Learn the building blocks
Short explanations of the building blocks of agent work: how agents plan searches, run multi-step workflows and act inside a browser.
Agent work rests on a handful of ideas that come up again and again: an agent that searches in steps rather than once, a workflow that adapts instead of following a fixed script, and a browser the agent can operate. Each page here explains one of them in plain terms, including where a person should stay involved.
The building blocks
- Agentic search: planning queries, reading and refining
- Agentic workflows and where decisions belong
- Browser agents and how to supervise them
- Headless browser APIs and when to use one
Where to go next
When a term is clear and you want to build something, the articles walk through complete workflows, such as a research agent that holds up. To see how approvals and evidence work around these blocks, read why an agent should ask before it acts.
4 resources
- Learn
Agentic search: an agent that plans, reads, refines and cites
How agentic search differs from one-shot search, what the plan, read and refine loop looks like, where it goes wrong, and what a team should control.
research - Learn
Agentic workflows versus fixed automation, and where people decide
What makes a workflow agentic, how it differs from scripted automation, and where human decisions belong so the agent can move without anyone losing control.
research - Learn
Browser agents: what they can do and where a person stays involved
What a browser agent is, the jobs it suits, the risks that come with logins and live-site actions, and how to supervise one with approvals and a visible trail.
browser - Learn
Headless browsers for agents: when rendering is worth the cost
When a headless browser API beats a plain HTTP fetch, how rendering and blocking work, and what running browsers costs you in time, money and upkeep.
browser