FundamentalsTokenization Explained: Why Your Bill Is Not About Words
Tokenizers decide what your prompt costs, how long your context really is, and why models miscount letters. Here is how subword tokenization works in practice.
ReadNotes on models, agents, and what inference costs
Practical writing for developers building with large language models — how they work, how to pick one, and how to keep the bill predictable.
Fundamentals
AI AgentsTool Calling: The Feature That Turns a Model Into an Agent
Tool calling is how a model asks your code to do something. Understanding the handshake — and designing good tools — is most of what makes an agent reliable.
Read
GuidesWhat "OpenAI-Compatible" Actually Means (And Where It Breaks)
Most AI tools speak one API shape. Understanding what compatibility covers — and the four places it usually leaks — saves hours of debugging a base URL swap.
Read
Cost & PricingWhat AI Actually Costs Per Developer Per Month
There is no single number, but there is a reliable way to derive yours. Work through the token arithmetic behind light, typical and heavy agentic usage.
Read
FundamentalsWhat Is a Parameter Count, and Does It Matter?
Parameter counts are the most quoted and least understood model spec. What they measure, why total and active differ, and when the number predicts anything useful.
Read
Cost & PricingWhat It Costs to Run a Fleet of Agents
Fan-out is the dominant cost term once you run agents in parallel. Here is the arithmetic for coordinator and subagent spend, plus what actually reduces it.
Read
Cost & PricingWhen a Cheap Model Is Enough
Most production LLM traffic does not need a frontier model. A practical framework for deciding which tasks can drop a tier without anyone noticing.
Read
AI AgentsWhen Not to Use an Agent
Most tasks handed to agents are better served by a fixed workflow or a single model call. A decision rule, the failure modes, and what to build instead.
Read
Cost & PricingWhy Agent Costs Are Unpredictable, and What to Do About It
Agent spend follows a heavy-tailed distribution, so the average is a poor planning number. Here is how to budget from percentiles and bound the tail instead.
ReadFundamentalsWhy LLMs Hallucinate, and What Actually Reduces It
Hallucination is not a bug that will be patched out. It follows from the training objective and from how we grade models. Here is the mechanism and the mitigations that work.
Read
FundamentalsWhy the Same Prompt Gives You a Different Answer
Temperature zero is not determinism, and a seed is only best effort. Here is where LLM nondeterminism really comes from and how to build around it.
ReadShowing 721–731 of 731 articles