AI engineering, one day at a time.

For anyone who has used an AI chat product: in five days, learn to turn a fuzzy requirement into a prompt the model executes reliably, and verify it with a small test set.

In five days go from chatting with Claude to running a real development task end to end with Claude Code — explore, plan, implement, verify, commit — then wire up the API and the Agent SDK.

In five days get fluent with the Codex CLI, cloud tasks, the Responses API, and the Agents SDK, then spend the last day running the same task through both Codex and Claude Code for a real side-by-side comparison.

For developers who write TypeScript or Python: in seven days, learn to read, implement, debug, and safely ship MCP servers and clients against the 2026-07-28 spec.

In seven days, learn to turn something you already do well into a SKILL.md, so any client that supports Agent Skills opens it up on its own when it's needed, and package a team's conventions for distribution.

Treat the context window as a piece of luggage with fixed capacity: in five days, learn to break down what fills it, work out its bill, and keep it under control with pruning, compression, and isolation.

For React developers: in seven days, hand-build an agent workbench in the browser. Align with the open AG-UI protocol and turn a raw event stream into an interface that renders without jank, pauses for approval, replays history, and reads correctly to a screen reader.

For anyone who has shipped an agent: build a zero-dependency evaluation framework in seven days. Turn non-determinism into pass^k, mine your incidents for a benchmark set, put the model judge itself through calibration, score the trajectory rather than just the ending, wire up telemetry and a cost panel, and finally block regressions before they merge.

Take one internal assistant that happens to hold all three parts of the lethal trifecta and turn it into a target: five days from threat modeling to architectural defense, runtime controls and a red-team loop that drives attack success to zero while keeping the agent useful.

For anyone who has already shipped at least one agent: hand-build an unattended runtime in seven days. Move the authoritative state out of the context window, keep progress alive on disk and in git, block premature completion claims with an end-to-end gate, intervene automatically on doom loops and blown budgets, and finally let it run through the night and hand you a report you can actually check in the morning.

For anyone with no AI background: four weeks from the LLM API all the way to multi-agent orchestration and a job-hunt sprint.

For developers who write TypeScript or Python: in fourteen days, turn Retrieval-Augmented Generation from a working prototype into an evaluable, attributable, production-ready service — get keyword search solid first, then push it layer by layer toward production with vectors, reranking, query rewriting, and evaluation.

For developers who write TypeScript: in fourteen days, turn a single sentence into one episode of a vertical short drama, then scale it into a production platform that can produce in parallel, be reviewed, and keep cost under control.

For developers who write TypeScript: in twenty-one days, start from an empty-shell REPL, add one capability a day, and end up with a terminal Coding Agent that can read code, edit files, run tests, and work in parallel on a real repo, installable and distributable, with zero framework and zero runtime dependencies throughout.