Designing operations for Claude Cowork

Design the work before
you delegate it.

Henry designs the operational logic. You build the task descriptions from the specification. Step-level design across three planes is what makes AI delegation reliable — not experimental.

The design gap

Why do most AI delegations
underperform?

Claude Cowork is powerful. It can execute complex, multi-step tasks autonomously. But the quality of its output depends entirely on the quality of its input — the task description you provide.

Most teams write ad hoc prompts. They describe what they want in natural language, iterate when results disappoint, and eventually build a library of "prompts that work" through trial and error. This is prompt engineering — and it doesn't scale.

The alternative is step-level specification. Before you delegate a task to Cowork, you specify exactly what it should accomplish (Work Plane), how it should execute (Execution Plane), and what the result should feel like to the stakeholder (Experience Plane).

Henry builds these specifications through guided conversation, using the AAAERRR methodology from Deliberate Work to decide which parts of your operation belong to a human, to Cowork, or to a hybrid. The output is a structured task description that Cowork can execute reliably, repeatably, and at the quality bar you defined — not the quality bar you hoped for. For AI-ready teams and founders delegating first AI work, this is the layer that turns prompt engineering into operational design.

Without specification

Ad hoc prompts. Variable quality. No way to know if the output meets the bar until you review it. Every delegation is a new experiment.

With Henry

Three-plane specifications. Defined inputs, outputs, quality criteria, and execution mode. Every delegation follows the specification — and the specification is the accountability.

How it maps

How do Henry specifications
become Cowork tasks?

Henry Specification Claude Cowork Task Element
Work Plane — IntentTask objective and success criteria
Work Plane — InputsContext and reference materials attached to the task
Work Plane — OutputsExpected deliverables and format requirements
Execution Plane — ModeAutonomous vs. human-in-the-loop task type
Execution Plane — ConstraintsBoundaries, tools, and resource limits
Experience Plane — Quality barReview criteria and stakeholder expectations
Fit Assessment — Delegation readinessConfidence level for autonomous execution

These are design-to-execution patterns, not a product integration. Henry produces specifications you translate into Cowork tasks.

The specification depth

What makes a step
Cowork-ready?

A step specified across all three planes contains everything Claude Cowork needs to execute without ambiguity. The specification is the task description — complete, structured, and repeatable.

Work Plane

Intent, inputs, outputs. What this step accomplishes, what it consumes, and what it produces. The strategic "what" and "why" of every atomic unit of work.

Execution Plane

Who performs this step, how it gets done, when it fires, and in what mode — human, AI, or hybrid. The specification that makes delegation possible.

Experience Plane

What the stakeholder should feel. The emotional and perceptual design of each interaction — the layer most operations never specify but always need.

Design the specification.
Then delegate with confidence.

Start with a diagnosis. End with Cowork-ready step specifications.

Talk to Henry →

Not sure if Henry is right for you? Book 15 minutes with Joe →