Lesson 16 · Claude Mastery Pro ~12 min read 4 delegation patterns

Working with Fable 5: delegation is the new prompting.

For two years the skill was writing a clever prompt and getting a clever answer back. Fable 5 changes the game: it can work on its own for hours, across many steps. The people who get the most from it stop writing prompts and start writing briefs — the way you'd hand a project to a capable new hire.

The mental model

Stop asking. Start delegating.

A prompt is a question: "Write me an email about X." A brief is an assignment: the goal, the constraints, what done looks like, and what to do when something's unclear. Earlier models needed prompts because they did one thing then stopped. Fable 5 can carry a whole task — so the quality of your brief now matters more than the cleverness of your wording.

The shift in one line

With older models you steered every turn. With Fable 5 you set the destination well, then check the work — like managing a sharp employee instead of operating a tool. Vague brief, wandering result. Tight brief, finished work.

Pattern 01 Lead with the goal, not the steps

1

Say what "done" looks like before you say how

Weaker models needed you to spell out each step. Fable 5 plans the steps itself — if you tell it the actual outcome you want. Define success first; let it figure out the route.

Instead of a step list, write a brief

Delegation briefGoal: a one-page competitive summary I can hand to my boss Friday. Cover our top 3 competitors. For each: what they do better than us, where we win, and one thing we should copy. Sources: the three URLs below, plus anything else you find that's from 2026. Done looks like: one page, plain language, no jargon, a short "what I'd do about it" at the end. If a competitor's pricing isn't public, say so rather than guessing.

Why it works

  • The model owns the planning, which is what it's now good at
  • "Done looks like" prevents the 80%-finished result that needs three more rounds
  • The "say so rather than guessing" line is your hallucination guardrail
Result: One brief instead of a ten-message back-and-forth.

Pattern 02 Set constraints up front

2

The guardrails that keep a long task on the rails

When a model works for a while on its own, small wrong assumptions compound. Constraints are how you prevent drift before it starts — the same things you'd tell a contractor before they begin, not after.

The constraints worth stating every time

Constraint stackAudience: non-technical executives. Length: under 500 words. Tone: direct, no hype. Must include: real numbers where you have them. Must NOT include: invented statistics, fake quotes, or filler. If you're unsure about a fact, flag it instead of stating it confidently. Stop and ask me before making any assumption that would change the conclusion.

Why it works

  • "Stop and ask before assuming" turns a runaway task into a checkpoint
  • The "must not include" line is the single most valuable instruction for trustworthy output
  • Constraints stated once at the start beat corrections made five steps in
If you don't set a length and tone, Fable 5 defaults to thorough — which on a long task can mean far more than you wanted. Bound it on purpose.

Pattern 03 Build in checkpoints

3

Catch a wrong turn at minute 5, not hour 2

The failure mode with autonomous work is discovering at the end that it went the wrong direction early. The fix is to ask for a plan first, approve it, then let it run — exactly how you'd manage a big assignment from a person.

The two-stage handoff

Checkpoint patternBefore you do the full task, give me: 1. Your understanding of what I'm asking for, in 3 bullets. 2. Your plan of attack, in 5 steps or fewer. 3. Any assumptions you're making that I should confirm. Wait for my go-ahead before doing the work.

Why it works

  • You correct a misunderstanding in 30 seconds instead of redoing an hour of output
  • The model's stated assumptions often surface a gap in your brief
  • For genuinely long tasks, add: "check in with me at the halfway point before continuing"
Result: Long tasks that finish right the first time.

Pattern 04 Make it check its own work

4

The single instruction that raises quality the most

One of Fable 5's genuinely new strengths is reflecting on and validating its own output — Anthropic specifically calls this out at high effort. You unlock it by asking for it explicitly.

The self-review close

Self-validationWhen you're done, review your own work as if you were a skeptical expert seeing it for the first time. List anything that's weak, unsupported, or that you'd want to double-check. Then fix the issues you found and give me the final version plus a short note on what you changed.

Why it works

  • The model catches its own thin spots before you have to
  • The "note on what you changed" tells you where to apply your own judgment
  • It mirrors how good professionals work: draft, critique, revise
Self-review reduces errors but doesn't eliminate them. You're still the editor of record — especially on anything that leaves your hands. Trust, then verify.

Final challenge: write one real brief

Take a task you'd normally hand Claude as a quick prompt — and instead write it as a full brief: goal, constraints, a checkpoint, and a self-review close. Run it on Fable 5. Notice how much less back-and-forth it takes to reach something you'd actually use. That gap is the skill this whole release rewards.

What you can do now

  • Tell the difference between a prompt (a question) and a brief (an assignment)
  • Lead with the goal and "done looks like," and let Fable 5 plan the steps
  • Set constraints up front to keep a long task from drifting
  • Use a plan-first checkpoint to catch wrong turns early
  • Add a self-review close to raise output quality
  • Stay the editor of record on anything that leaves your hands
Pro
Up next in Claude Mastery

Lesson 17 · Fable 5 vs Opus vs Sonnet: which Claude for which job now

The most powerful model isn't always the right one. A clear decision framework for picking the right Claude — and spending usage credits wisely after June 23. See pricing →