Build for AI Agentss
Building an agent means turning the concepts into running code using a model SDK or an agent framework, plus optional computer and browser control. This unit covers choosing between raw API calls and frameworks, structuring the loop, and letting an agent operate a screen or web page.
Stage: Building · Part of the AI Agents roadmap · Last reviewed 19 November 2026
What to learn in Build
- Implement a basic agent loop by hand using a model API with tool calling
- Compare a framework against hand-written code for a given task
- Pass structured outputs between steps using a typed schema
- Run a browser-using agent in a sandbox with limited access
- Handle a tool failure without crashing the whole run
Build interview questions
Should you use an agent framework or write the loop yourself?
Start by writing the loop yourself against the model API, because it is short and teaches what is happening. Adopt a framework when you need its tracing, state handling or integrations. Avoid layers that hide prompts and make debugging harder.
What are the risks of computer and browser use agents?
They act through screens and web pages, so they can click the wrong thing, follow instructions hidden in page content or expose logged-in sessions. Run them in isolated sandboxes, use throwaway accounts, restrict domains and require human approval for irreversible actions.
Where Build fits in the AI Agents roadmap
Build belongs to the Building stage (advanced level), which takes about 18 hours. Frameworks and runtime.
Other units in this stage
Back to the full AI Agents roadmap