The Model Call
WritingOne API request that takes eight seconds, streams, costs money, and can fail halfway. Timeouts, retries, queues, and state.
Part I — Foundations
What a language model is, and what has to exist around it.
Foundation Models
WritingHow models are trained, what they can take in, and why the same input gives different answers.
Part II — Controlling the Model
How to get the output you want, and how to know when you have it.
Prompt Engineering
WritingWriting the instruction: what to say, what to show, what to ask for.
Context Engineering
WritingEverything else in the window, and how to spend a limited budget.
Evals
WritingMeasuring quality so you can tell whether a change helped.
Part III — Connecting the Model
Giving the model access to your code, your data, and its own past.
Structured Output & Tool Use
WritingGetting back objects your code can act on, and calls it can run.
Multimodal
WritingImages, audio, and documents in. Images and speech out.
Retrieval
WritingFinding the right information and putting it in front of the model.
Memory & State
WritingWhat persists between calls, and where it actually lives.
Part IV — Designing the System
How the pieces fit together, what happens when they break, and how you see it.
Orchestration & Failure
PlannedSplitting work across calls you control, and handling the ones that go wrong.
Designing for Uncertainty
PlannedInterfaces for output that might be wrong: confirmation, editing, undo.
Observability
PlannedTracing requests in production, versioning prompts, rolling out changes.
Part V — Safety
Unreliable output, hostile input, and data you're responsible for.
Guardrails & Prompt Injection
PlannedChecking input and output before either reaches somewhere it shouldn't — and the attacks that arrive as text.
Privacy & Data Governance
PlannedWho sees user data, where it goes, how long it stays.
Part VI — Cost & Speed
What each request costs, how fast it feels, and which model handles it.
Token Economics
PlannedWhere the money goes per request, and how caching changes it.
Latency & Streaming
PlannedTime to first token, perceived speed, and what to show while waiting.
Model Selection & Routing
PlannedChoosing a model per request, with fallbacks when one fails.
Part VII — Judgment
The expensive options, and deciding what you actually need.
Agents
PlannedHanding the loop to the model instead of driving it yourself, and the cost of that decision.
Model Adaptation
PlannedFine-tuning, LoRA, distillation, and why prompting is usually enough.