Guides to AI agent web access: tools, errors, budgets
Six guides, one per decision you make when an agent reads the web. Each gives a default you can copy, a section on where it breaks and links to the field notes that go deeper.
Fetch tool design
How to design a web fetch tool for AI agents
The contract between the model and its web tool: schema, description, output shape, metadata and the knobs to keep out of the model's hands.
0 related posts
Errors and retries
Agent tool error handling: a field guide for web fetch
Retry transient failures in code, hand the rest to the model as one actionable sentence, and stop before the loop burns your budget.
0 related posts
Prompt injection and trust
Indirect prompt injection from web pages: an agent's threat model
Fetched pages are data, not instructions. A threat model and the controls that hold up: allowlists, SSRF checks, read/act separation and human approval.
0 related posts
Cost and token budgets
Token budgets for agent web pages: caps, limits and costs
Cap each page, budget fetches per task and keep long pages out of the main context. Starting values and how to measure your own.
0 related posts
Fetch, search or browser
Browser agent vs fetch: which tool each agent step needs
Fetch, search, browser agent or computer use: a per-step decision, with the signals that tell you a static fetch was not enough.
0 related posts