From request to task · chapter 4 of 17

Context and sources: what AI sees and does not see

A good answer depends on the right context. The model only sees what the conversation, project and tools actually expose, and time-sensitive information must be checked against a source.

4 of 17 chapters

4.1 · plain explanation

Which information does the model need, and what can it not know unless we expose or verify it?

A good answer depends on the right context. The model only sees what the conversation, project and tools actually expose, and time-sensitive information must be checked against a source.

Important: More context is not always better. A small, precise package makes gaps visible and reduces the risk that stale information drives the work.

4.2 · building blocks

Three concepts to keep separate

01

Context

The objective, background, relevant files and prior decisions needed for this particular task.

02

Source

The document, system or measurement that can support a claim.

03

Freshness

History can be a useful lead, but current code, runtime state and official documentation have higher authority.

4.3 · method

A practical workflow

  1. 1State the objective in one sentence.
  2. 2Select only the files and facts that affect the task.
  3. 3Mark what is observed, assumed and still unknown.
  4. 4Check time-sensitive claims against a current primary source.

4.4 · concrete example

A website defect

Scenario

A website defect

Which information does the model need, and what can it not know unless we expose or verify it?

Starting point

A screenshot shows the symptom but not necessarily the cause.

Controlled result

Combine the screenshot with the live DOM, current CSS, correct viewport and a test after the correction.

4.5 · try it yourself

Exercise

Create a context package of no more than ten lines for a real task: objective, relevant files, known boundaries, fresh observations and open questions.

4.6 · checklist

Before you continue

  • Does the model have the concrete objective?
  • Are sources relevant and current?
  • Are assumptions labelled as assumptions?
  • Are secrets and unnecessary history excluded?
Remember:More context is not always better. A small, precise package makes gaps visible and reduces the risk that stale information drives the work.

4.7 · sources

Sources

Reviewed 23 September 2026. Official sources support product-specific claims; the practical contract, memory and verification patterns are Finnandre project methods.