Upskilling your existing engineers for AI
Upskill engineers by having them build a real internal AI project, with an experienced practitioner guiding them through problem framing, architecture, evals and deployment. Courses alone produce certificates, and building on your own problems produces skills and a working system at the same time.
Why courses don't transfer
Teams complete online AI courses and still can't ship an AI feature, because the courses don't include their data, their constraints or their deployment environment. Your senior engineers already know your systems, and adding AI to that is usually faster than hiring people who know AI but not your business.
Format
| Week | Focus | Output |
|---|---|---|
| 1 | Problem and data | Problem statement, metric, data access, a raw solution without AI if one exists |
| 2 | First working version | End-to-end path on real data |
| 3 | Evaluation | Eval set with a domain expert, first error analysis |
| 4 | Fixes | What broke, what changed, how it was checked |
| 5 | Deployment | Running in a real environment with tracing |
| 6 | Demo | Results to stakeholders, with eval numbers and limitations |
Theory sessions happen only where the build needs them, for example tokenization and embeddings before retrieval work.
Rules that make it work
- One real problem per team, with a business owner.
- Weekly reviews of what broke and how it was fixed.
- Engineers do the engineering work: issues, pull requests, reviews, tests, deployment.
- A demo at the end with evals, not only a walkthrough.
- Leaders join at least one session, so they understand the trade-offs too.
Measure it
Count systems in production after the program, and engineers leading the next one, rather than completion certificates.
From the floor
In the AI agents bootcamp orientation, I said there's a way to build agents for a very specific task, and that I had recently run a training with a company where we built an AI agent for their sales division, and after implementing that very specific agent, a lot of hours were reduced.
Related
- Learn by building
- Structuring an AI team
- AI readiness assessment
- Backend to AI engineering