Systems on Glass is an AI-automation side project of mine. Every episode is generated by Claude (Anthropic's AI, working through Claude Code in my own video studio): the script, the motion graphics, the edit, and the narration in an AI clone of my voice. I choose the topics, bring the stories from my own work, and review each episode before it goes up. Not affiliated with Anthropic.
Every Monday at nine, hundreds of people open the same link at once, and the model behind it can only answer so many a minute. A small bakery on a hill already knows what to do.
The short version
- The jar is a token bucket. Ten wooden tokens, one per loaf. A new token drops in every second, never more than ten: the jar's size is the burst you allow, the drip is your steady rate.
- Bursts are fine. When the 9 o'clock tram arrives, a full jar serves ten people at once. Then bread goes out at the oven's pace, one a second.
- The bench is a queue. Whoever can wait, waits their turn, and someone in a real hurry goes first. In my design, live chats beat new arrivals.
- The sign is a 429. When even the queue is full, say no politely and say when: "Back at ten, we'll ring the bell". In HTTP, that's 429 Too Many Requests with a
retry-after. - It's real. Anthropic's API uses the token bucket algorithm: capacity refills continuously up to your limit, and going over gets a 429 that says how long to wait.
then: 1 a second, as each token arrives · the queue holds who can wait · past that: 429 + retry-after
my design's peak (illustrative): quota 3,000 calls/min · calls wait up to 25 s · 16% of new sessions get a text instead
The Monday-morning wave is from my design for a car-service intake system, Four Agents Walk Into a Service Bay, which is my own recreation, not Nissan's system. Drag its wave and switch on "Token bucket + queue". The bakery is a fable told over a real simulation: every count in the video is computed, not typed in.
Further reading
- Rate limits (Claude API docs) — a real token bucket, and what a 429 with retry-after means.
- Token bucket (Wikipedia) — the algorithm, with the burst and the rate as two numbers.
- 429 Too Many Requests (MDN) — the status code and its Retry-After header.
Next episode: Let code decide. Previously: Shape ≠ Meaning. All episodes →