claudia.pa
Put Claudia to work ↗

CLAUDIA / YOUR FIELD GUIDE

Bring the ambition.
Give it a useful brief.

A personal agent is most useful when it knows what you want, what it may use and where it should stop. Personality helps a collaboration feel natural. A clear task makes it productive.

Six missions worth starting.

1. The week that got away

Give the agent a non-sensitive list of commitments and priorities. Ask for a draft weekly plan, clashes and decisions you need to make. If you later connect a calendar, decide separately whether it may only read, propose events or create them after approval.

2. The research rabbit hole

State the decision, deadline and acceptable sources. Ask for a comparison that shows evidence, dates, assumptions and unanswered questions. Agree how current each claim needs to be. Review original sources before relying on important conclusions.

3. The communication knot

Supply the intended reader, purpose and tone. Ask for a concise first draft and a stronger alternative. Specify what must stay private. Writing an email is different from sending one; keep recipients and final approval visible.

4. The process everyone dislikes

Describe the repeated business task from input to completion. Ask for the simplest improved workflow, exceptions and a small pilot. Check permissions, failure handling and the person who owns the result before connecting systems.

5. The trip with too many tabs

Give a destination, dates, budget and preferences without payment or identity details. Ask for a shortlist and trade-offs, then verify current availability and terms. Booking is a separate action with a named provider and your approval.

6. The subject you keep postponing

State what you can do already, what you want to learn and the time available. Ask for a short lesson, a practice task and a way to check understanding. A good explanation should leave you more capable, not more dependent.

The “let’s actually finish this” agreement.

  1. Define done. Name the result, audience, format and deadline.
  2. Give the constraints. Include budget, sources, exclusions and the choices already made.
  3. Choose the idea level. Essentials, a few alternatives or a broad brainstorm.
  4. Use checkpoints. Review the plan before expanding the task or connecting a tool.
  5. Make blockers explicit. Missing information, failed tools and uncertainty should be visible.
  6. Set a stopping rule. Agree time, cost and retry limits, and who can authorise the next step.
“I’ll keep looking for a way through. But if we need a key, I’m going to tell you—not pretend the door opened.”

That is the intended character of Claudia’s persistence. It is not a claim that an agent works indefinitely, guarantees every outcome or should ignore a request to stop.

A personal agent needs personal boundaries.

Memory with a purpose

Separate stable preferences from temporary working context. Set the owner, source, scope and retention period of stored information. Let people inspect and remove it. Do not assume one client’s knowledge belongs in another client’s task.

Tools with limited authority

Start with the minimum access a task needs. Distinguish a read, a draft and an external change. A model’s suggestion is not a permission grant; the application must enforce the actual boundary.

A place to work

Cloud, client systems and personal machines involve different operational responsibilities. Review where information travels, what is stored, who can access it and which external services are involved. “Local” alone is not a complete privacy policy.

A little quiet

Useful proactivity means relevant suggestions at a chosen time. Agree quiet hours, channels and idea frequency. Stop, pause and deletion controls should be understandable and effective.

Proof before promises

Test a real implementation on representative tasks and failure cases. Measure useful completion, evidence quality and escalation. Keep a human owner and a rollback plan before extending autonomy.

What this website does today.

The mission studio creates an illustrative plan locally in your browser. It does not invoke a model, use tools, access accounts, book travel, send messages or save persistent memories. The enquiry form is a separate way to discuss a real deployment.

Let’s scope your first useful mission ↗