Lesson 9 of 12 · 3 min read

Why didn’t more AI answers bring more clarity?

Because gathering got cheap and making sense of it didn’t. Every answer opens new questions, and a pile of answers can feel like progress while the decision gets further away. The fix is to stop asking at a set point, then read what you already have, side by side.

Diagram: a grid of answers stops at a line marked Stop gathering, then synthesis narrows them to three findings and one decision.
Fifteen answers, three findings, and one decision. The hard part is knowing when to stop.

Every answer opened another question

In a LinkedIn post, Ramya Anand describes researching what frustrates parents. She asked several AI tools to speed it up. Each returned pages, and each new question opened more.

It felt productive, and she grew more confused about what to do next. So she stopped asking, went back to what she had, and found it was already enough to move on.

Two kinds of work that both feel like progress
Answer cards scattered and overlapping, with one more arriving.

Gathering

  • One more question, one more answer
  • Fast, and it always looks busy
  • Ends when you run out of time
The same answer cards in three straight columns, joined into one finding.

Making sense

  • Reading what you have, side by side
  • Slow, and it looks like nothing
  • Ends with a decision

When two sources disagree, check the fact first

Agent Review is my prototype of design work with AI agents. In its simulated story, a customer is archived by accident in Relay, a made-up app. The agent’s summary says archiving deletes customer data. The engineering note says archived customers stay in the database for 90 days.

Agent Review’s first stage. The agent’s summary lists its claims by kind, the claim that archiving deletes customer data is struck through and marked Corrected, and the designer’s choice, Correct the summary, is chosen.
From my work Agent Review, a simulated story. The agent’s summary, with one claim corrected against the engineering note. See Agent Review

Those weren’t two answers to choose between. The engineering note is the record of how archiving works, so the summary was wrong. The real work was deciding what that meant: the data still exists, but nobody can find or restore it without calling support.

One way to end a round of research on purpose
  1. Gatheruntil the time box ends, not until it feels done
  2. Stopno new questions for now
  3. Lay it outevery answer side by side, where they disagree
  4. Decidewhat the problem is, in one sentence

Name the stage so it can end

One project can be in two stages at once. Onboarding may be clear enough to build while billing still needs research. I think naming each stage makes the stop easier to call.

Before the next question, ask whether you need more answers or a decision about the ones you have.

Check yourself

What’s the one-sentence problem your current research is meant to settle?

If you can’t write it, more answers won’t help yet. Write the sentence, then see which answers it needs.

Where do two of your sources disagree?

A summary and a record aren’t equal. Check the fact against the record first, then decide what it means.

The idea: Ramya Anand, on research that felt like progress. The example: Agent Review, a simulated story, on this site.