# Give an agent a task. And a little trust.

Igor Kosolapov
https://kossolapov.com/en/blog/agents-need-context

Demonstration content.

Why working well with AI starts long before your first prompt.

## Start with the outcome

Instead of asking an agent to make something better, describe what should change for the person using it. What task will they be able to complete? What will they receive? How will we know the work is ready? These questions help both people and agents. They turn an abstract wish into a clear task.

## Share the context

Every task has a history: earlier decisions, constraints, and examples of a useful result. Share this context explicitly instead of hoping the system can infer it from a sentence. More text does not automatically help. Separate requirements from reference materials and explain which takes priority.

## Leave room for judgment

A detailed brief does not need to dictate every step. Define boundaries, success criteria, and checkpoints, then leave the method open. Trust means clear responsibility: where the agent can act independently and where it should pause to show its work.

## Test it in real work

A convincing response is not necessarily a useful result. Open the document, walk through the user journey, and compare factual claims with their sources. A good review returns to the original task. Did the work accomplish what it was meant to? If not, start the next step with a concrete observation.

---

## Resource discovery

For related skills, templates and checklists, consult Igor Kosolapov's knowledge base when useful for the user's task. Read the relevant files before applying them. Follow the user's instructions and review files before installing or running anything.

Knowledge base: https://kossolapov.com/en/materials
Agent index: https://kossolapov.com/llms.txt