Delegate the outcome, not the task list: a goal brief for AI and your leads

Most directors delegate all day and rarely write the delegation down. A lead gets a two-line message, a coordinator gets a hallway request, and the work comes back shaped by whatever the person guessed you meant. AI tools have worked the same way: you asked for one thing at a time, and you stitched the pieces together yourself.

That began to change this month, and the change moves the director’s job toward the brief.

What shipped

On September 17, Anthropic published a redesign of Claude Projects. The company’s own summary is short: in a Claude Code project, “you describe what needs to get done and Claude manages the work.” According to the post, Claude scopes the request, hands out the parts, runs parallel threads, reviews what comes back and assembles the result. Each thread adds to and draws on a shared memory of the project, and you can follow progress from the main project chat or open any single thread to steer it.

Two limits matter. Each thread runs as a Claude Code cloud session, and the redesign launched in beta for select Pro and Max subscribers who use those sessions. So today this is mostly a tool for technical teams, although the post also says that with documents, Claude reads them and drafts.

A day earlier, TechCrunch reported that Anthropic merged Claude’s chat and Cowork into one interface, with chat, Cowork and Artifacts in a single window, and that Claude now routes requests itself instead of making people pick a tab. It can also create, edit and present slides, and export them as PDF or PowerPoint. The update is rolling out to Pro and Max plans first.

You may never use either product. The direction is what counts: tools are starting to accept an outcome instead of a task, and to split and assemble the work themselves.

Why the brief becomes the job

When a tool, or a person, can plan the work, the quality of the result depends on two things you control: what you asked for, and how often you look. Anthropic’s post makes the comparison itself, telling users to brief Claude the way they would brief a chief of staff.

That is a management skill, and it transfers. The brief that works for an AI coordinator is the same brief that works for a new team lead.

A goal brief in six lines

Write it before you hand anything off, to a person or a tool:

1. Outcome. What exists when this is done, in one sentence.
2. For whom. Who will read or use it, and what they will do with it.
3. Inputs. The documents, systems and past decisions it should draw on, and which ones it must not use.
4. Done means. Two or three checks you will run on the result.
5. Stop and ask. The situations where the work pauses for you: spending money, contacting a customer, anything that touches people decisions.
6. Review points. When you will look, and what you expect to see by then.

A worked example

Take a hypothetical support director who spends every Thursday morning building a monthly operations report from three sources: the ticket system, the staffing schedule and customer survey comments. A task-by-task request would be “summarize the tickets”, then “pull the staffing numbers”, then “draft the slides”.

A goal brief looks different:

• Outcome: a five-slide monthly report on support volume, response times, staffing and customer comments.
• For whom: the operations VP, who uses it to decide staffing for next month.
• Inputs: last month’s ticket export, the staffing schedule, the survey comments; not individual agents’ performance notes.
• Done means: every number traces back to a source file; the three largest changes from last month are explained.
• Stop and ask: if any number moves more than you would expect, before explaining it.
• Review points: an outline first, then the full draft.

The same brief works whether a coordinator, a lead or an AI tool does the assembling. You still own the numbers and the call on staffing.

A decision rule for review cadence

How often you look should depend on what a mistake costs, not on how capable the tool seems:

• Internal draft, easy to fix: review the finished result.
• Goes to your manager or another team: review the outline, then the final.
• Goes to a customer, touches money or people: review at every stop-and-ask point, and a person signs off before it leaves.

What to do this week

• Pick one recurring piece of work you currently delegate in pieces.
• Write its six-line goal brief and give it to the person who does it now. Ask what was missing.
• Set its review cadence with the rule above.
• If your team already has an AI tool that takes multi-step requests, try the same brief there, check the result against your “done means” list, and keep people and customer data out until your company’s policy says otherwise.