Care, Coordination, and What a Practice Is Paid For

Lesson one established that the standard is care rather than correctness. The practical consequence is that a practice needs to be able to show the care, and generated work makes that harder unless it is deliberate.

The difficulty is specific. When a junior produces a drawing and a senior redlines it, the redlines are evidence. There is a marked-up print, a comment thread, a revision. The review left a trace as a by-product of how it happened.

When a professional reviews generated output on screen and edits it in place, the review leaves nothing. The final document is identical whether it was scrutinised for an hour or accepted in ten seconds.

That matters years later, when a defect surfaces and the question is what the practice did at the time. Absence of evidence of care is not the same as evidence of absence, but in a dispute it is uncomfortably close.

What works, proportionate to a small practice.

Record which decisions were yours. A short note in the project file saying the system selection, the performance criteria and the standards were determined by the design team, and the specification text was drafted from those decisions.

Keep the verification artefacts. If you checked every cited standard, the list of what you checked is the evidence.

And note the checks in the review record you already keep, rather than inventing a new system.

None of that is heavy. It is the difference between saying we reviewed it and being able to show what reviewing meant.

Professional indemnity insurance is the financial backstop of a design practice, and it is worth understanding how it interacts with new tooling rather than assuming either that it covers everything or that it covers nothing.

The basic structure. Professional indemnity responds to claims arising from negligent acts, errors or omissions in the performance of professional services. It is built around the standard of care: if you met it, there is generally no negligence to respond to; if you fell below it, the policy is what pays.

Using a tool does not by itself change that analysis. A negligently produced specification is negligent whether a person or a system drafted it, and the policy responds to the professional service, not to the method.

Where the interaction becomes real is in two places worth checking rather than assuming.

Warranties and guarantees. Most policies exclude liability assumed under contract that goes beyond the ordinary standard of care. If a practice tells a client its AI-assisted process guarantees code compliance or eliminates coordination errors, that is a warranty, and warranties are commonly outside cover. The marketing claim creates the uninsured exposure, not the tool.

And insurer expectations, which are still settling. Insurers are actively developing their position on AI-assisted design, and policy wordings, application questions and renewal terms are changing across the market.

So the practical advice is short and honest. Read your policy, tell your broker what your practice actually does at renewal rather than after a claim, and do not make claims in marketing that your policy will not stand behind. A practice that promises more than the standard of care has bought an exposure no tool created.

Two intellectual property questions arise, and they run in opposite directions.

What comes out. Copyright in many jurisdictions requires human authorship, and purely machine-generated material may not attract protection in the way a drawing by a person does. For a practice this matters mainly at the margins, since the valuable, protectable work is the design expressed in the documents, which is substantially human. But a practice that licenses its drawings to clients should understand that the ownership it is granting may be less clean for wholly generated content than for authored content.

What goes in is the more immediate risk. Practices routinely hold material that is not theirs to disclose: a client's confidential brief, a developer's unannounced scheme, a consultant's proprietary detail, drawings received under a confidentiality agreement, and information about buildings whose security matters. Uploading any of that to a general consumer tool is a disclosure to a third party, and under many consumer terms it may be retained or used to improve the service.

The project-specific angle is sharper here than in most professions. An unannounced development, a competition entry, or the security arrangements of an occupied building are exactly the material a client would be most alarmed to find outside your office.

The practical positions. Use tools procured by the practice with terms addressing retention and training use. Check what your appointment agreements and confidentiality agreements actually permit, since some now address AI processing explicitly. Strip client identifiers where the task does not need them. And treat drawings of secure or sensitive buildings as a category that does not leave your systems at all.

A building is designed by several separately appointed professionals, and knowing where the boundaries fall determines who checks what.

The client appoints the architect and, depending on the procurement route, the engineers directly or through the architect. Each consultant is responsible for their own discipline and seals their own work. The architect typically coordinates, which means making the disciplines fit together, not becoming responsible for their engineering.

The contractor is responsible for means and methods, and for the work conforming to the documents.

Why that matters for tooling. A generated output crossing a discipline boundary is a specific hazard. A model asked about a structural question inside an architectural workflow will answer, fluently, and that answer has no engineer behind it. Using it, or passing it to a client as though it were coordinated advice, blurs a boundary the appointments deliberately drew.

So the rule that follows. Assisted work stays inside the discipline that is competent and insured for it. A question that belongs to another consultant gets routed to that consultant, exactly as it did before, and the speed with which a tool can produce a plausible answer is the reason to be deliberate about this rather than a reason to skip it.

Coordination remains a human act of reconciling professionals, not of merging outputs.

