AI Agents

Lesson content

AI Agents Learn AI Agents as part of AI Fundamentals, including its purpose, practical setup, common development workflow, limitations, and production considerations. Learning objectives Explain AI Agents in clear language. Recognize when it helps and when it does not. Apply it in a small, reviewable development workflow. Validate AI-generated output before accepting it. Practical developer workflow State the desired outcome and acceptance criteria. Provide only the relevant repository context. Ask Copilot for a plan or a small change. Review every suggestion and generated file. Run tests, linting, builds, and security checks appropriate to the change. Easy example Begin with a narrow request that explains one concept or proposes one small change. Plan this task in small verifiable steps. Edit only in-scope files, run the relevant checks, and report assumptions, changed files, and remaining risks. Easy-example verification Check that the response addresses the exact request. Compare technical claims with the repository or trusted documentation. Do not apply a suggestion until you understand it. Advanced example Use AI Agents in a production task with explicit scope, constraints, review gates, and recovery requirements. Plan and implement a production use of AI Agents. Limit changes to the named files and preserve public behavior. Include normal, edge, and failure tests. Run the relevant lint, test, build, and security checks. Report assumptions, evidence, tradeoffs, and rollback steps. Real-world example A development team uses AI Agents while working on a customer-facing application. The team supplies repository rules and acceptance criteria, keeps changes small, reviews the generated diff, and verifies behavior with automated checks and manual inspection. Common mistakes Using a vague request without constraints or success criteria. Providing too much irrelevant context or omitting the files that define behavior. Accepting generated code, commands, or claims without verification. Including secrets, personal data, or restricted source material in prompts. Allowing a large change to proceed without checkpoints and rollback. Production perspective Bound tool permissions, review the plan and diff, require validation, and never treat a successful command as proof of correct behavior by itself. Review checklist Is the intended outcome explicit? Is the supplied context relevant and safe? Does the result follow repository architecture and standards? Were edge cases, security, and accessibility considered? Is there test evidence and a safe recovery path?