A property listing description reads well — clean rhythm, no exclamation marks, one nice line about the light.

The light is the problem. The copy says all-day sun through the north-facing living area. The apartment faces south-west.

Nobody lied. Nobody sat down and decided to say a false thing about the property. The description was drafted from the agency’s own listing data, some of which was itself inherited from a previous campaign, and by the time it reached a human it was already fluent, already formatted, already 90% right. Reading it felt like proofreading, not like checking. Those are different activities and they use different parts of your brain, and the sentence about the sun sailed straight through the first one.

If that description had gone live and a buyer had relied on it, the Complaints Assessment Committee would not have been interested in any of the above. It would have been interested in one thing: the name at the bottom.

The manifesto

In June a group of about forty-five product leaders — including people whose names you’d know if you work anywhere near software — published something called the Makers Manifesto. It takes its shape openly from the Agile Manifesto: a preamble, four value pairs, sixteen principles. It’s short. It’s readable. You can get through it in five minutes.

The fourth value pair is Human accountability over full automation, glossed like this: you can delegate decisions; you cannot delegate accountability. Under it sits principle fifteen, match the standard to the stakes — guardrails should scale with the severity of the risk and with whether it can be reversed. And then the last principle in the document, three words long.

Humans sign the work.

I read it twice and then went and looked at the listing description again.

Because here is the thing about that sentence. For the product world it is a hard-won position. Forty-five people flew to a room in London in May to argue about it, and one of them has since written publicly that most organisations will nod along, paste it into an onboarding deck, and change nothing structural — because companies rarely fail at knowing what good looks like. They fail at building the conditions that make good the cheapest path.

For a licensed salesperson in New Zealand, “humans sign the work” is not a position. It’s the Act.

We’ve been running this experiment for 15 years

The Real Estate Agents Act does not regulate agencies in the abstract. It licenses people. When a Complaints Assessment Committee finds unsatisfactory conduct, it finds it against a named licensee, and that finding attaches to that licensee, and in most cases it is published with the name attached. There is no corporate veil to stand behind and no process to blame.

Two features of that regime matter enormously right now and almost nobody in the technology conversation knows about them.

The first is that unsatisfactory conduct does not require intent. You do not have to have meant to mislead. The Client Care Rules require a licensee to exercise skill, care, competence and diligence, and to refrain from withholding information that in fairness ought to be provided. A salesperson who publishes a description they did not write, did not verify, and did not understand has not accidentally avoided the rule. They have walked into the centre of it. “The system generated it” is not a defence. It is barely a mitigating factor. If anything it is an admission that the diligence obligation was never engaged.

The second is supervision. Section 50 requires that a salesperson’s work be properly supervised and managed by an agent or branch manager. The statute imagines a chain: the agency supervises the salesperson, and the salesperson exercises judgement over their own output. That chain now has a layer in it that Parliament never contemplated — a layer that produces work product, is not licensed, cannot be supervised in any meaningful sense, and answers to nobody.

And the supervision obligation did not move. It still lands on the branch manager. Whose review capacity did not increase, while the volume of material to review just multiplied.

The stakes are the wrong way round

Principle fifteen says to size your guardrails to the risk and to whether the damage can be undone. It’s a good principle. It is also, I’d argue, the most easily gamed line in the whole document, because “proportionate” is a judgement and judgements made without evidence quietly become preferences and preferences reliably converge on whatever is cheapest.

Apply it honestly to our sector and you get an uncomfortable result.

For the agency, a listing is one of several hundred this year. For the vendor it is the largest asset they will ever sell, sold once, with roughly three weeks of genuine market attention that cannot be repeated. For the buyer it is a decision made under time pressure with imperfect information, financed by a thirty-year commitment, and effectively irreversible the moment the contract goes unconditional.

So the asymmetry runs exactly the wrong way. The party with the low stakes owns the tooling decisions. The party with the irreversible stakes has no visibility into how the material in front of them was produced.

Look at where these tools are being pointed. Not at back-office admin, where a mistake is cheap and recoverable. At the appraisal, at the listing copy, at the marketing photographs — the three artefacts with the greatest capacity to induce reliance. Enhancement software that lifts shadows and warms a grey afternoon crosses into misrepresentation somewhere along its slider, and the software does not draw that line. A licensee does. Or doesn’t.

What a signature needs underneath it

I want to be direct about my own interest here, because I build in this space.

Read enough disciplinary decisions and a pattern emerges that has very little to do with dishonesty. The recurring failure is not the misstatement. It’s the absence of a record. The vendor mentioned the deck was consented; did anyone ask to see the paperwork, was the answer written down, and can the licensee produce that note eighteen months later when someone asks what was known and when. In the decisions that go badly, the licensee is very often telling the truth about what they believed at the time — and has nothing whatsoever to show for it.

That gap was survivable when a listing description took an hour to write and every sentence passed through a human who had stood in the property. It becomes structural when the description takes eleven seconds.

This is what I mean when I talk about compliance memory, and it’s why Resaido exists. Not a tool that produces the answer. A record of the judgement: what was decided, on what basis, by whom, and what you would have to produce if a Committee asked. Principle three of the manifesto says features no longer create moats. In regulated work the feature was never the moat. The file was.

You cannot sign what you cannot reconstruct.

One last thing

The manifesto’s website offers two download buttons. One says Download for humans. The other says Download for agents.

In my industry that word means something else, and I sat with the collision for longer than I’d like to admit. A document about human accountability, being downloaded at nearly equal rates by people and by machines, in a profession where the word “agent” already denotes a person who acts for another and is answerable for how they do it.

The old meaning has a duty attached. The new one doesn’t.

I would send the agency one line for the listing description: Check the aspect. If it came back corrected in 3 minutes, it would be fine, and which is also the entire point. The tool was never the risk. The 3 minutes were.

Written from the perspectives of real estate agency in Auckland — compliance practice, disciplinary decisions, and the tools being built for regulated work in New Zealand. Less commentary on the industry than notes from within it.

Subscribe if your name goes on things.