TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · The Brief · AGE 16—18 · POSITIONAL · HELD
The prompt is a brief
A prompt needs the same discipline as a brief to a consultant or a junior.
Our position
A prompt is not a wish. It is a brief handed to a machine, and it needs the same discipline as a brief handed to a consultant or a junior: intent, constraints, exclusions, criteria. When an image comes back plausible and empty, do not assume the model failed before you have examined the brief — most of the time it read exactly what it was given, and what it was given was thin. And do not assume a better brief can beat a capability limit: a model that cannot count bays will not count them for a better sentence. Writing the programme before the form is the oldest skill in practice. It is now the first step in using the newest tool.
> The tool is not the problem. The programme is.
Why we hold it
William Peña made the argument in 1969: programming is problem seeking, design is problem solving, and most design failures are programme failures. Drawing used to absorb a vague programme — you sketched your way to clarity. A model cannot do that. It returns the most probable version of what you wrote. With AI, Peña's fifty-year-old argument now runs inside the machine: a thin programme returns a thin result.
The strongest objection
The junior-architect analogy anthropomorphises the model. A junior asks questions back, reads your silences, knows the office, and carries responsibility for the answer. A model does none of that, and it reacts to things no human brief contains: word order, token position, the statistical weight of a phrase. Teaching students to brief it "like a junior" may teach them to credit the receiver with understanding, intention and professional responsibility it does not have. The same briefing discipline does not make the two readers alike, and calling the prompt a brief may hide how strange the reader is. Some practitioners argue that "prompt engineering" is the honest name for exactly that strangeness.
What would make us revise it
Models that reliably ask the clarifying questions a good consultant asks — questioning a thin brief instead of filling it. If the machine starts doing the programme's work for you, the skill moves again, and this card moves with it. We check it against what the tools actually do, every edition.
Try it
Take a one-line wish: "a contemporary house with natural light and warm materials." Run it. Now rewrite it as a brief — the spatial type, the material logic, where the light enters and when, the sequence of arrival, the moment, what must not appear. Run that. Put the two outputs side by side and name what the second has that the first could not. Then brief the same room twice more — once in Kannada, once in English, same content — and compare which output is closer to the room you meant, and where each one drifted.
Take it to crit
Shown a weak output, did the student reach for the re-roll or for the brief — and can they point to the missing part of the brief that caused the failure, or say honestly that the brief was sound and the tool could not do it?
How it works
Large generative models do not work out what you left unsaid. They compute a probable answer from what you wrote and what they learned in training. So a vague prompt returns a probable house, not your house. A thin brief is the commonest cause of an empty image; it is not the only one — spatial relations, counts and geometry can fail under a sound brief, and the spatial-check card on this map is how you tell the two apart. Every other card in this strand is a part of the brief — the words (space, material, light, conventions), the structures (six layers, seven slots), and the discipline around them (criteria, exclusions, versions, the crit loop).
Lineage
Peña's Problem Seeking (1969; fifth edition with Parshall, 2012) established architectural programming as a discipline separate from design. The profession has written programmes for clients, teams, planning departments and contractors ever since. AI is a new audience for a document architecture already knows how to write.
What this idea builds on
What this idea opens up
- The seven-slot brief
- Decode the messy brief
- Curation is diagnosis
- The red pen is authorship
- The batch is a contact sheet
- Set the rubric before you generate
- Was the brief the failure?
- Name the job, not the deliverable
- Criteria, or only plausibility
- The six-layer reasoning brief
- What a reference image says
- The brief is a versioned record
- Three prompt families, three registers
- Negotiating the model's defaults
- The brief-to-crit loop
- Write so it can be found
- The space programme as brief
- Brief spatially, verify spatially
- Prompting is thinking about thinking
Sources
Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.