Learn
How AI models and agents work
Plain-English explanations of models, agents, cost, and risk, written for the person who has to run it in production.
New here? Start with What happens when an app calls an AI model
Requests and tokens
What an app sends a model, and what comes back.
Models
How models differ, how they fail, and how to pick one.
- Explainer 8 min Why models get things wrong The distinct ways an AI model fails: invented facts, stale knowledge, drift in long documents, agreement bias, and run-to-run variation. What reduces the rate.
- Decision guide 8 min How to choose a model for a task How to pick an AI model for a task: what differs between models, whether to start cheap or strong, and why retirement dates belong in the decision.
- Decision guide 8 min Open-weight and closed models What open weights means, why the licence matters more than the label, the three ways to run an open model, and what self-hosting requires.
Cost
What you are charged for and what changes the total.
- Explainer 8 min What a request to an AI model costs How model pricing works: two separate rates, why a conversation costs more as it goes, what caching and batching change, and the line items that aren't tokens.
- How-to 8 min How to reduce what you spend on AI models A practical order of operations for cutting model spend: measure first, then the two changes that never touch your product, then the ones that trade something away.
Agents and tools
Models that use tools and run in a loop.
- Explainer 6 min What an AI agent is An AI agent is a model in a loop with tools, choosing its own steps. What separates it from a chatbot or a workflow, and where the cost and risk come from.
- Explainer 5 min How tool calling works Tool calling lets a model ask your app to run an operation and use the result. The round trip, who executes what, and what the model knows about a tool.
- Explainer 7 min What MCP is The Model Context Protocol standardises how apps expose tools and data to AI clients. What it defines, what it doesn't change, and what its spec says on security.
Data and privacy
Where your prompts go and who keeps them.
Security
How AI systems get attacked, and what limits the damage.
Infrastructure
The layer your requests pass through.