ai-platform/roadmap
All pages

Start here

  1. How to use this roadmap
  2. Two kinds of role
  3. Read code, write it rarely
  4. Target profile
  5. Eight principles
  6. Work on a real product

Phases

  1. 01Direct coding agents and read their code
  2. 02Learn how models behave and what they cost
  3. 03Build retrieval that finds the right evidence
  4. 04Build tool servers with MCP
  5. 05Design agent workflows that survive failure
  6. 06Run agent workloads on AWS
  7. 07Trace every step an agent takes
  8. 08Build evals that catch regressions
  9. 09Run agents reliably in production
  10. 10Secure agents and the code they write
  11. 11Build an internal AI platform
  12. 12Get teams to adopt the platform
  13. 13Build Forge, the capstone platform
  14. Elective: Kubernetes

After the phases

  1. Staff skills
  2. 12-month plan
  3. A typical week
  4. Show your work
  5. Tools to know
  6. What to skip
  7. Final checklist

Start here

How to use this roadmap

This roadmap is a curriculum for becoming an AI platform engineer. Each phase gives you something to learn, something to build, and a checklist that tells you when you are done. At the end of every phase you should have work another engineer can inspect: running software, traces, evals, dashboards, runbooks, and written design decisions.

The roadmap assumes that coding agents write most of the code. You decide what to build, write the spec, check what the agent produces, and run the result in production. Typing code by hand is no longer the slow part of the job. Knowing what is correct and safe is.

The order of the phases matters

Phase 1 teaches you to work with coding agents, because you use them to build everything after it. The next phases cover model cost, retrieval, tool servers, agent workflows, AWS, tracing, evaluations, reliability, and security. Only after those do you build a platform for other teams. If you learn a framework before you understand the problem it solves, you cannot tell when the framework hides a bug.

Plan for about 9 hours a week

The plan takes about 12 months at 8 to 10 hours a week outside work. Agents do most of the typing, so the build work goes faster than it did a few years ago. Learning to operate what you build is what takes the time. If your day job covers the same ground, count those hours.

The phases overlap on purpose. For example, you add tracing in Phase 1, although tracing gets its own phase later.

Count only evidence

For each skill, ask what artifact proves you have it. A certificate proves little. A deployed service with a cost dashboard, a regression suite, failure tests, and a written design decision proves a lot. A product that real people use proves the most. For that reason the roadmap puts as much work as possible into a real product. The page "Do the work on a real product" explains how to pick one.

Work on two projects all year

You work on two projects for the whole year. The first is a real product with real users, where you practise running AI in production. The page "Do the work on a real product" explains how to pick one.

The second is a platform for engineers who build software with coding agents. This roadmap calls it Forge, but the name is only a working title. You start Forge in Phase 1 with your own agent setup, add to it in most phases, and finish it as the capstone project in Phase 13.

Four stages from builder to Staff engineer

You move through four stages. Each stage adds to the one before it.

  1. Builder. You ship useful AI products in days instead of weeks by directing coding agents well.
  2. Operator. You run those products in production and find the cause of failures from evidence.
  3. Platform engineer. You turn what you built into something other teams can reuse.
  4. Staff engineer. You set technical direction, write standards, and get teams to adopt them without formal authority.