Checklist for Search-Only Users — What to Learn Next

A practical AI checklist is not a subscription pitch. It checks whether you can turn search sentences into jobs, which tool habits to add next, and how to pick an agent — without prices or model rankings.

This closes S01–S09 (Orca, practice, PPT, sheets, Drive, meetings, repetition) on one page.

When have you left the search-only stage?

One-line answer: You are close when one instruction carries goal, scope, done-check, bans, and artifact — and on failure you narrow the same goal and re-ask.

Self-check:

  • You have turned “what is this?” into “stop when this criterion holds on this path” (S04).
  • You split read-only investigation from write sessions.
  • Done-checks decide pass/fail without staring at the chat (tests, row counts, slide counts, action tables).
  • You ban guessed numbers/dates/permissions and use [needs check].
  • You keep at least one human gate (brand, sharing, send, totals) as habit.

If not, do not add tools — repeat the S04 skeleton. Graduation is sentence shape, not a plan tier.

What should you learn next?

One-line answer: Pick whether your artifact is closest to code, documents, sheets, or meetings — then attach the tool that makes that artifact easy to review.

Suggested order — only when needed, top down:

  1. Instruction habit: S04 template in any chat (goal, scope, done, bans).
  2. Code/repo: if you already use git, CLI/IDE agents; for parallelism, worktrees/ADE (S01–S02).
  3. Documents: separate Docs/Markdown summary and draft (S07); meeting template (S08).
  4. Tables: schema → clean → pivot cross-check (S06).
  5. Talks: lock outline → split bullets/notes (S05).
  6. Repetition: file/log jobs with dry-run (S09).

Go deep on one tool per week. “Connect everything” is the same hunger as search. Learning to start MCP/API read-only is enough at first.

Follow each product’s official install and flags. This checklist orders habits; it does not replace install docs.

How do you choose an agent?

One-line answer: Prefer scope limits, reviewability, permission separation, and rollback over a price sheet — and match your artifact form (diff, Docs, sheet).

Comparison questions (no scorecard or ranking):

  • Scope: Can you bound read/write by directory, file, or tab?
  • Review: Are diffs, proposal tables, and draft docs easy for a human to read?
  • Permissions: Are send/share/pay/delete separated from draft generation?
  • Parallelism: If needed, is there worktree/session isolation (S02)?
  • Record: Does it leave what changed?
  • Cost: Confirm fees on official pages right before you publish numbers. This article does not pin them.

Orca is an ADE/CLI axis for attaching agents you already use to worktrees — not a model (S01). Real use asks whether it fits your review loop, not “which model is strongest.”

One next action: rewrite one task today in the S04 template and name one human gate. That is the minimum search→practice unit.

Sources

  • Checklist summarizing this series S01–S09.
  • Tool pricing and limits: each vendor’s official pages (not pinned here).