AI directory
A structured directory for understanding AI coding agents, runtime architecture, tools, permissions, MCP, sessions, and migration discipline.
Use this directory
Use this directory when you want the operating model behind AI coding agents rather than another quick product comparison. It connects the agent runtime stack, tool systems, permissions, MCP integration, plugin/session design, and migration practices into one editorial entry point.
Understand the loop, state, tool execution, and trust boundary before comparing model scores.
Look at file access, shell execution, MCP integrations, hooks, and permission gates as product features.
Treat clean-room rewrites, parity audits, and session continuity as first-class engineering work.
Topic hub layer
Start here
Claw Code's public docs and parity repo offer a useful blueprint for how modern AI coding agents are actually structured beyond the model layer.
Claw Code's parity repo shows why modern coding agents often split responsibilities between Rust for runtime-critical paths and Python for orchestration and migration.
A coding model becomes a real agent only when tool execution, permission policy, and MCP integration are designed as one coherent system.
Hooks, plugin registries, and persistent sessions are what turn an AI coding assistant into an extensible platform instead of a one-shot demo.
Claw Code's parity workflow offers a strong model for teams rebuilding or migrating complex agent systems without drifting into vague rewrites or cargo-cult copies.
Supporting analysis
These articles deepen the directory without turning it into a thin generated list.
Announcements from NVIDIA and Anthropic between March 16 and March 17, 2026 show the AI race shifting toward long-running agents, policy runtimes, and physical-world execution.
Microsoft's Agent 365 points to a new enterprise AI battleground: a control plane for agent identity, policy, observability, and governed tool access.
Google's March 2026 Pixel Drop suggests Gemini is moving beyond chat and becoming a cross-app action layer for mobile, which may matter more than another model benchmark.
On March 6, 2026, Anthropic and Mozilla said Claude-assisted research helped uncover 22 Firefox CVEs, signaling AI's move into real security work.
OpenAI's March 2026 GPT-5.4, Codex Windows, and Codex Security releases point to a bigger shift: the company is building a full agent stack for professional work.
The definitive comparison of 2026's hottest AI agents—coding assistants, task automators, and personal AI butlers. Find your perfect agent stack.