Q: What are AWS AI agents for enterprise? A: AWS AI agents for enterprise act on goals rather than just prompts: they read context, call tools, and take steps, and on AWS they can pull documents from knowledge bases, call APIs, run code actions, and log every move. They operate across code, data, and tickets to find patterns of debt, suggest changes, open pull requests, and write tests. Q: How can AWS AI agents for enterprise help cut tech debt? A: They can scan repositories, build logs, incident notes, and dependency files to find and rank problems like dead code, duplicate modules, outdated SDKs, and flaky tests. Agents can suggest changes, open pull requests, generate missing tests, and automate safer changes to speed releases and reduce manual handoffs. Q: Which AWS services are recommended to build agent workflows? A: Key AWS building blocks include Amazon Bedrock for hosting foundation models and planning steps, Amazon Q Developer for reading code and suggesting tests, AWS CodePipeline and CodeBuild for running builds and tests, and Step Functions with EventBridge for orchestration. Monitoring and security services such as CloudWatch, X-Ray, Amazon Inspector, Security Hub, IAM, and CloudTrail provide signals and audit trails for agent actions. Q: What governance and guardrails should teams set for safe agent use? A: Teams should scope agent access to specific repositories, services, and environments, require human approvals for code, infra, and data changes, and enforce pipeline policies like linting and security scans. They should also keep sensitive data out of prompts, use private knowledge bases with access controls, and log prompts, tool calls, and outputs for audits. Q: What does a 30-60-90 day roadmap look like for adopting agents? A: In the first 30 days pick one service with clear debt, enable an IDE agent to explain code and write tests, set up a Bedrock agent to catalog dependencies, and add pipeline policy checks. Over the next 60 days scale to 3–5 services with repeatable templates, automate dependency upgrades with scheduled PRs, add canary deploys and auto-rollback, and publish weekly metrics; by 90 days create a debt playbook, offer golden pipelines and starter repos, and review guardrails, budgets, and access scopes. Q: How should teams manage dependency hygiene and automated rollouts with agents? A: Automate checks on libraries and SDKs to flag risky versions, propose safe bumps, bundle small upgrades to reduce churn, and roll back quickly if tests fail. Use pipelines, canary or blue/green deployments, policy-as-code, and automated rollback on health alarms to make change boring and reduce blast radius. Q: What metrics should teams track to measure impact on tech debt? A: Pick a few clear metrics tied to business goals and track them weekly, such as PR cycle time, review load, number of libraries upgraded per month, and test coverage. Also monitor operational indicators like flaky tests reduced, build stability, incidents, MTTR, change failure rate, cost per deploy, and engineer satisfaction. Q: What common pitfalls should teams watch for when deploying agents and how can they be avoided? A: Common pitfalls include poor context from missing documentation, over-automation that removes human oversight, hallucinations in agent outputs, change fatigue, and secret leaks. Avoid these by investing in a clean knowledge base, keeping humans in the loop for risky changes, gating outputs with tests and policy checks, rolling out in small waves, and masking tokens while scanning prompts and logs.