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 13 · Months 9 to 12

Build Forge, the capstone platform

What Forge is

Forge is an internal platform for software delivery with AI agents. It runs a small number of useful engineering agents and handles the hard production concerns for all of them in one place: context, tools, identity, budgets, tracing, evaluations, approvals, and reliability. Its first users are you, on your product and your other work, and ideally a small team.

Start with four agents

Start with a PR review agent, a dependency-upgrade agent, a documentation agent, and an assistant for incident investigation. Add a security agent only after your permissions and evals are mature. Every agent uses the same platform interfaces for models, tools, context, tracing, and policy.

Use this reference architecture

Forge has these parts:

  • A CLI and SDK, and optionally a GitHub App.
  • An agent gateway and control plane that send work to a durable runtime.
  • Separate components for context retrieval, MCP tool servers, and stored state.
  • One place that all model provider calls go through.
  • OpenTelemetry traces sent to your tracing tools, and experiments and evals recorded in an eval system.
  • A shared coding-agent setup.
  • Deployment through infrastructure as code.

Meet these operating requirements

Every run has a hard budget and a timeout. Forge authorises every side effect. High-risk actions need approval. You can identify every workflow version. Traces connect model calls, retrieval, and tools. Production failures become eval cases. You can replay jobs from the queue. Secrets never appear in prompts. Every team has quotas and audit logs.

Write the Staff-level documents

Write an RFC for the platform architecture and a decision record for your choice of workflow engine. Write a threat model, an SLO document, an incident runbook, a model provider strategy, an eval method, a rollout plan for coding agents, and a cost report. Draw an architecture diagram and record a short demo.

Check it against this test

A reviewer who reads the repository should conclude four things. You understand how agent workloads behave in production. You know how to limit what varies between runs. You make sound trade-offs in cloud, reliability, security, and developer experience. You explain those decisions clearly. The platform should also be in real use.

You are done when

At least one real user besides you uses Forge every day, it brings the earlier phases together, and every claim about it on your CV links to an artifact in the repository.

Exit checklist

0 of 8 done