Last week, a demo with the Dev team tested an AI Debugging Agent that was integrated directly into the developer workflow through the Kiro IDE.
Developers only need to include one question, for example:
"Why did this API start failing after the latest deployment?"
Instead of just giving an answer based on available knowledge, agents can orchestrate multiple steps to investigate:
Kiro IDE → MCP → AgentCore Runtime → Debugging Agent → Skills → Internal Tools
Some current capabilities:
🔎 Retrieve API / application information
📂 Source code and GitLab testing
📋 Retrieve information from Jira/Confluence
☁️ Audit AWS resources and CloudWatch
🧠 Maintain context via AgentCore Memory
🤖 Invoke LLM on Amazon Bedrock for Analytics
📊 Execution Observation via AgentCore Observability
🧪 Evaluate responses through AgentCore Evaluations
The interesting point here is not just "AI answers dev eloper's questions".
The agent began to be able touse tools, collect evidence from multiple systems, and perform a debugging workflow to get to the root of the problem.
This is one of the ways tobring Agentic AI into the software engineneering, where AI not only supports developers to think, but alsobegins to participate directly inthe process of doing things.
Top comments (0)