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

Skills a Staff engineer needs

Technical depth is necessary for Staff, but it is not enough. A Staff engineer moves from solving hard problems alone to making a whole organisation more effective. You still ship, mostly by directing agents. Your most valuable work becomes standards, architecture, mentoring, incident leadership, and systems that make other engineers and their agents faster.

Decide under uncertainty

Make reversible decisions fast. Take your time over decisions you cannot reverse. Write down assumptions, constraints, and alternatives. When requirements are unclear, prefer designs that keep options open. After you build something, review the decision with real data instead of defending it. Agents make it cheap to prototype two options before you choose, so use them for that.

Write RFCs and decision records

Write a short RFC for a change that affects several teams. Write a decision record for an important implementation choice. A good document states the problem, constraints, options, recommendation, trade-offs, rollout, and how you will measure success. An agent can write the draft. The decision and the reasoning must be yours.

Lead without authority

Pair with teams, listen to their problems, and earn adoption by being useful. When you propose a standard, make moving to it easier than staying on the old way. Collect cases where the standard does not fit, and document a way out. Rolling out coding agents across a team is a clear current example of this skill, and you can practise it now.

Lead incidents

During an incident, reduce uncertainty. Establish the impact, assign investigation threads, keep a timeline, decide on mitigations, and communicate clearly. Afterwards, fix the system instead of blaming people, and turn the failure into a test, an alert, a runbook, or an architecture change.

Teach others

Teach through design reviews, pairing, and reusable examples. Run workshops on agent security, tracing, eval design, and working with coding agents. Review other teams' agent architectures. Raise the level of the whole organisation, instead of becoming the only person who can run the platform.

Practise with these documents

Write each of these at least once:

  • An RFC for a standard way to trace agents.
  • An RFC for rolling out coding agents to an engineering organisation, covering permissions, telemetry, cost control, and training.
  • An RFC for the security model of internal MCP servers.
  • A decision record on a workflow framework versus a custom state machine.
  • A decision record on how to abstract model providers.
  • An SLO document for the PR review agent.
  • A threat model for agent tool calls and coding-agent permissions.
  • A postmortem for a runaway token-spend incident.
  • A runbook for a model provider outage.
  • A guide on how to add an eval before you change a prompt.