DQ Measure your DQ
The framework · in full

Delegation has exactly three moves.

Everything about working well with AI agents reduces to how you brief, how you hand off, and how you review. This page is the full framework: what each move actually is, what good looks like, where it fails, the management theory each one stands on — and how the score works.

№ 01 — The premise

Everyone has the same models. The same subscriptions, the same launch-day access. What separates 10× output from a slightly better search engine is not the AI — it's the quality of the delegation. And delegation is not one skill. It's three distinct moves, each with its own failure modes.

№ 02 — The loop

It's a cycle, not a checklist.

Brief feeds Handoff. Handoff produces output for Review. And Review — this is the part almost everyone skips — feeds the next Brief. One pass through the loop is using a tool. Repeated passes, where each Review improves the next Brief, is running a workforce that gets cheaper to manage every week.

Move B

Brief

The work you do before the agent starts: scope, curated context, constraints, and a definition of done — written the way you'd brief a sharp new hire on their first week.

What good looks like

  • A definition of done. A sentence the output can be checked against. If you can't write it, you don't know what you want yet — ask the agent for three interpretations first.
  • One example beats three paragraphs. A sample of good output communicates format, depth, and taste better than any description of it.
  • Curated context, not dumped context. The two files that matter — with a note on what to ignore. Nine files where two matter actively misleads.
  • Ambiguity resolved in the brief, not the chat. When a mid-task question reveals a misunderstanding, fix the brief itself and restart — the chat patch evaporates, the brief fix persists.

Where it fails

  • The mega-prompt. Every detail you can think of, in one breathless paragraph — detail without structure. Length is not clarity.
  • "Use your best judgment." Passing your ambiguity to the agent instead of resolving it. You'll dislike the result and blame the tool.
  • Delegating the decision instead of the groundwork. For judgment calls, brief for options-with-trade-offs you can interrogate — not for the answer.
THEORY ROOT Covey's stewardship delegation (1989): desired results, guidelines, resources, accountability, consequences — the five elements of a brief, named 35 years early. Prompting is gofer delegation →
THE METRIC IT DRIVES First-Pass Yield — the percentage of delegated tasks accepted without rework. Weak briefs show up here first.
Move H

Handoff

Actually letting go: work running without you — in parallel, behind gates you chose deliberately — while you do something only you can do.

What good looks like

  • The second task. The moment one agent starts working, you start another stream. Two parallel streams is the line between using a tool and running a workforce.
  • Calibrated gates, not vibes. Autonomy is set per task type, from evidence: client-facing work gets tight gates, research runs free. Full autopilot everywhere is not high trust — it's no calibration.
  • Batched review. Outputs processed at set times rather than on arrival, so finished work doesn't interrupt the work only you can do.
  • Fortnightly re-tests. Every model release moves the line of what can run unattended. The task you kept manual in March may run free in June — but only if you re-test.

Where it fails

  • Watching it generate. Supervision theatre. If you're watching, you haven't delegated — you've just changed who types.
  • The frozen trust setting. Calibrating once — usually on a bad early experience — and never re-testing. Most professionals' trust is set to the first model that burned them.
  • Becoming the bottleneck. Outputs queueing on you for hours. The agents are parallel; you're serial; the system runs at your speed.
THEORY ROOT Graicunas (1933) proved span of control caps at ~6 because relationships grow geometrically. Agents have no relationships — the cap moves to how many briefs you can write and verify. Span of specification →
THE METRIC IT DRIVES Span of Supervision — your median concurrent delegated workstreams — and with it, the True Leverage Ratio.
Move R

Review

Judging output against the brief — not your mood — then sending it back with feedback, and folding what you learned into the next brief. The only move that compounds.

What good looks like

  • Verification by risk, not by page. With ten minutes and twenty pages, check the three claims that would be most expensive if wrong. Reading linearly is review theatre.
  • Send-backs with management notes. Three specific gaps named against the brief, plus one example fix — not "try again," not a silent rewrite.
  • The sixty seconds after. When the task ships, improve the saved brief while the failure modes are fresh. Ten tasks in, you have a playbook — delegation capital.
  • Gate-level fixes. When something wrong ships, trace which gate failed — brief, checkpoint, or final review — and fix that gate. Not the model, not your faith in AI.

