The Knowledge Graph: Slim Context for Sharp Agents

The biggest productivity jump we have seen with coding agents did not come from a larger prompt. It came from giving agents a knowledge graph they can navigate — and then keeping their context small on purpose.

Slim context beats big context

Many AI coding failures look like reasoning failures but are really context failures. The model gives a vague answer because it has vague context. It edits the wrong file because every directory looks equally plausible. It invents architecture because the real architecture was never loaded. Dumping the whole repo into the window does not fix that — it pays in cost, latency, and lost attention.

Context is a critical but finite resource for AI agents.

A graph, not a prompt blob

The Knowledge Graph is a folder of small notes that link to each other, the way an Obsidian vault does. Each note holds one fact, decision, convention, or relationship, and points at exact code paths or assets. SE Studio indexes the folder and gives each agent its root, so the graph is a navigable resource the agent can search and follow — not a hidden blob stuffed into every prompt.

The agent does not need the entire repository in context. It needs the right neighborhood: the relevant route, the component conventions, the service boundary, the migration caveat, and the decision that explains why the code is shaped this way. File paths, note titles, and links all become predictive clues. The agent forms a fast hypothesis about where to look, then verifies in the source.

Example: a brand folder

Say an agent is building a new screen. Without a graph it produces something technically correct that feels disconnected from the product. With a brand neighborhood in the graph, it loads the visual system, approved colors, voice, and guidelines first — and builds something that fits.

knowledge/
  index.md
  brand/
    visual-system.md
    palette.json
    voice.md
    guidelines.pdf
  product/
    onboarding-flow.md
    workspace-navigation.md
  engineering/
    auth-boundaries.md
    event-contracts.md

The brand folder is just one neighborhood. The same shape works for engineering: service ownership, API contracts, schema migration rules, test commands, and security boundaries. The agent moves from task to note to code with a much smaller search radius.

Why it matters

• Smaller context — the agent loads a neighborhood, not the whole repo.
• Sharper output — answers are anchored in known facts instead of guesses.
• Lower cost — fewer tokens spent on irrelevant context every task.
• A feedback loop — useful learnings become new notes that future agents inherit.

The threshold for usefulness is low. Ten good notes save repeated orientation. Fifty can encode a product system. A few hundred, kept clean and linked, make a very large codebase feel navigable to both humans and agents.