What it means
AI tokens are the units of text that a language model reads and writes. A token is roughly 4 characters, or about 3/4 of a word. Every model prices its input and output by the token, so tokens are the meter that decides what running AI actually costs and how much text you can feed a model at once.
Why it matters
Token pricing is what makes large-scale AI work economical. When you can categorize 20,000 keywords for under $20, the low unit cost of a token is the reason. As prices per token fall, jobs that were once too expensive to automate become routine, which is the practical engine behind most of the productivity gains you hear about.
Say a mid-market B2B SaaS company wants to classify 100,000 support tickets by theme. At older per-token prices the job might have cost $2,000; after prices dropped, the same job runs for $150. That moves it from a once-a-quarter project to a weekly habit.
How to use this knowledge
Estimate token cost before you commit to an AI workflow: Count the input plus output tokens per run, then multiply by the model's per-token price. Batch and trim inputs to cut spend on repetitive jobs. Track cost per output (per keyword categorized, per draft written) so you can compare models and re-price a workflow as token rates keep falling.
Growth Memo guidance
“Ai-fficiencies” are incremental improvements from AI. Instead of doing new things, the biggest value add from AI so far is doing things faster and better. — Aifficiency
This week, I categorized almost 20,000 keywords into 8 core topics for a client and paid less than $20 in one hour. AI is NOS for no-code. — Aifficiency
Aifficiency — Kevin's term for the speed and cost gains from AI, which cheap tokens make possible.
Token limit — the cap on how many tokens a model can read and generate in a single request.
Programmatic content — large-scale AI generation where token cost per page decides whether the project is worth running.
AI Overviews (AIOs) — Google's AI answers, produced by processing tokens across many retrieved pages.