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

Phase 11 · Months 10 to 12

Build an internal AI platform

Run the platform like a product

A platform has users, workflows, onboarding problems, reliability expectations, and a roadmap. Interview developers and find the problems they keep hitting: model integration, tracing, tools, budgets, evaluation, prompt deployment, approvals, identity, context, and getting consistent results from coding agents. Build the smallest shared solution that removes a real problem.

Build both halves of the platform

An internal AI platform has two halves. The agent runtime runs the agents your organisation ships. It holds agent definitions, versions, policies, budgets, eval settings, and telemetry. The coding-agent setup makes every engineer's coding agents safer and more useful, with shared skills, context-file templates, hooks, MCP servers, permission policies, and telemetry. The Staff role is mostly the second half, and most roadmaps cover only the first.

Offer small, safe building blocks

Offer small APIs such as create_agent, register_tool, run_eval_suite, require_approval, set_budget, and trace_workflow. Provide templates for common patterns instead of one framework with endless options. Make the defaults safe: retries have limits, tracing is on, every run needs a budget, and privileged tools need a policy.

Support many teams

Separate teams and tenants. Add quotas, audit logs, feature flags, staged rollouts, and version compatibility. Keep migrations routine. A platform that breaks every team that uses it slows the organisation down.

Build the first slice of Forge

Put the PR reviewer, the context engine, and the tool servers behind one developer interface. Include an SDK and CLI, a documented local environment, an example agent, standard tracing, the model gateway, and an eval hook. Add your shared agent setup from Phase 1: skills, hooks, and MCP settings. Then build a second agent using only the public interfaces, to find parts that depend on internals by accident.

You are done when

Another engineer can create and run a small agent on your platform, and get a well-configured coding-agent setup from it, without knowing its internals. You still see spend, reliability, and policy for every agent in one place.

Exit checklist

0 of 6 done