Free · sourced · kept current
Guides for senior engineers moving into AI
A free reference library for senior software, DevOps, and cloud engineers crossing into AI engineering and architecture. Every guide is answer-first, sourced, and dated. No sign-up required. The guides, the Weekly AI Engineering Brief, and the Engineering Workflow Library are the free, public layers of the aiArch Continuous AI Professional Development membership.
Certifications & Career
Are certs worth it, which one, and how to actually make the move — the transition, the credentials, and the job search.
Which certs are worth it
Honest worth-it / maybe / skip verdicts on CCAR-F, the AWS exams, and the Cloudflare reality.
AI certification exam guide
Confirmed CCAR-F and AWS AIP-C01 domains, weights, and pass marks — the factual blueprint.
The Anthropic (Claude) certification
The CCAR-F credential explained — exam domains, pass mark, cost, and whether it's worth your time.
AWS Certified AI Practitioner
The foundational AIF-C01 exam — the five domains and weights, who it's for, cost and format, and how to prepare.
AI certifications compared
The major AI certs side by side — AWS, Google, Microsoft, Anthropic — by level, focus, cost, and who each is for.
Agentic AI certification
What the new agentic-AI credentials cover, the options emerging, and whether they're worth your time yet.
The Cloudflare AI certification question
No official cert exists — the honest answer, plus the fully published CAPA blueprint: 30 doc-anchored task statements and a practical capstone.
The AI skills split
The tech job market is bifurcating — AI roles up, generalist roles down, seniors holding steady. The sourced data and what it means for you.
You don't need machine learning
Why becoming an AI engineer doesn't require learning to train models — and what to learn instead.
Forward Deployed Architect
A real title at Palantir, Google Cloud and NVIDIA — what it asks for, what it pays, and why the certifications sold for it come from third parties.
What the research actually says
Stanford AI Index, WEF, and OECD on AI and tech jobs — even-handed and sourced.
From DevOps to AI engineer
What transfers from ops, what's actually new, and a realistic transition path for senior platform engineers.
The AI engineer roadmap
A stage-by-stage skills map — what to learn in order, from LLM fundamentals through agents, evals, cost, and deployment.
AI engineer interview questions
The questions that actually come up — fundamentals, agents, RAG, evals, system design — with senior-level model answers.
The AI engineer resume
What to show, how to reframe your existing experience, and the mistakes that get a strong engineer filtered out.
AI engineer vs ML engineer, data scientist, SWE
How the AI engineer role differs from adjacent roles, where they overlap, and which to aim for.
The AI roles, decoded
Engineer, architect, builder-architect, AgentOps, solutions architect — what each does and who it suits.
Where to find AI jobs
Curated channels for AI/agentic roles, plus how to decode a posting and spot the real requirements.
AI architect
Designing, shipping, and operating production AI systems — the senior role the engineer path grows into.
AI engineer
What an AI engineer does, the skills and certs, and how a senior software engineer moves into building production AI systems — no ML required.
AI developer
How the AI developer role differs from an AI engineer — the skills, certs, and how a working developer starts building AI-powered applications.
AI solutions architect
The client-facing architect role — how it differs from an AI architect and engineer, the skills, certs, and how to move into it.
Forward deployed engineer
The fastest-rising AI role — embedded with customers to build and ship AI systems fast. Skills, salary, and how to get there.
How to become an AI engineer
A practical roadmap for experienced engineers — what transfers, what to learn, how long it takes, and the portfolio that hiring screens on.
AI engineer salary in 2026
What AI engineers actually earn by seniority and role — engineer vs architect — with honest caveats on the numbers.
AI platform engineer
Building the platform agents run on — inference, gateways, retrieval, observability. The role for devops and cloud engineers.
The agentic AI engineer
The specialism in building agentic systems — where prompt-engineering energy went — its skills and demand.
Is prompt engineering dead?
No — but it was absorbed into context engineering and agentic work. What changed, and where the careers went.
AI product manager
The more-technical PM role — evals, model trade-offs, cost, agentic UX — and how a technical person moves into it.
Agentic Systems
How agentic AI actually works — the loop, the frameworks, MCP, and where to run agents in production.
What is agentic AI?
A precise definition, the agentic loop, and how agentic AI differs from generative AI.
Agentic AI design patterns
The augmented LLM, workflows vs agents, and the five patterns — with when to use each.
Agentic AI architecture
The six components of a production agent system — reasoning core, orchestration, memory, tools, retrieval, operational plane — mapped across Anthropic, AWS, and Cloudflare.
AI agent frameworks
The landscape — LangGraph, CrewAI, AutoGen, the Claude and OpenAI agent SDKs, Google ADK, Strands — and how to choose, or skip one.
