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

Tools to know

This table lists the tool categories an AI platform engineer uses, current examples in each, and the pattern each category implements. The examples are current as of September 2026.

CategoryCurrent examplesPattern to learn
Coding agentsClaude Code, Codex, Cursor, OpenCode.Context files, planning first, skills, hooks, permissions, and headless runs.
Agent multiplexersHerdr, tmux, Zellij. Git worktrees for isolation.Isolate each agent, show agent state, keep sessions alive, and gate risky actions.
Agent SDKs and frameworksClaude Agent SDK and LangGraph. Vercel AI SDK, Pydantic AI, and Microsoft Agent Framework when they fit.Agent loops, state machines, and supervisors. Start with the model provider's SDK.
Durable executionTemporal. Lighter options are DBOS, Inngest, and Restate. Step Functions on AWS.Record each step, replay after a crash, wait a long time, and run each side effect once.
Model gatewaysLiteLLM, Portkey, Bedrock gateway options.One place for routing, keys, quotas, and cost tracking. That also makes the gateway a target for attackers.
Code sandboxesE2B, Modal, Daytona, Vercel Sandbox, Cloudflare sandboxes.Isolation model, outbound network control, and environments you throw away after use.
Tool protocolMCP and the MCP Inspector.Narrow tool contracts, authentication, audit logs, and distribution to teams.
Tracing and monitoringOpenTelemetry with its generative AI conventions. Langfuse, which ClickHouse acquired in January 2026 and which is still open source. New Relic.Traces instead of logs, cost and quality for each workflow version, and one instrumentation for any backend.
EvalsLangfuse datasets and experiments, or an equivalent.Versioned datasets, calibrated judges, and CI gates.
Cloud and modelsAWS Bedrock, Bedrock AgentCore, Lambda, SQS, Step Functions.Portable agent interfaces on top of managed services.
Local developmentAspire, formerly .NET Aspire. LocalStack, which is no longer open source. Moto. Docker.Environments that start with one command, recorded model calls, and fast, repeatable feedback for agents.