Best laid plans
New planning tools are rolling out to local authorities. It's a big test of AI for government.
Everyone is hunting for AI-shaped problems.
You can see it in startups engaged in a gold rush to crack automating all kinds of workflows, no matter how niche. You can see it in big businesses, aware they need adoption. And you can see it in government, where everyone knows there are legacy systems that could benefit from a technical breakthrough, but fears things may already be too broken.
At present, there are several projects inside the UK government driven by an optimistic view that yes, we can solve some issues with tech. In these corners, the finding of AI-shaped problems is treated in a suitably mathematical way: teams come up with options, whittle them down through elimination, then build and test their ideas to see if they work.
A couple of weeks ago, I saw a handful of these demonstrated at the Society for Technological Advancement (SoTA) Frontiers Night featuring the Incubator for Artificial Intelligence (i.ai), a small but determined set of engineers and project managers in the heart of Whitehall. It was my third time seeing members of this cohort present, and one particular product has come up every time. It’s called Extract.
Here’s how the pitch usually goes. At this very moment, underneath your local council building, there is probably a dingy cellar filled almost to the brim with paper.
This is where planning documents go to never see the light of day again. Local authority employees have allegedly been known to wish for a convenient flood.
But what if they could all be turned into data? And what if that data could be used to help speed up planning decisions and, through a meme-worthy domino effect, eventually get the Government closer to achieving its 1.5 million new homes?
Extract, developed by i.ai alongside the digital planning team in the Ministry for Housing, Communities, and Local Government (MHCLG) is an AI tool that converts old – sometimes hand-edited, sometimes poorly photocopied documents – into a usable geospatial database.
The product has become one of the poster children for the Government’s embrace of AI. Almost a year ago, the Prime Minister spoke about it at London Tech Week, pledging that it would be made available to all councils by Spring 2026.
From what I understand, this rollout is now within reach and should be happening next month. “Spring”, after all, ends in late June, astronomically speaking – and in Whitehall terms.
The other big development is that work on an augmented planning tool is now getting underway, after Google DeepMind and Faculty won an £8.2m contract to develop it in February. The contract for that procurement has now been signed.
What happens next with Extract and the augmented planning tool will be a significant test case for this era of nimble AI projects hoping to have outsized impacts on government. The rollout of Extract will be the biggest expansion of an i.ai tool beyond Whitehall (a transcription app, Minute, is used by just over 20 councils, but this will be going out to all 300+). Meanwhile the AI-assisted planning decisions tool raises a lot of questions. I’ve been speaking to people who watch this area closely in the past couple of weeks and though many would like to be optimistic, they foresee challenges ahead. It could either prove controversial – giving political fodder to campaigners on either side to blame a “bot” for decisions they don’t like – or may even be ineffective for the ultimate goal of getting more homes built, if it simply replicates past decisions.
Green and pleasant data
Extract is nowhere near the first government tool to be aimed at upgrading planning data. MHCLG already funds local authorities to do data remediation work through the Open Digital Planning programme. There are also other projects like PlanX, a software co-designed with local planning authorities (LPAs).
That image of the basement overstuffed with paper is an evocative one, but it isn’t the full story. Plenty of work has already been done to digitise planning documents.
What Extract could do is level this up. Using the advanced visual reasoning of Google’s Gemini model, it can turn documents into easily usable maps. Even if a document is already available in PDF form, the proposition of Extract is that it makes it far easier to turn that into geospatial data.
Steve Messer, who worked on Extract from the MHCLG side, told me it’s a “growth hack” for the existing planning data platform. Some successful work has already been done to get councils publishing their planning data to that platform more quickly. The hope is that Extract will speed things up further.
This might have effects beyond government. A problem that proptech software providers face in this space is the lack of reliable, up-to-date data. That’s a barrier for creating innovative AI tools. “Realistically, any other AI won’t have any trustworthy data to use unless the planning data platform has datasets,” Messer said.
The big bet on planning
The origin of Extract was in meetings between i.ai and MHCLG in late 2024. But the inception of the augmented planning tool stemmed from the arrival of Jade Leung, the Prime Minister’s AI adviser, to Number 10.
She pushed the incubator team to look at some big bets. This led to a focus on planning, and eventually to a procurement process. Housing minister Matthew Pennycook confirmed in the Commons in March that the government intended to award Google and Faculty an £8.2m contract to develop an AI-powered planning tool. Trials in England should start this month.
The big hope for this initiative is that it will dramatically reduce planning application processing times, with the initial goal of slashing the eight-week wait in half to four weeks. Part of what would make this possible will be the data digitised by Extract.
This tool though will probably face its fair share of skepticism, not least from planners themselves.
Messer has previously pointed out in a blogpost that automating decisions which have traditionally been made by highly accredited professions is not necessarily straightforward. The Royal Town Planning Institute (RTPI), for example, requires members to set out their decision-making process.
Indeed, Daniel Slade of the RTPI told the FT this week that AI should never replace professionals, and that there was a danger of redesigning the planning system itself for the benefit of the technology. “There’s no value in processing applications more quickly if the developments that follow are low quality,” he said.
It could also be a tricky sell with the general public. One can imagine the headlines: “Bots build on green belt!” Councillors looking for a reason to block a development that their constituents don’t like might see an opportunity in rejecting the claims of a computer, even though with this system, humans will make the final decisions once AI has compiled a recommendation.
You might expect the people advocating for planning reform and building new homes to be excited by this development, but many of those I spoke to gave something closer to a shrug. Speeding up decisions won’t help solve the planning system’s biggest issues or help gather the political will to change it, they said.
“Like with so much else in the government’s AI strategy, it will be a well-intentioned distraction from fixing the bigger problems, which are constrained not by technology but by political economy,” Jamie Rumbelow, formerly founder of proptech company Tract, told me.
The project might be a big swing, but many think we need an even bigger one to fix planning. Still, if planning decisions really do become twice as fast, there’s an opportunity for something the government is looking for: ways to show the public real, tangible benefits of AI.
Teatime scroll Each week I share links to writings, events, tweets and other conversation-starters. If you have something you think should be in here, feel free to email or DM me.
🪺 Coming up next week, I’ll be at Land Talks on Monday and Restless Egg on Thursday.
⏳ The next Long Now London meeting is taking place on Wednesday.
🧑🎨 Really enjoyed this piece from M.J. Hines, arguing that the UK needs a new creative director.
🗳️ For the New Statesman, I explored the world of synthetic polling and asked where the line should be with this technology.
📳 The return of Pornhub to the UK is an interesting development in the online age verification debate. Management says it can now rely on device-level age checks thanks to an Apple update. Is this going to resolve some of the angst over how to enforce age controls online?
🕚 Elevenlabs revealed some new investors, including Eva Longoria.
🔨 I wrote about the skills gap for The Morning Intelligence this week. Tom is also hosting an event on Monday to discuss the fallout of local elections for tech policy.
🧟 My friend Jonathan wrote this fascinating essay about technological necropolitics.
🍎 Over on the Progress Post, I argued that Oak Apple Day should be revived and made into a new bank holiday. Cheers!