LangChain vs LangGraph
How the two differ, how they fit together, and when to reach for each.
The Claude Agent SDK
Anthropic's SDK for building agents — the loop, tools, MCP, and subagents — and when to use it.
Multi-agent orchestration
Coordinating multiple agents — the topologies, when multi-agent is worth it, and the hard parts.
Browser agents
AI that drives a real browser — how it works, what it's good for, and the limits.
Computer use
Giving a model the ability to see a screen and control mouse and keyboard — how it works, the limits, and the risks.
What is MCP (and how to build a server)
The Model Context Protocol — tools, resources, prompts; MCP vs an API; and building a server.
How to build an MCP server
A step-by-step build — tools, resources, prompts — and connecting it to an agent.
Agentic AI coding tools
How agentic coding tools work, how they differ from autocomplete, and how to choose one.
Where to run your Claude agent
Anthropic API vs AWS Bedrock vs Cloudflare — cost, latency, lock-in, and the abstraction boundary. Kept current.
Amazon Bedrock AgentCore
AWS's managed runtime for production agents — Runtime, Memory, Identity, Gateway, Observability — and when to use it.
Context engineering
Designing what enters the model's context window — retrieval, memory, tools, compaction — the discipline that succeeded prompt engineering.
Production & Ops
Running agents in production — evals, observability, guardrails, gateways, routing, and cost control.
How to evaluate an LLM agent
Evals as the test suite for probabilistic systems — golden sets, offline vs online, and LLM-as-judge.
LLM observability
The traces, metrics, and logs that make agents debuggable in production — and how it differs from evals.
AI guardrails
The input and output controls that keep LLMs and agents safe, on-topic, and valid — types, trade-offs, and frameworks.
Cloudflare AI Gateway
Caching, rate limiting, observability, and cost control for LLM calls — features, pricing, and setup.
What is an AI gateway?
The LLM control plane — routing, caching, observability, and cost control between your app and the model providers, and when you need one.
LLM routing
Sending each request to the right model — the strategies, build vs buy, and how it cuts cost without losing quality.
LLM cost optimization
Why input tokens dominate multi-turn agents, and the levers: caching, routing, turn caps, context management.
Your ops instincts are the moat
How DevOps/SRE/cloud instincts — observability, cost, reliability, security — map onto running production agents.
AI agent security
The threat model for agentic systems — prompt injection, excessive agency, tool exposure — and how to design an agent that's safe to run.
Prompt injection
How direct and indirect prompt injection work, why it's unsolved, and the defenses that actually help.
What a Claude agent actually costs
Per-token rates, runtime add-ons across API/Bedrock/Cloudflare, and how to estimate a multi-turn agent. Maintained.
Opus vs Sonnet vs Haiku
Price and best-fit workload for each Claude model, and the Opus → Sonnet → Haiku routing pattern.
Data & Retrieval
Grounding models in real data — RAG, embeddings, vector stores, and when fine-tuning is the better call.
What is RAG?
Retrieval-augmented generation explained — why it grounds models in real data, how the pipeline works end to end, and its limits.
Embeddings explained
The vectors that capture meaning — how they power semantic search and RAG, and how to choose an embedding model.
RAG vs fine-tuning
What each does, the cost and effort trade-offs, when to use which, and why most production systems combine them.
Agentic RAG
Making retrieval a decision the agent controls — how it differs from a classic RAG pipeline, and how to evaluate it.
Vector database comparison
Choosing a store for RAG — Pinecone, Weaviate, Qdrant, pgvector, Chroma, Milvus — and whether you even need one.
Governance
The regulatory and oversight layer for agentic systems — the EU AI Act, ISO 42001, and agent governance controls.
The EU AI Act for engineers
Risk tiers, provider vs deployer, transparency duties, and how obligations map to your architecture.
ISO 42001
The AI management system standard — who needs it, how certification works, and how it compares to the NIST AI RMF.
AI agent governance
Identity, audit, oversight, and policy — the controls for running autonomous agents accountably.
LLM subprocessors
Adding a model changes who processes your users' text — the serving path, the router's DPA, and the check to run before you merge.
Ask the guides
Ask a question and get an answer grounded only in these guides, with a link to the source. This is a small, public preview of the senior-calibrated coach inside the membership — it answers from our published material, not the open web.
Grounded in the aiArch guides. Answers can be imperfect — follow the source link to read the full guide.
The guides are the map. The membership is the path.
The membership turns this material into a mastery-based program with a senior-calibrated coach, evals on your own work, and a portfolio — across Anthropic, AWS, and Cloudflare.