Developing the Design Concept

Start with the boundary, because everything else in this lesson depends on it. The seed of a concept - the intuition that the house should turn its back on the road and open to the ravine, that the clinic should feel like a home and not a hospital, that the shop should be one calm room and not a corridor of shelves - comes from you. It comes from the site walk, the brief, the client's half-said hopes, your training and your taste. Claude cannot have those intuitions, because it has not stood on your site, met your client, or felt the particular pressure of this budget and this street. It also has no stake in whether the building is any good.

What Claude can do, extraordinarily well, is be the studio-mate you think out loud with at eleven at night: the one who never tires, always has another reference, and will happily argue the side you have not considered. That is a real and valuable role. But notice what it is not. It is not the author of the idea, and it is not the judge of it. The moment you type "give me a concept" into an empty chat and take back what appears, you have swapped your reading of the world for the smooth statistical average of every concept the model has ever read - which sounds profound and means nothing.

Held as a partner, Claude makes your concept stronger without making it theirs. Held as an idea vendor, it makes your concept weaker and un-owned. Every technique that follows - widening, pressure-testing, grounding, articulating - is a way of keeping the intelligence in the room pointed at an idea that started, and stays, with you. If at any point you cannot say which parts of the concept are your judgement and which are Claude's suggestion, that is the signal to close the laptop and go back to the sketchbook.

Two moves make Claude genuinely useful at concept stage, and they pull in opposite directions. The first is widening - opening more directions than you would reach alone, so your choice is made against real alternatives rather than the first idea that stuck. The second is pressure-testing - trying to break the idea you have, so what survives is defensible. Do them in that order, and always after you already hold a seed to react to.

To widen, give Claude your seed and your context, then ask for range, not answers:

I am designing a small municipal library on a tight urban corner in
Pune. My instinct is a quiet vertical stack of reading rooms around a
top-lit stair. Give me eight distinctly different conceptual
directions for the same brief and site - vary the ORGANISING idea, not
just the style. For each: one line on the core idea, one on what it
would cost me.

You are not hunting for the winner here. You are hunting for the three directions you would never have reached, the one that clarifies why your instinct is right, and the language that helps you name what you are actually after.

Then switch. Feed your chosen direction back and ask Claude to attack it:

Here is the concept I am committing to: [paste]. Act as a hard but
fair design critic. Where is this idea weak, generic, or fighting the
brief? What would a sceptical client or jury ask? Give me the five
questions I would least like to be asked.

This is where a foil earns its keep. It surfaces the soft spots while they are still cheap to fix: the concept that reads beautifully on paper but ignores the west sun; the "community heart" with no reason for anyone to linger. You are not obliged to accept a single point - you are obliged to have an answer to each. The loop in the figure captures the rhythm: you seed, Claude widens, Claude presses, you decide - round and round, with the idea staying yours the whole way.

The failure mode to fear is not a wrong concept - it is a plausible, well-dressed, generic one. Large language models are fluent by nature, and left loose they drift to the smooth middle: "a dialogue between old and new", "a seamless connection to nature", "a play of light and shadow". Every one of those is true of almost every building ever made, which means it is a real concept for none of them. The antidote is to keep dragging Claude back to the specifics only you hold.

The practical move is to feed the constraints, not just the ambition. Instead of "help me develop a concept for a farmhouse", give it the ground truth: the plot is a long north-south strip, the client cooks obsessively and hates air-conditioning, the budget is real and tight, the site floods once a decade, and there is one magnificent existing tamarind tree. Now the same tool that produced platitudes has something to bite on, and its questions and variations sharpen instead of drift. Whenever an idea comes back sounding like it could be pinned on any project, treat that as your cue to re-ground: "that is too general - tie it specifically to the tamarind and the flood line."

The figure lays out the ownership split to hold in your head throughout. On one side is what is yours and never delegated: the core idea, the design intent, what the place is really about, and every decision. On the other is what Claude can genuinely help with: more variations to react to, precedents and analogies to test against, the devil's-advocate questions, and clearer first-draft words for an idea you already hold. Read the verbs in that second column - help, react, test, draft. None of them is decide. When you can look at your concept and name exactly which parts came from your judgement and which parts Claude only helped you articulate, you are using it well. When you can no longer tell, you have crossed the line.

Here is the whole loop in one sitting. You come back from a site visit for a weekend home with a fuzzy conviction: the house should feel discovered, not announced - you approach through scrub and meet the view only at the last moment. That spark is entirely yours. Now you put Claude to work on it.

First you widen: "Here is my instinct - the reveal of the ravine withheld until the end of the approach. Give me six spatial strategies that could deliver a withheld-then-released sequence, each in one line." Claude offers compression-and-release at the entry, a blank street wall, a bent circulation spine, a level change, a screen that dissolves, a courtyard you cross blind. Two are obvious, one you already had, and one - the bent spine - makes you sit up. You combine the bent spine with the level change. That judgement is yours.

Then you pressure-test: "I am committing to a bent entry spine that hides the ravine until a final turn. Attack this - where does it fail?" Claude asks the useful, uncomfortable things. What does the daily user feel on the hundredth arrival, not the first? Does the drama cost you the morning light? Is a bent plan worth its structural and cost penalty on this budget? You keep the idea but add a plain side-door that skips the ceremony for everyday use - a real improvement you now own.

Finally you ask for articulation, not authorship: "In three sentences, state this concept plainly, in my voice - direct, no jargon." You take the draft, cut the one word that sounds like a brochure, and you have a stated concept you can defend to a client and hold across a hundred drawings. The idea was yours at the spark and yours at the finish; Claude only made the middle faster, sharper and less lonely - which, at eleven at night, is worth a great deal. That habit - spark yours, middle shared, decision yours - is the concept workflow in miniature.