Guides

Short, visual explainers for topics that don't fit a single quiz question — written for how PMs actually decide (tradeoffs, not hype).

• Model selection attributes

  Context, latency, throughput, pricing signals, capabilities, and compliance — with scenarios and archetypes.
• Prompt & policy design

  Instruction precedence, output contracts, and when strict schemas beat prose for shipped AI.
• RAG vs fine-tuning vs prompt-only

  Decision flow for factual grounding vs behavior tuning, plus comparative tradeoff charts.
• Retrieval architecture cheat sheet

  Chunk-to-answer pipeline, retrieval patterns, and complexity vs quality visuals.
• Evaluation strategy for AI features

  Layered eval maturity, judge vs automation radar, and signal vs latency curves.
• AI pricing & unit economics

  Token COGS stacks, margin sensitivity, and finance checkpoints for AI features.
• Caching & request shaping

  Semantic and prompt cache tiers, dedup, batching — TTFT and COGS levers with trust caveats.
• Embeddings vs keyword vs hybrid

  Lexical vs semantic retrieval, decision flow, and fit-by-query-shape charts.
• Latency budget & streaming UX

  TTFT timeline, streaming cadence, and UX tactics for perceived responsiveness.
• Agent boundaries & human-in-the-loop

  Risk tiers, approval lanes, and HITL patterns for tool-using agents.
• Tool & integration design

  Schema discipline, normalized errors, idempotency, and timeouts for LLM-side effects.
• Hallucination & grounding playbook

  Citations, hedging, abstention, and escalation — UX plus backend patterns.
• Security basics for AI PMs

  Threat sketch across prompts, tools, and data — plus controls and owners.
• AI governance & data lifecycle

  Collection through retention, subprocessors, and artifacts enterprise procurement expects.
• Observability for shipped AI

  Debug loops, triage signals, and trace sketches for production AI.
• Experimentation & rollout

  Shadow, canary, and A/B ladders — signals, exit criteria, and rollback discipline.
• Multi-tenant AI SaaS

  Isolation for prompts, retrieval, billing, and telemetry across tenants.
• Vendor & routing strategy

  Multi-provider failover, tiered models, and integration vs ops burden.
• Working with ML teams

  Ownership handoffs across the build lifecycle, and where PM/ML collaboration breaks down.
• Synthetic data & augmentation

  When synthetic data helps vs hurts, core patterns, and the model-collapse risk.
• Model monitoring & drift

  The detect-diagnose-respond loop for shipped models, and four kinds of drift to recognize.