How-To Guides

Write an effective Objective

The quality of an objective determines whether the agent executes it cleanly or comes back with a miss. These steps cover every field and why it matters.

1. 1

   Write the title as an outcome

   Describe what changes about the world when the objective is done — not what you're doing to make it happen. The title is the agent's primary directive. It should be specific enough that a senior engineer (or a coding agent) could start without asking a clarifying question.

   Not this

   “Implement OAuth”

   “Auth work”

   “Fix the payment flow”

   “API improvements”

   This

   “Add GitHub OAuth so new users can sign in without a password”

   “Replace manual card entry with Stripe Checkout on the upgrade page”

2. 2

   Write acceptance criteria

   Add a bullet list of conditions that must all be true before the objective can reach Done. Each criterion should be verifiable — a reviewer should be able to answer yes or no without judgment. Agents use this list as their target; reviewers use it as their checklist. Without it, agents guess at completeness.

   Good: “Users can sign in with a GitHub account on any page with a login prompt.” Not: “Auth feels polished.”

3. 3

   Refine it with your agent

   Before scheduling, open a conversation with your coding agent and share the draft objective. Ask it to think through edge cases, user scenarios the acceptance criteria don't cover, implementation caveats it would want resolved before starting, and anything ambiguous in the title or scope. The agent will surface gaps you didn't anticipate. Update the objective — title, criteria, description — with what you learn before moving on.

   This is refinement, not execution. The agent isn't working on the objective yet — it's helping you make the spec precise. Schedule it only after this conversation.

4. 4

   Fill in the description

   Add what the agent needs to understand scope and constraints that the title doesn't cover: related files or services, links to external docs, known edge cases, things that must not change. The refinement conversation in step 3 will tell you exactly what belongs here. Agents don't accumulate context from standups or Slack — the description is their complete briefing.

5. 5

   Scope it to a single agent run

   An objective should be completable in one agent session. If you have more than four or five acceptance criteria, or if the work spans multiple independent areas, you likely have two objectives. Split them — smaller, tighter objectives get cleaner agent execution and faster acceptance cycles.

6. 6

   Set priority

   1 is urgent, 5 is someday. In multi-agent workspaces, priority determines claim order — no sprint planning required. Default to 3 if you're unsure.

7. 7

   Set strategic focus

   Choose BUILD, GROW, TRUST, or SCALE. This field is required before an objective can be claimed. If you skip it, the agent will propose a classification at claim time — but setting it yourself keeps portfolio allocation data accurate from the start.

   Optionally assign to an initiative if the objective belongs to a named theme or release milestone. Standalone objectives are fine — initiatives are optional groupings, not required structure.