Positioning
The AI Architect Roadmap
Eight rungs from foundations to enterprise AI, with an honest map of what this site covers deeply, where it is strong, and where the gaps still are.
6 of 8 rungs covered by the writing here. The flagship is rung 7, observability and cost, where the depth is deepest.
-
Foundations
Not yetThe Python and tooling that actually matter in production, reframed for working engineers rather than taught as an intro course.
Reframe planned, not yet written. The senior lens on a rung others tutorialise.
-
Learn AI & ML
Not yetThe machine learning you need to operate LLMs, not train them: tokenisation, context windows, and the production knobs that move cost and behaviour.
Reframe planned, not yet written. Operate, do not re-derive the maths.
-
Master GenAI
CoveredPrompting, RAG, tools, and the SDKs that wire an agent together on Bedrock.
- Part 1: Intent vs State. How AWS DevOps Agent Closes the Gap Between What Your System Is and What You Decided It Should Be
- Part 2: The MCP Server: Turning ADRs and Incidents into a Queryable Org-Knowledge Surface
- Part 3: Wiring It Into AWS DevOps Agent: AgentSpace, register-service, and the IAM Trust Policy That Ate My Afternoon
- Part 3: Building the AI Agent with Strands Agents SDK, Prompt Caching, and AgentCore Memory
-
Design AI Systems
CoveredArchitecting agents that survive contact with production: intent versus state, the harness that bounds a runaway loop, and the boundary the model never gets to cross.
-
Build AI Infrastructure
StrongThe infrastructure around the model: CDK, containers, local development, CI/CD with OIDC, and the cost controls baked in from the start.
- Part 1: Why I Chose Amazon Bedrock AgentCore (And What Lambda Gets Wrong for AI Agents)
- Part 2: CDK Infrastructure for Amazon Bedrock AgentCore (And Every Gotcha You'll Hit)
- Part 3: Building the AI Agent with Strands Agents SDK, Prompt Caching, and AgentCore Memory
- Part 4: Running Your AgentCore Agent Locally with Docker (The Right Way)
- Part 5: CI/CD for Bedrock AgentCore with GitHub Actions and AWS OIDC (No Stored Credentials)
- Part 6: Cost & Performance for Bedrock AgentCore: Prompt Caching, Model Selection, and CloudWatch Alarms
-
Security & Governance
CoveredThe controls that actually hold: deterministic authz boundaries in code the model never touches, guardrails as a backstop, and credential-free deploys.
- The LLM is not a security boundary
- Your LLM security diagram defends the wrong layer
- Part 2: CDK Infrastructure for Amazon Bedrock AgentCore (And Every Gotcha You'll Hit)
- Part 3: Wiring It Into AWS DevOps Agent: AgentSpace, register-service, and the IAM Trust Policy That Ate My Afternoon
- Part 5: CI/CD for Bedrock AgentCore with GitHub Actions and AWS OIDC (No Stored Credentials)
-
Observability & Cost
★ FlagshipWhere the depth is. Measuring what the probabilistic layer hides: runaway cost while the dashboard stays green, caching that ships off by default, latency that is not model speed, and a memory write that reads back empty.
- Field Notes: Three things I learned diagnosing a production Bedrock workload
- Field Notes: Turning prompt caching on for a production Bedrock workload
- Every dashboard was green while the agent burned six figures a year
- Field Notes: The AgentCore Memory write that returns success and reads back empty
- Part 6: Cost & Performance for Bedrock AgentCore: Prompt Caching, Model Selection, and CloudWatch Alarms
-
Enterprise Solutions
StrongProduction builds a business can actually adopt: an org-aware DevOps agent, a support assistant, and a study system that survived an architecture pivot.
- Part 2: The MCP Server: Turning ADRs and Incidents into a Queryable Org-Knowledge Surface
- Part 3: Wiring It Into AWS DevOps Agent: AgentSpace, register-service, and the IAM Trust Policy That Ate My Afternoon
- What a Year 10 study system taught me about production AI failure modes
- Part 1: Why I Chose Amazon Bedrock AgentCore (And What Lambda Gets Wrong for AI Agents)
Building on AWS and want a second set of eyes on where you sit on this ladder? Read the field notes, or get in touch.