TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · The Brief · AGE 18—20 · POSITIONAL · METHOD
When the receiver plans
A process brief adds checkpoints, stop-conditions and permissions to the six layers.
When to use
When you hand work to a system that will act over several steps — search, read, write files, call tools, come back — rather than answer once.
The method
Start with the six layers: intent, context, constraints, criteria, output form, iteration loop. Then add three. Checkpoints: where the process must pause and show you — "after the list of candidate sites, stop and wait." Stop-conditions: what ends the work — a count, a time, a finding, a failure. Written in the brief, a stop-condition tells the process when it ought to stop; it does not make it unable to continue. What enforces the stop is the system's own controls — an execution budget, a maximum number of steps, a timeout, a tool allowlist, a human approval gate — so set those where the tool lets you, and write the stop-condition as well. Permissions: what it may read, what it may change, and what it must never touch — the client folder, the sent mail, the live drawing. A permission in the brief asks; a permission set in the system is what holds. Write the three as plainly as constraints. Run a small version first, read the log of what it did, and only then give it the real job.
Watch for this
A brief that states the goal and trusts the process to know when to stop — or trusts a sentence to stop it. A multi-step system that is wrong at step two carries the error through every step after it. One wrong answer from a chatbot is one wrong answer. One wrong answer inside a process is a chain.
The Lab's note
Six layers plus three is the Lab's extension of its own reasoning brief for systems that act. Vendors publish their own guidance on agents, and it agrees on the spirit — simple loops, human checkpoints, limits set in the harness rather than asked for in the prompt. The nine-part shape is ours, and it will be revised as the tools change.
Try it
Brief an agent-style tool to collect ten references on a topic and write a one-page note. Version one: goal only. Version two: goal plus a checkpoint after the references, a stop at ten, and a rule that it may read but not write outside one folder — and, where the tool allows, the same limits set in its settings, not only in the words. Compare the two logs, not just the two notes.
Prove it
Show one process brief with all nine parts labelled and the log of a run, pointing to where a checkpoint or a stop-condition did its job — and to the system control (budget, step limit, approval gate, permission) that backed it.
How it works
An agent is a model inside a loop: plan, act, observe, repeat. Each pass through the loop reads your brief again, so instructions that were clear for a single answer become ambiguous over twenty actions. What an agent can reach — files, mail, money — is set by its permissions, which someone configured, and how long it can run is set by its budget; those are the controls. The brief's stop-conditions and permissions ask; the configured ones enforce. The Studio Practice strand carries the responsibility side of that. Tool capabilities and safeguards change by the month, which is why this card is reviewed each edition.
What this idea builds on
- A starting idea.
What this idea opens up
Sources
Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.