Where it fails

  • Skim-and-ship. Agent output is fluent and confident whether it's right or wrong. Polish is not evidence. This is the single most expensive habit in the agent workforce.
  • The full rewrite. Fixing everything yourself feels faster — and guarantees the next output has the same flaws, because nothing flowed back into the brief.
  • Review without memory. Honest checking that never updates anything. Every task starts from scratch; nothing compounds.
THEORY ROOT Principal-agent theory: AI agents deleted the incentive problem and doubled the verification problem. All delegation cost collapsed into this move. The verification inversion →
THE METRIC IT DRIVES Delegation Capital Rate — the share of tasks started from an existing playbook. The compounding curve lives here.
№ 03 — The four tiers

Where the moves put you.

  1. 0 – 3.5

    The Chat-Tab Operator

    Asks AI questions, copies answers. No briefs, no handoffs, no review — a better search engine, experienced as transformation. The way up: the three-line brief — what, what good looks like, what to ignore — before every task. That single habit moves a tier.

  2. 3.6 – 6.0

    The Power Prompter

    Real technique, elaborate prompts — but one task at a time, watched while it runs, judged by vibes. The human is still the bottleneck. The way up: the second task. Start another stream before the first finishes; the discomfort is the skill forming.

  3. 6.1 – 8.0

    The Delegator

    Proper briefs, genuine parallel handoffs, review against a definition of done. Work finishes during meetings. What's missing is the flywheel: every task still starts close to scratch. The way up: sixty seconds after every task improving the saved brief. The only move that compounds.

  4. 8.1 – 10

    The Agent Manager

    A portfolio of delegated streams, evidence-calibrated trust, playbooks that improve with every run. Output has decoupled from hours — the 1933 org-chart math no longer applies. The way up from here is organisational: raise the ten people around you and become the multiplier.

№ 04 — How the score works (and its honest limits)

A measurement you can argue with.

Every item is a workplace caselet — a board deadline, a confidently wrong report, an ambiguous ask from a founder — with four responses scored 0–3 against the framework. Your DQ is the total, normalised to a 0–10 scale, with sub-scores for each move.

Three item types resist aspirational answering: recall items ("in the past two weeks, how many…") that ask what you did, not what you'd do; cost trade-offs where the textbook-virtuous answer is deliberately not the best one — asking for an extension scores below checking the three riskiest claims; and forced trade-offs where you can only keep one thing in the brief.

The limits, stated plainly: this is structured self-report, not observation. The Quick Score (12 items) is an estimate — useful mostly for the gap between your guess and your measurement. DQ Professional (27 items, nine per move) is the measurement worth standing behind. The highest-validity version — scoring your actual briefs as artifacts — is on the roadmap.

A non-obvious scoring choice worth knowing: full autopilot scores below gated autonomy. Shipping everything an agent produces untouched isn't high trust — it's the absence of calibration, and the framework treats it as such.

№ 05 — The lineage

A century of management theory,
pointed at agents.

DQ is not a new philosophy. It's the oldest discipline in business — delegation — re-derived for a workforce with no self-interest, no relationships, and capabilities that change fortnightly. Where the classical theory transfers, we kept it. Where it breaks, the breaks are the framework.

Classical theoryWhat agents changeWhere it lives in DQ
Covey — stewardship vs gofer delegationTransfers intact: prompting is gofer, briefing is stewardshipBrief
Jensen & Meckling — principal-agent costsIncentive problem deleted; verification problem doubledReview
Graicunas — span of controlRelationship math obsolete; specification is the new capHandoff
Hersey-Blanchard — situational leadershipReadiness now changes per model release; calibration becomes a loopHandoff
McGregor — Theory X / Theory YBoth miscalibrated for AI; trust becomes per-task evidence ("Theory E")Handoff · Review
Drucker — knowledge-worker productivityThe brief is a file: delegation compounds for the first timeReview → Brief

That's the framework.
Now measure where you stand.