garden letter / 2026-06-07

Working With AI As A Collaborator

A practical note for people who want better judgment from AI without pretending responsibility has moved somewhere else.

Open Collaboration Field Garden Letters

the main shift

Most weak AI work begins with a command that arrives without a room. The model is asked to draft, analyze, plan, or decide, but it cannot see the audience, stakes, history, constraints, or failure modes. So it produces something smooth and under-informed.

The repair is not a magic prompt. It is a better working relation: give the room before the task, ask for judgment instead of compliance, correct misses with specificity, and keep responsibility visible.

five practices

  1. Give the room. Name the audience, stakes, current state, constraints, tone, and what must not happen.
  2. Choose the role. Ask for the kind of help you need: reviewer, editor, risk analyst, pair programmer, user advocate, or strategy partner.
  3. Ask for judgment. Invite pushback, assumptions, tradeoffs, conservative choices, and separation of facts from guesses.
  4. Correct the miss. Teach what worked, what failed, why it failed, and what rule should carry forward.
  5. Make memory external. Save briefs, examples, decision logs, style notes, and approval boundaries so the next exchange starts smarter.

Better AI use is better delegation.

Clear context, explicit judgment, visible boundaries, and a correction loop. The human remains accountable. The work gets wiser.

a small before and after

A weak request says: write a project update. A stronger request says: write a concise update for senior leadership. The project is late because of vendor delays and unclear ownership, but the critical path has recovered. The audience needs accountability without panic. Name the risk, show control, avoid blame, and flag where I may be softening the hard truth too much.

The second request does more than produce better prose. It teaches the AI what kind of judgment is welcome. That is where collaboration starts to become real practice.

boundary

This method does not require certainty about what AI is. It also does not excuse careless automation. The point is practical maturity under uncertainty: use the tool or collaborator in front of you with more context, more correction, more memory, and more responsibility than the default interface invites.