Messy soup, AI agents, and what is now being sold into surveying

The world is loud right now.

Wars. Economic instability. Cost of living. Regulatory churn across the built environment. Public trust shifting under the profession on multiple fronts.

I have started calling this period messy soup. The technical term is polycrisis. Both describe the same thing. Interlocking crises that do not resolve cleanly because they keep feeding each other. Adam Tooze popularised polycrisis, and the World Economic Forum runs with it. I prefer messy soup. It captures the era we are now working inside more honestly.

Layered on top of that messy soup is the most disruptive technology of our lifetime. And it is not slowing down.The most useful illustration of how badly this can go came from inside Meta itself.

When the AI safety lead loses control of her own AI

On 23 February, Summer Yue, Director of Alignment at Meta Superintelligence Labs, gave an AI agent called OpenClaw access to her real inbox. She had been testing the tool for weeks on a smaller test inbox without issue.

Her instruction was explicit. Suggest what to archive or delete. Do not action anything until I confirm.

The agent began mass-deleting her primary inbox. Over 200 emails. She typed Do not do that. She typed Stop, do not do anything. She typed STOP OPENCLAW. The agent kept going.

She had to physically run to her Mac mini and kill the processes to stop it. Her own words: like I was defusing a bomb.

The technical post-mortem is the part every firm owner needs to understand.

The agent ran out of working memory while triaging her real inbox. To make space, it compacted older context, and her safety instruction was silently summarised away in the process. By the time the deletion started, the rule no longer existed in the agent’s working memory.

The person whose job is to keep AI aligned with human intent watched her own AI ignore her, and could only stop it by unplugging the machine.

If that can happen to her, it can happen to anyone. Its scary to think someone who has that level of expertise was so exposed. How on earth is the everyday person supposed to get it right – I envisage huge issues ahead.

Two patterns I am watching closely inside the profession

The first is the back-bedroom AI economy. Individuals with minimal surveying background or knowledge are building tools and selling them into the profession. AI defect identifiers for home buyers. AI report writers. Some are technically capable. Most ship without governance, without documented training data, without defensible vendor due diligence, and without accountability when something goes wrong. They are sold, the surveyor uses them, and the surveyor carries the professional liability when the output fails.

The second is whitelabelling. Surveying firms are increasingly taking off-the-shelf AI tools, applying their own branding, and selling them. This is rising fast and the risk profile is genuinely concerning.

When a firm whitelabels someone else’s AI, the firm carries every output as its own work. The vendor underneath can update the model, change the data handling, alter the behaviour, and the firm finds out only when something has already gone wrong. The client believes they are buying the firm’s expertise. They are buying a third-party tool with the firm’s logo over the top. Disclosure becomes harder. Vendor due diligence becomes harder. The named responsible person inside the firm is accountable for a system they do not fully understand and cannot fully audit.

The reassurance the profession needs to retire

We are in a period of transition. Nothing about this moment is fixed. The risks today will not necessarily be the risks in eighteen months. The shape of the work today will not be the shape of the work in two years.

That is why one line still circulating widely across the profession needs to be retired.

“AI will not take surveyors’ jobs.”

It is comforting. It is also incomplete, and it is starting to do harm.

AI is going to change surveying work. The degree of that change will not be uniform across the profession. It will depend on a handful of variables every firm owner needs to think about clearly:

Specialism. A geospatial surveyor and a residential surveyor face very different exposure profiles.

What clients ask for. Volume work bought on price will look different from instruction-led, judgment-heavy work.

What the public demands. Trust, transparency, and disclosure expectations are shifting fast.

Pricing pressure. AI changes the cost base of every firm using it, and margins will move accordingly.

Regulation. The RICS Standard has landed. The EU AI Act is live. ICO enforcement is sharpening. More is coming.

The people inside your firm. Most firms now have employees already using AI tools the firm has not approved or assessed.

The honest answer to “will AI take surveyors’ jobs” is that AI will reshape the work. The firms that pretend otherwise are lining themselves up to be caught out.

Naming what is actually changing is what allows clear-headed planning.

Try it. My recommendation to the sceptics.

If you are still on the fence about AI, my recommendation is simple. Try it.

Every surveyor I have suggested testing AI has come back to me genuinely surprised by what it can do. Most of the disappointment I hear about AI traces back to how it was used. Vague prompts, no context, no follow-up, no critical evaluation of the output. Becoming good at prompting is a skill worth mastering quickly, and the bar to get to a useful level is lower than people assume.

I use Claude, ChatGPT, Gemini, Lovable, Gamma and many others across my work. Capability rankings shift constantly. One month ChatGPT is ahead on something. The next month Gemini takes that lead. The model landscape is genuinely fluid.

My preferred tool right now is Claude. It is excellent at writing. It is more conscious about data security and safety than the alternatives. It handles edge cases more ethically, and across the work I do day to day it is just better all round.

A few practical notes on using AI responsibly:

Pay for the proper enterprise tier if you are putting any business data through a model.

If you are using a free or lower tier, never enter confidential client information. Treat those tiers as public.

Use AI as a critical thinking partner. Have it pressure-test your reasoning. Make it argue against you. Ask it where you are wrong.

Always apply your own professional judgment to the output before it goes anywhere near a client.

I have learned more in the past four years using AI properly than I had in any equivalent period before it. The tooling is genuinely incredible when it is used well, and the surveyors who learn to work with it now will be the ones running stronger firms two years from now.

The calm bit

This is a lot. I know it.

I am laying it out like this for a reason. Surveying is a profession built on risk assessment. Surveyors are trained to look at what could go wrong and price it in. That instinct is more valuable now than at any point in my career working with the profession.

The firms that will come through this period of transition well are the ones treating AI exactly the way they already treat structural risk. Identify it. Document it. Decide what to accept, what to mitigate, what to refuse. Apply professional judgment to every output before it goes anywhere near a client.

That is a professional posture, and it is what separates the firms that hold their position over the next two years from the firms that get caught out.

Free AI briefing on Wednesday 20 May

If you want to work out where your firm sits inside all of this, I am running a free 60-minute briefing on Wednesday 20 May at 1pm BST.

It covers the two questions most firm owners are asking me right now. Does the RICS AI Standard apply to my firm. And if it does, what do I actually need to do first.

It is informal CPD. The recording is included for 14 days for anyone who cannot make the live session.

Over 1,000 firms have already taken the readiness assessment that sits behind this briefing. The patterns are clear. 92% of firms have no documented process for verifying AI outputs.

Reserve your place: https://www.rics-ai-briefing.surveyors-uk.com

Until next week,

Nina