A £10.64m decision, modelled by AI in 78 minutes. What's left for you?

On 1 October Antony Slumbers published an account of a morning's work. He asked Codex, OpenAI's coding agent, to build a tool comparing three options for an office building: sell now, hold with essential works, or refurbish and re-let. A working demonstration took 49 minutes. A second round of 29 minutes added guided interrogation of the numbers. Seventy-eight minutes in total, for a model that puts £10.64m on selling now.

Two cautions are worth making first. This is a first-person account of an experiment, not a formal study. The asset is synthetic; it has not been tested on a real building, and Slumbers makes that clear. He is not a surveyor either. On his own site, he describes himself as a speaker, advisor and writer on proptech and says he has founded and exited several proptech software companies.

It still deserves your attention, because the question he ends on is the one every valuer, QS and cost consultant will eventually hear from a client: if the analysis takes 78 minutes, what am I paying you for?

The tool, Asset Strategy Lab, models Alder House, a fictional 1998 regional UK office of 36,000 sq ft, 27.8% vacant, with £692,000 of passing rent. In the base case, selling now returns £10.64m, holding with essential works has a present value of £8.01m, and refurbishing comes in at £9.96m.

The domain knowledge is the striking part. Slumbers is not an asset manager. He supplied the purpose and the three strategies.

With light steering, the model decided what the spreadsheet should contain.

The most useful passage is where the model behaved oddly. Changing only the refurbishment letting delay and a longer void improved the refurbishment result. The explanation is mechanical. The delay shifts later lease expiries and rent-free periods, which changes the income falling into Year 11 and, in turn, the exit price estimated by the model. At a three-month delay, the higher calculated exit value more than offsets the cash lost during ownership.

Slumbers draws the right distinction: "The arithmetic follows the chosen rules. The economic question is whether a purchaser would value that temporary income pattern in the same way or assess the longer income stream differently."

He links RICS's practice information on discounted cash flow valuations, published in November 2023 with Neil Crosby as technical author, which says in paragraph 4.32 that "It may be inappropriate to make broad assumptions that every tenant leaves or stays unless there is strong evidence to support those positions." The tool's own model notes state that all tenants vacate and that renewals, early breaks and defaults are excluded. As built, the demonstration does the thing the guidance cautions against.

Except, and this is the honest part of the essay, the AI helped find and explain the problem too. Slumbers concludes that his role was persistence rather than expertise.

The challenge mattered, and once discovered, it became a reusable instruction and a test that the next user gets for free. As he puts it, "Some of today's human intervention becomes tomorrow's process."

Slumbers identifies four areas where he would still rely on a person:

He does not promise this adds up to a living. "Greater analytical capability does not guarantee higher fees or the same number of jobs. Clients may expect more, prices may fall, and value may concentrate elsewhere."

This lands alongside the first Litmus report, which we covered last week. Chris de Gruben, Senior Director, Head of Property for the UK and EU at Artefact and a chartered valuation surveyor, told that group: "I can automate eighty per cent of the valuation process with the tools available to me today." That is one practitioner's estimate; the report carries no survey data.

Its fifth finding, that the judgement which catches an error before it becomes a claim is being spent faster than it is built, is the exit-capitalisation quirk in one sentence.

Disclosure: Project Flux co-founder James Garner is a founding Litmus contributor in a personal capacity and sits on the RICS construction professional group panel.

As we cover elsewhere this week, Anthropic's robotics research puts office-based roles in the front line before the trades. Asset strategy modelling is about as office-based as work gets.

If a strategy comparison takes 78 minutes to build, an hourly fee for building it is finished. What the client is buying is the judgement to brief the model properly, the evidence the model cannot reach, and a signature that carries liability under the RICS professional standard on responsible use of AI, in force since 9 March 2026. Those three things are now the whole service, and they need pricing as such.

We will keep testing claims like this one against real surveying practice. If you want the next worked example before the fee conversation reaches you, subscribe to the Project Flux newsletter at projectflux.ai.

All content reflects our personal views and is not intended as professional advice or to represent any organisation.