Two structural questions for practices, and the second is the one the profession has not resolved.

Fees first. Much design work is priced as a percentage of construction cost or as a lump sum, not hourly, so time saved in documentation flows to margin rather than to a smaller invoice. That is a straightforward gain, and it lands in the part of the business that has been squeezed for decades.

The harder question is what happens to the people who used to do the documentation.

Historically, a junior learned the profession by drawing details, writing specification sections, and processing submittals. Those tasks were the training, not just the output. You learn how a building goes together by drawing how it goes together, and you learn what a specification means by writing one and then watching a contractor price it.

If that work is compressed, the profession has to answer where the judgement comes from that lesson one said cannot be delegated. Judgement is not taught in the abstract. It accumulates from having done the detailed work and seen the consequences.

This is a genuine open question rather than one with a settled answer, and it is worth naming as such. Practices are experimenting with keeping juniors on review rather than production, on the theory that reviewing generated work builds a similar skill. Whether reviewing teaches what producing taught is not yet known.

What can be said is that a practice which uses the tools to remove junior roles entirely is consuming a training pipeline it will need, and that the effect will not be visible for a decade.

Naming what is not substitutable, which in this profession is unusually concrete.

Someone who is answerable. A client with a defective building has a licensed professional, a regulator, and an insurer. That chain of accountability is a substantial part of what is purchased, and no system provides any link in it.

Judgement about a particular place. Lesson one covered this: the tool is strong on what is typical and blind to what is particular, and a building is a response to a specific site, a specific brief and a specific set of constraints that exist in project documents and local knowledge.

Reconciling irreconcilable requirements. Real projects are over-constrained. The budget, the programme, the planning position, the client's aspirations and the physics do not all fit, and the professional act is deciding what gives. There is no stated objective function to optimise, because the client has not decided their own priorities and often discovers them through the process.

Advocacy. Arguing a scheme through a planning committee, negotiating with a contractor over a claim, holding a position with a client who wants something unwise.

And the relationship over time. Projects run for years, and much of the value is continuity: knowing why a decision was made in year one when it is questioned in year four.

The pattern matches the other professions in this catalogue, with the boundary drawn more explicitly than most. Elsewhere it is inferred from how the work behaves. Here it is a seal, a licence and an insurance policy, all naming a person.

What a practice can write down, derived from obligations that already existed.

The seal. Nobody seals work they have not personally reviewed and are not prepared to defend. Provenance changes nothing about this. Rubber-stamping generated output is plan stamping.

Codes and life safety. No generated conclusion about code compliance, structural adequacy, egress or fire performance enters a drawing set, a permit application or a client letter. The tool finds candidate provisions; a person reads the adopted code and decides.

Numbers. Quantities from the model, structural numbers from analysis software and a licensed engineer, energy from the calculation tool, costs from a cost plan. The language tool explains numbers and never produces them.

Specifications. Drafted from decisions the design team made. Every standard verified against the issuing body, every product verified as real and current, and a check for requirements nobody intended.

Discipline boundaries. Assisted work stays inside the discipline competent and insured for it. Questions belonging to another consultant go to that consultant.

Confidentiality. Approved tools only. Client briefs, unannounced schemes, competition work and anything about a secure building do not go into general tools.

Evidence. Which decisions were the design team's, and what was verified, recorded in the project file at the time.

And claims. The practice does not promise clients an outcome better than the standard of care, because that is a warranty and warranties are commonly uninsured.

Eight paragraphs, and every one of them restates a duty that predates the technology.

Pulling the three lessons together into what a design professional should actually take away.

The compression is real and it is concentrated in documentation and contract administration, which is where most of the hours are and where almost none of the professional distinction lies. A practice that captures that gets its time back in the part of the week nobody enjoyed.

The boundary is unusually clear here, and it is drawn by the seal rather than by anyone's opinion about capability. Design decisions, code conclusions, life safety and the act of sealing stay with a licensed person, because the licensing regime has no mechanism for anything else.

The verification burden is real and specific. Invented standards, invented products, code provisions without their exceptions, numbers that look right, and requirements nobody intended. Each of those is invisible in the output, which is why the checks have to be deliberate rather than a general sense of having looked.

The standard of care moves. Doing what careful practitioners do with these tools is defensible; the definition of careful is going to keep shifting, and a practice should expect to revisit its position rather than settle it once.

And the thing worth watching is not liability, which the existing framework handles. It is the training pipeline. The judgement that the seal certifies was historically built by doing the work that is now being compressed, and no profession has yet shown how it will be built otherwise.

That is the open question. Everything else in this cursus is an application of duties that were already there.