Decide what to build

Research a question

Turn a question into an evidence-graded answer, with the strongest objections written down.

desk-research

Ask your agent a question in plain words, and each section below writes one file named after your topic.

Scope the question

To split the question into the sub-questions a thorough answer must cover, ask:

Decompose this question into the sub-questions a thorough answer has to cover: should we adopt server-side rendering for our marketing site?

The agent writes <topic-slug>-outline.md, where the slug is a short name for the topic. Before it gathers evidence, it asks you to confirm the scope and depth. For a contested question, ask for identify-perspectives to name the camps and source-map to rank the sources.

Run the research

To gather evidence at a depth you choose, ask:

Research this at applied depth: what do teams actually report after moving a marketing site to server-side rendering? Cite primary sources.

The agent writes <topic-slug>-survey.md with a confidence tag on every finding and a list of what the evidence could not answer.

Challenge the result

To find where the survey is weakest, ask:

Argue against this survey. Find the strongest counter-evidence for each finding and tell me which ones should be downgraded.

The agent writes <topic-slug>-counterpoints.md, giving each finding a verdict. For a decision between options, compare-hypotheses scores each option against the same evidence.

Run it as a project

For a question that needs weeks, start a research project instead:

Start a research project on how teams operate design systems at scale. This runs over several weeks.

The agent creates a project folder with an overview and a working hypothesis. Ask for desk-research-project-status at any point to see the phase and what to do next.

Write the final brief

When new sources stop changing the findings, ask:

Synthesize the project into its verdict and write the governance brief.

The agent writes the verdict file and <topic-slug>-brief.md in the project folder, with gaps named rather than smoothed over. You review the synthesis before you act on it.

Next, analyse the market with what you found.