Upskilling for the agentic shift
The roles are being redrawn, and the governance has to come with them
A companion cheat sheet. The roles are being redrawn: more than 180,000 people have gone in tracked layoff rounds this year, over half citing AI. The people staying are the ones who can supervise agent work. This sheet maps the upskilling that matters, and the governance that has to come with it.
What the cheat sheet covers
A three-page companion for people learning to work with agents. It starts from the line the market is drawing: on one side, people who can work with these systems; on the other, people who work around them. It treats two skills as being tested. The first is building with agents, meaning not prompt tricks but wiring an LLM to tools, giving it a task and staying in control of what it does. The second is checking what it produces, because judgement is the half of the job that survives automation.
Its case for carrying governance alongside is direct. Upskilling to ship faster without upskilling to check is how the failures happen, and the roles being removed are often the ones where a human sat between an action and its consequence. Learn the tools, and learn where the checkpoint has to stay.
Key points
- Free and cheap beats prestigious. Vendor certifications teach you one product; building teaches you the shape of the problem.
- Three recommended Udemy courses from Ed Donner: the AI Engineer Agentic Track over six weeks, marked as the place to start; AI Coder over three weeks, on using coding agents such as Claude Code, Cursor, Copilot and Codex; and the eight-week LLM Engineering core track for the deeper foundation.
- Free to start: the full course code is public on GitHub, and Anthropic’s “Building effective agents” is named as the seminal short read.
- How to spend the time: build one thing end to end before starting the next module, and break it on purpose. Watching what the agent does with a bad instruction is the governance lesson.
- Six questions for checking agent output: Provenance integrity, Supervisory clarity, Scope definition, Review substance, Constraint fidelity and Capability calibration. Review substance is where the theatre gap shows up first.
- Each is scored from Absent to Substantive. Most teams stall at Level 4, documented, and call it done. Used as a question rather than a scorecard, the ones you cannot answer are where the risk lives.
New material lands week by week
Research, the kernel devlog, news and the papers are all published here first. Get them by email, or follow the feeds.
The newsletter is moving in-house and relaunching soon. Until then, email us to be notified when it opens, or follow the RSS feed.
Or follow the work directly: Research · Kernel devlog · News · CDI assessment