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

Show your work to employers

A hiring manager should not have to guess your experience from buzzwords. Structure your product's case studies and the Forge repository so that an artifact backs every claim.

Structure the Forge repository

A strong Forge repository has these top-level folders: /platform, /agents, /tools, /context, /evals, /infra, /observability, /agent-config for skills, hooks, and MCP settings, /docs/rfcs, /docs/adrs, /runbooks, and /security. Add an architecture diagram and a quick start that takes ten minutes.

Write these case studies

  • How the product cut cost per task and kept measured quality.
  • How a real production failure became a regression eval.
  • How I run several coding agents in parallel without losing control or quality.
  • How a team adopted a shared coding-agent setup: what worked, what did not, and the adoption numbers.
  • Why write tools require approval, and how the policy layer works.
  • How I evaluate retrieval separately from answer quality.
  • How workflow state survives worker restarts and duplicate deliveries.
  • How tracing made a multi-step agent failure faster to diagnose.

Practise these system-design questions

  • Roll out Claude Code and internal MCP servers to 300 engineers. Cover permissions, telemetry, cost control, and training.
  • Design an internal coding-agent platform for 500 engineers.
  • Design a multi-tenant agent execution service with a budget for each team.
  • Design a PR reviewer that keeps severe false positives below a threshold.
  • Design tools that let agents deploy to staging but never to production.
  • Design a context service over repositories, tickets, and documents that respects access control.
  • Design tracing and incident response for a workflow whose output varies between runs.

Lead with the problem, not the tool

Do not open with framework or tool names. Open with the constraint and the outcome. "We used a durable state machine because jobs run for 20 minutes and workers can disappear" says more than "I used LangGraph". "I run four isolated agents and gate merges on evals" says more than "I use Herdr". Show that the tool followed from the problem.