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

After the phases

What to skip

These topics look useful but cost more time than they return. Review this list at every quarterly checkpoint, because tools in this field change fast.

Writing code by hand as a training goal

Build reading fluency and understand the concepts. Let agents type. The exception is a short, deliberate block of interview preparation.

Every new agent tool

A new multiplexer, editor, or agent appears most weeks. Use one tool per category, review the choice every quarter, and spend the rest of your attention on the patterns. Every candidate knows the tools. Judgement is what sets you apart.

Deep ML research before systems skills

You need enough machine learning to reason about model behaviour and inference. This roadmap targets platform engineering, so do not postpone production systems until you can train a transformer from scratch.

Collecting frameworks

Do not build one tutorial in every agent framework. Pick one or two after you have seen the underlying loop. Ask which concepts would still matter if the framework disappeared next year. Be careful with frameworks that hide a lot of prompting, such as CrewAI. You get a demo running in an hour, and then you struggle to debug it.

Prompt engineering on its own

Prompts matter, but reliable systems depend on retrieval, tool contracts, policy, structured output, evaluation, and tracing. Treat a prompt change like a code change: version it, evaluate it, and roll it out safely.

Certificates without production experience

An AWS certificate can give your study a structure. A running agent service on a queue, with least-privilege IAM and a game-day report, proves much more.

Kubernetes before you need it

The roles this roadmap targets rarely ask for it. Use the simplest runtime that fits the workload.

Tools to learn only when a job needs them

Learn durable execution, sandboxes, and gateways properly. Learn Aspire and LocalStack only well enough to apply for AWS and .NET platform roles. Do not study New Relic on its own, because OpenTelemetry covers what you need. Learn everything else when a specific job needs it.