Most writing about AI is either theory about how models are trained or news about the latest release. This series is neither. Applied AI is written for engineers who use AI rather than build models: you call model APIs, use AI coding tools, or build products on top of large language models.
Each post takes one practical topic, explains how it actually works at the level a builder needs, and is direct about trade-offs, failure modes, and cost. No machine learning background is assumed. Every post carries a datestamp, because this field moves fast, and names the current tools and versions so you can check them yourself.
Where to start
If you are new to the series, start with Tokens and Context Windows. Tokens are the unit behind every limit, price, and delay in an AI application, and almost every other post builds on them. From there, AI Agents and RAG cover the two patterns most applications are built from.
After that, read in any order. The posts are grouped below by theme, and each one stands on its own, with links to related posts where they connect.
Foundations
How models handle text, which model to use, and how to get reliable output from them.
- Tokens and Context Windows: the unit behind every limit, price, and delay, and how to budget for it.
- Choosing a Model: why leaderboards mislead, how to run your own eval, and how to plan for retirements.
- Reasoning Models: what the thinking dial controls, what you pay for, and when it is worth it.
- Structured Outputs: getting data your code can trust, not text it has to parse.
- Embeddings: how meaning becomes numbers, and what that makes possible beyond search.
Knowledge and context
Getting the right information in front of the model, and making sure answers stay tied to it.
- RAG: answering from your own documents with retrieval-augmented generation.
- Context Engineering: deciding what goes into the context window, and what stays out.
- Hallucinations and Grounding: why models make things up, and the pipeline that keeps answers honest.
- Documents, Images, and Multimodal Input: reading PDFs, scans, screenshots, and charts reliably.
- Memory and Skills: giving agents knowledge that lasts between sessions.
Agents
Models that act: calling tools, connecting to systems, and working on their own.
- AI Agents: the loop behind every agent, and when you need one at all.
- MCP: the Model Context Protocol, and how agents connect to tools and data.
- Multi-Agent Systems: when several agents beat one, and how they coordinate.
- Computer Use: agents that operate software by looking at the screen.
- Voice and Realtime Agents: latency budgets, turn-taking, and interruptions.
- Long-Running and Background Agents: jobs, checkpoints, and safe side effects for work that takes hours.
- AI Coding Assistants in Practice: scoping, checks, and review for coding agents.
Safety and trust
Keeping AI applications from being tricked, misused, or leaking what they should not.
- Prompt Injection: the attack every agent builder needs to understand, and the defenses that work.
- Guardrails: checks on inputs, outputs, and actions, matched to the risk.
- Privacy and Data Governance: where data goes in an AI application, and how to control every copy.
Running in production
Measuring, monitoring, and paying for AI systems once real users depend on them.
- Evals: testing AI behavior so changes are decisions, not guesses.
- Observability: tracing what your AI application actually did, call by call.
- Cost Control: caching, routing, and batching to keep the bill in check.
- Fine-Tuning: when training a model on your data is worth it, and when it is not.
- Open-Weight and Local Models: running models yourself, and when the trade pays off.
Applied AI: AI from the builder's side. Also on this blog: the Explained series and the Architecture series.