Redesigning the Workflow

Picture the old process as a line of stages: brief, research, concept, documentation, checking, issue. The bolt-on approach picks one stage - say documentation - and drops an agent into it, so that stage runs faster while everything around it stays the same. This is genuinely useful and it is the right way to begin, because it is safe and it teaches. But its ceiling is low, for a simple reason: you have optimised one link in a chain whose length is set by all the other links and by the hand-offs between them. If documentation was a fifth of the effort, halving it saves a tenth of the project, and even that gets eaten if the faster draft now waits longer for a human to get around to checking it.

Redesign asks a different and larger question: given that agents can now take on research, synthesis, drafting, cross-checking, reformatting and coordination across many stages, what should the whole process look like? The answer is not the old line with a faster box in the middle; it is a different shape. Several stages that used to be sequential and manual can run in parallel and largely unattended - an agent compiling precedents while another drafts the room schedule while a third cross-checks the drawing set against the brief. The long tedious middle of the project - the part that used to consume junior time and calendar weeks - compresses. And the human work rearranges itself around the machine work rather than being interrupted by it.

The practical test of whether you are bolting on or redesigning is where your people spend their time afterwards. If they are doing the same tasks slightly faster, you have bolted on. If the nature of their day has changed - less time producing the middle, more time defining intent sharply at the front and scrutinising output hard at the gates - you have redesigned. The redesign is more work to think through and it disturbs comfortable habits, which is exactly why most studios stop at bolting on. But the studios that get the transformational gains, rather than the marginal ones, are the ones that were willing to ask what the project should look like now, not just how to make last year's project a little quicker. And redesign is not a one-time event: because the tools move fast, the sensible cadence is to keep revisiting the shape of the process as agents get more capable - a standing question, not a project you finish.

Redesign is not uniform - some stages transform, some barely move, and knowing which is which is the core of the redesign. The stages that change most are the multi-step, tool-using, information-heavy ones that sit around the creative core: research and precedent-gathering; turning a messy brief into a structured programme; generating and comparing options against stated criteria; producing and coordinating the mass of documentation - schedules, room data sheets, specification drafts, drawing annotations; cross-checking sets for consistency and against rules; and the endless reformatting and administration. These compress hard because they are precisely the bounded, delegable work agents are good at, and because much of it can now run in parallel and largely unattended rather than as a manual relay from desk to desk.

The stages that change least are the ones that are irreducibly human, and it is worth being just as clear about these, because a redesign that tries to hand them to agents is a redesign that fails. Setting the actual design intent - the concept, the idea that is right for this place and these people - does not compress; if anything the redesign asks you to do it more sharply and earlier, because agents can only act well on intent that has been made clear. The core judgements about what is good, the taste and spatial sensibility, the ethical and safety calls, the client relationship, and the professional sign-off all stay human and stay roughly as effortful as before. What changes is that they are no longer diluted by the grind around them.

Between these two categories sits a third that the redesign transforms most interestingly: verification. In the old process, checking was woven through the work and often under-done because everyone was busy producing. In the redesigned process, because agents produce so much more, so much faster, verification becomes a distinct, load-bearing, deliberately-staffed activity - arguably the most important human stage in the whole flow. This is the central trade of agentic redesign: you move human effort off producing the middle and onto defining intent at the front and verifying output at the gates. Get that trade right and the process is both faster and safer; get it wrong - compress the middle but skimp the verification - and you have simply built a faster way to ship confident errors under your name.

If agents run more of the middle, the human role does not shrink - it relocates and, done well, becomes more valuable. It concentrates at two poles of the project, and the redesign is largely about staffing and honouring those two poles properly. The first is the front: defining intent. An agent is only as good as the goal, context and constraints you give it, so the quality of the whole downstream flow now depends heavily on how well a human frames the design intent, the brief, the criteria and the standards at the outset. This is a shift in where senior judgement is spent: less on producing documents, more on articulating - clearly enough that agents and juniors alike can act on it - what the project is trying to be and what good looks like. Time spent sharpening intent at the front pays off multiplied across everything the agents then do.

The second pole is the gates: verification and decision. In a redesigned flow, work should pass through defined human checkpoints where output is scrutinised before it is relied on, coordinated or issued - a review of the compiled research before it shapes the concept, a check of the drafted schedule before it goes into the set, a verification of anything touching code, cost or safety before it becomes a commitment, and a professional sign-off before anything leaves the door. These gates are not bureaucracy; they are where the human owns the output the agent produced. Designing them explicitly - who checks what, against what, at which point - is one of the most important parts of the redesign, because agents fail silently and fluently, and a gate is what catches the failure before it becomes your liability.

Between the poles, the human role becomes one of direction and orchestration rather than production: setting agents going, monitoring, handling the exceptions and hard cases they escalate, and stitching their outputs together with judgement. This is genuinely more skilled and more interesting work than much of the grind it replaces, which is the honest good news of the redesign for designers - but it is also more demanding of exactly the judgement a tool cannot supply, which is why the human is more essential in this shape, not less. The studio that thrives is the one that recognises this relocation and invests in it: training people to define intent sharply, staffing the verification gates seriously, and treating the ability to direct and orchestrate agents as a core professional skill rather than an afterthought.

Put the pieces together and the felt rhythm of a project changes. The old rhythm was a long, fairly even grind: weeks of steady production between the early ideas and the final issue, with checking squeezed in at the end. The redesigned rhythm is punchier and more uneven - intense, high-judgement bursts at the front (getting the intent and brief truly right) and at the gates (verifying hard), with fast, largely-supervised agent production compressing the stretches between. Iteration gets cheaper and therefore more frequent: because an agent can regenerate a set of options or redraft a schedule quickly, you can afford to explore more and commit later. Lead times on the documentation-heavy stretches shrink. And the studio can take on work, or reach a depth of exploration, that a hand-production process could not afford.

That new rhythm brings real traps, and naming them is part of designing the workflow well. The first is skimped verification: the faster the middle runs, the more tempting it is to wave output through, and a redesign that compresses production without seriously staffing the gates is a machine for shipping errors at speed. The second is losing the thread: when agents produce a lot quickly, a human can lose the deep familiarity with the work that judgement depends on, reviewing output they did not really engage with - so the redesign must keep humans genuinely in the loop, not nominally rubber-stamping. The third is fragility and lock-in: a workflow wired tightly around one fast-moving tool can break when the tool changes, so sensible redesign keeps the process legible and the human able to step in (a theme 9.3 turns into durable studio assets). The fourth is treating the redesign as finished: the tools change, so the right shape of the process keeps shifting, and the studios that win keep asking the redesign question rather than answering it once.

The honest summary is that redesigning the workflow around agents is where the transformational gains live, but it is a redesign that must be built around the human at the two poles, not one that quietly writes the human out of the middle and hopes for the best. Compress the grind; sharpen the intent; staff the gates; keep the human genuinely directing, judging and signing off. Do that, and the new shape of the project is faster, deeper and still fully professional - which is exactly the outcome the studio system in the next lesson is designed to make durable rather than dependent on one clever person.