Continuing the tokenmaxxing theme.
Google says 75% of its code is written with AI. Meta says 90%. Uber burned a whole year’s worth of tokens in the first four months of the year.
Programmers tokenmaxx. Managers token-flex — about how much their team spends on tokens per month, as if the API invoice were a performance review.
Everything’s better with AI
Need a status report? AI bot. Monitoring? AI bot. Code review? AI bot. Dashboard? AI bot. Every action a programmer takes, every automation, every cron job that used to be twelve lines of bash — must now go through an AI. Because everything’s better with AI. Obviously.
So what do we do with all of this?
The obvious next step
The answer writes itself. For all of this, you need an orchestrator.
Something that sits in the middle and coordinates every bot. Something that stores all your AI conversations — every status report, every review, every dashboard refresh — in one place. Just imagine it. A single pane of glass for the entire firehose of prompts your organization now emits.
And of course the orchestrator itself could use AI — to suggest prompts. Why would it not? At this point not wiring AI into the AI-coordinator would be the only un-AI thing left in the building, and we can’t have that.
If you’re going to do it, do it properly
Look — if we’re going to tokenmaxx, let’s at least do it with distributed systems.
If the goal is to burn a year of tokens in four months and call it a strategy, then let’s not be cheap about the architecture. Sharded conversation storage. A consensus protocol for which bot gets to comment on the PR first. Backpressure for when the prompts come in faster than the model can flatter them. Exactly-once delivery for your status reports, because losing one would be a tragedy.
It’s the same instinct from the first post: the metric got wired to a dashboard, the number must go up, and now we’re building serious infrastructure in service of a contest nobody should be running. But if the contest exists, you might as well win it with good engineering.
So: distributed orchestrator, AI-suggested prompts, all your conversations in one durable, replicated, horizontally-scalable store.
What could possibly go wrong?