A clear, visual map of how modern AI actually works — from tokens and context windows to agentic coding, configuration, and MCP. Built for software engineers, not researchers.
1 GitHub / Wakefield Research survey, cited in source material.
AI tools no longer just finish your line. They plan, edit many files, run commands, and iterate in a loop — governed by repo-level instruction files and connected to live tools through MCP.
No single model wins everything. Claude leads agentic coding, GPT-5.x leads abstract reasoning, Gemini 3.x leads long context — and open models lead cost and privacy.
A perceive → plan → act → observe loop, with conventions like CLAUDE.md, AGENTS.md, Cursor Rules and SKILL.md giving agents persistent context.
An open USB-C-for-AI standard, now stewarded by the Linux Foundation, that lets any model talk to any tool through one protocol.
Each area is its own page — short, visual, and readable. Start anywhere; they build on each other.
Next-token prediction, tokens, context windows, sampling, the model landscape, reasoning models, hallucinations and knowledge cutoffs.
Explore foundations → 💸Input vs output pricing, prompt caching, and the cost-optimization stack that can cut your bill by 90%.
See the economics → 🛠️The autocomplete → chat → agent taxonomy, the 2026 editor and CLI landscape, and AI code review.
Tour the tools → 🤖What makes an agent, the action loop, autonomy levels, background agents, and multi-agent orchestration.
Meet the agents → ⚙️Skills, Cursor Rules, CLAUDE.md and AGENTS.md, custom instructions — and the configuration-stack mental model.
Configure your agent → 🔌What MCP is, why it matters, common servers, and its rapid cross-vendor adoption in 2026.
Plug into MCP → 🚦When to use AI and when not to, verifying output, keeping a human in the loop, vibe coding vs spec-driven development, and how the SDLC is changing.
Build the right habits →