Forward-deployed
/
Learning zone
Hands-on tracks for PMs and engineers moving into AI — from first principles to production.
The engineering discipline under production LLM systems — inference, retrieval, evals, observability, safety, cost. Designed reading editions.
ModuleBuild a coding agent's harness from scratch — loop, tools, context, memory, subagents — then use the real SDK. Build it / use it, ships an artifact each lesson.
ModuleProcess automation from scratch — build a token engine, wait states, and a job executor by hand, then run real BPMN on the Flowable engine. Concept-first for PMs, with a build layer for engineers.
ModuleReason from fundamentals and build range across disciplines — the method, a latticework of mental models, becoming a polymath, and learning how to learn.
ModuleThe instinct for what makes a product succeed, for APMs & PMs moving into AI PM — motivation, empathy, creativity, communication, domain expertise, and product sense for AI.
ModuleRead systems like an engineer — architecture, APIs, data, latency, reliability, and tech debt — with a diagram in every lesson. For APMs & PMs moving into AI PM.
ModuleThe operating discipline of shipping — the role, specs, prioritization, execution, metrics, and releases — with a diagram in every lesson. For APMs & PMs moving into AI PM.
ModuleWhat agents actually are — the loop, tools, memory, planning — plus reliability, security, and economics. Opens with a knowledge graph; a diagram in every lesson.
ModuleTreat what the company knows as a product — entities and ontologies, the construction pipeline, GraphRAG, governance, and the business case, in product leader language.
Generative AIWhat makes AI "generative," the five modalities, why output is probabilistic, the four-layer product stack, and build vs. buy vs. fine-tune. Opens the Generative AI family.
Generative AITokens, the context window, the jagged frontier, prompting, sampling, choosing a model, and the order to reach for prompting, RAG, or fine-tuning.
Generative AIThe request/response contract, authentication, rate limits, streaming, retries, structured output, webhooks, and fitting a model call into a real system.
Generative AIGrounding models in your data — embeddings, vector databases, chunking, retrieval quality, and when to reach for long-context, fine-tuning, or a graph.
Generative AIMemory as a product decision, the three shapes it takes, retrieval as its most common implementation, and the trust failures it has to be designed against.
Generative AIThe line where an AI product stops talking and starts doing: designing a tool worth trusting, and keeping the permission boundary around it real.
Generative AIThe loop behind every agent, how much autonomy a task needs, what keeps it reliable across many steps, and when not to build one at all.
Generative AIOrchestrating more than one agent when a single loop isn't enough, and what it takes to make a workflow durable enough, and valuable enough, to own end to end.
Generative AIWhy the eval set is the product spec for a non-deterministic system, and the order to actually build the eval and observability stack in.
Generative AIJailbreak, injection, extraction, and poisoning are four different attacks — and why governance only counts once it becomes compliance evidence a regulator or buyer can check.
Generative AIWhich lever fixes which cost driver, the build-vs-buy breakeven done as arithmetic, and the FinOps practice that turns attribution into governance before the invoice, not after.
ModuleHow real systems are designed at scale — from rate limiters to stock exchanges — with the architecture, tradeoffs, and failure modes that shape product decisions. 28 systems across 8 lessons, diagrams included.
ModuleTreat what the model gets to see as a product decision — instructions, retrieval, memory, and live state, spec'd, governed, and evaluated with the same rigor as the output it produces.
ModuleThe craft of writing input a model will reliably act on — from ChatGPT productivity patterns, through few-shot and chain-of-thought, into the prompts that drive Claude Code and other coding agents. 9 lessons.
ExploreEvery page across all nine modules as one interactive map — 405 pages, 1550 cross-references. Search it, filter by track, click any node to jump in.