text box, but you’ll also expect to be able to have a conversation with it. So answers to questions like the one I tried asking Google Maps above will be answered flawlessly and with the ability to maintain context.

This capability will obviously be of tremendous benefit to the billions of people that use consumer mapping apps every day, but where it’s really going to make a huge difference is in professional apps, in many cases eliminating the need for professionally trained “GIS analysts”.

Now the department or individual that normally has to rely on the geospatial geeks in the basement will be able to get the answers they’re looking for themselves. You only have to look at the queries used in the “Demoland” app that the Alan Turing Institute highlighted to understand where it’s all going.

Gone will be the days of contingent values, complex edges, feature barriers and topological unfathomables. You’ll just ask questions.

For example, I can imagine how a map-centric site selection app might work in this world. No buttons. No pull-downs. No ribbons. Just an interactive map, a query text box and an icon to invoke the microphone:

• “show me the areas in west London that have a median income greater than £65,000 and have a higher density of Millenials”…
• “now show me all available retail locations less than 200 square meters that are within a five minute walk of these areas”…
• “now eliminate all potential locations that are within a three minute walk of a Lululemon store”…
• “ok. now give me a list of those locations together with the contact details of the leasing agent”

Get ready. It’s going to happen.

Locate Me and Orient Me — Immediately, Anywhere

When they work, satellite positioning systems are great. But they don’t work in concrete canyons. And they don’t work indoors. They don’t work underground. They get spoofed. Unless you invest in very expensive equipment, they’re mostly accurate to only few meters. And, when you start your journey, they’re not good at telling you which way to turn.

VPS is beginning to change that. But it relies on pre-scanning the area (outside or inside) to make it all work. And to do that globally, everywhere is plainly not achievable — especially for areas which are not drivable. We need to add some intelligence to fill in the blanks. And it’s organizations like Niantic that are showing the way.

In this new world HD location will be a pervasive entitlement. Not being able to locate yourself precisely, accurately, anywhere — particularly indoors — will be expected. Just like Wi-Fi, it’ll be a new layer in Maslow’s hierarchy of needs.

A World of Geospatial Agents

Now I haven’t mentioned AI agents thus far. But there’s a lot of buzz about them across many companies and industries.⁴ As you might expect these creatures will be lurking in the background, monitoring, checking, acting on your behalf. You can imagine their use in consumer apps for keeping an eye on on prices, availability, and taking actions when certain specified conditions are met — think reservations for that table at a popular restaurant, or perhaps that amazing Airbnb villa in Tuscany that you’ve always wanted to stay at.

In the professional world it will enable you to be productive in other ways while the geospatial AI agents are doing grunt work on your behalf. In a site selection app, keeping an eye out for properties that meet your requirements. Or in a public safety app, automatically alerting you and others should dangers arise.

Agentive AI for geospatial is going to be huge. And we haven’t even started.

---

Footnotes

1. BTW, with the exception of American spelling this article is written in English, so it’s “maths”, not “math”. Q: did you sadly only study a singular “mathematic” at school? ↩︎
2. I can envisage a new marketing campaign for an innovative new mapping startup: “Not stodgy and dodgy — easy peasy, lemon squeezy”. Who says I don’t have a second career as a CMO? ↩︎
3. ChatGPT’s comparison of SAM vs. Grounding DINO. ↩︎
4. And more only yesterday: Google’s AI enters its ‘agentic era’ ↩︎