Pair Programming with Ai Agents & AI Coding Workflows by Meenakshi Sundaram, Manager @Google

​Gemini CLI,Claude Code, Codex, and OpenClaw are turning software engineering into an agent-orchestrated workflow — where AI writes features, hunts bugs, fixes failures, and reviews pull requests before you finish your coffee.

​This session explores how builders are shipping faster by running autonomous coding loops instead of traditional dev cycles.

​What’s happening now

​⚡ Gemini , Claude Code & Codex generating multi-file features and refactors
⚡ OpenClaw executing real dev tasks across shell, browser, and system tools
⚡ Agents debugging failures and proposing verified fixes
⚡ MCP making tools, context, and systems agent-aware
⚡ AI reviewing PRs and enforcing architecture & security rules

​What we’ll explore

​🔍 Debugging in the agent era
• plan → execute → observe → repair loops
• tracing failures across services & agent workflows
• semantic correctness vs runtime correctness

​🏗 Architecture & System Design for agent workflows
• designing systems that are debuggable by design
• building extensible, tool-aware architectures
• orchestrating agents safely across environments

​⚡ MCP (Model Context Protocol) as the power layer
• making tools & services discoverable to agents
• context injection & intelligent runtime decisions
• enabling secure, scoped tool execution

​🛡 Automated Code Review Pipelines
• AI PR reviews & architectural linting
• policy, security & performance guardrails
• continuous quality gates & self-improving codebases