Building production vertical agents in 2026 — without the hype
A question-driven series for builders and decision-makers. Read the preface first — it explains what this series is actually for and what you will be able to do after reading it.
A working mental model of how custom AI agents are built and run in production — who this series is for, what you will be able to do, and a map of the eight sections ahead.
3 min · Updated June 2026
Read the preface →What is an agent, really — and does my problem even need one?
The stripped-down definition of an agent, the single most important distinction in the field (workflow vs. agent), the six concerns every production agent decomposes into, and the first design question every team should ask.
5 min · Updated June 2026
Read this part →Why did “prompt engineering” stop being enough, and what replaced it?
The shift from prompt engineering to context engineering, the five failure modes to learn by name, the practical toolkit for managing what the model sees, RAG in 2026, and prompt caching as the most underrated cost lever.
9 min · Updated June 2026
Read this part →How does an agent remember things — and why is “a vector database of old messages” the wrong answer?
Context vs. memory, the layered architecture from thread checkpointing to episodic and procedural long-term memory, the memory framework landscape, why vendor benchmarks are unreliable, and the pragmatic default recommendation.
7 min · Updated June 2026
Read this part →How do agents interact with the real world — and what security problem came with the answer?
What MCP is and why it took over, MCP vs. plain function calling, designing tools agents can actually use, Agent Skills, the OWASP Top 10 for Agentic Apps, and the full defensive posture for production.
10 min · Updated June 2026
Read this part →When do you actually need multiple agents — and what does a well-structured system look like?
The multi-agent reality check, the shared session-state mechanic, five patterns with ADK code examples, the planner/generator/evaluator trio, and the canonical within-agent design patterns from the Anthropic taxonomy.
10 min · Updated June 2026
Read this part →Why do agent demos fail to become products — and what does the gap actually consist of?
Durable execution, observability and eval pipelines, layered guardrails, human-in-the-loop design, governance and compliance (EU AI Act, ISO 42001, sector floors), and the six-step cost optimisation order.
11 min · Updated June 2026
Read this part →If I’m building a production vertical agent in Python today, what should I actually use?
The full recommended stack table covering orchestration through guardrails, a four-phase rollout sequence, and four trigger conditions that should change your choices.
5 min · Updated June 2026
Read this part →What do real deployments actually look like — and what should I discount from everything I’ve read?
The pattern across marquee deployments, four things to stay skeptical about, and a tight one-paragraph synthesis of everything in the series you can carry with you.
5 min · Updated June 2026
Read this part →