At an Onam festival last night, I was talking w/ a couple engineering colleagues. We started w/ the old story of a small god who asked for 3 paces of land & ended up measuring 3 worlds, then somehow ended on 2 numbers.
Detour was my fault. I asked what their token spend looked like. I love Pro & Max plans, but was curious how much more intel their employers put within reach. One works at a large Bay Area tech company, w/ roughly $300 a day available across all models. Other works at a major payments company, w/ about $1,500 a month across models. Both called these internal norms, not published policy. Numbers like these age quickly.
At first $300/day vs $1,500/month seems like the story. Then calendar starts to matter more. A daily allowance expires at midnight. Use it or lose it. A monthly allowance stretches across four weeks. Week one feels loose. By week four, a long run has a shadow price. An engineer can look at work worth running, check what remains & decide it can wait until the first.
Call 2026 the year companies discovered they needed a currency for thinking. Models came first, then coding agents, research agents & workflows running long after a prompt. Then bills showed up, and finance had to set limits.
Tokens work b/c they can be counted. Dollars work b/c accounting systems already understand them. Neither says much about value. A thousand tokens might finish useful work, repeat context already seen or explain why an agent failed. Meter records consumption. Meaning comes later.
Conversation moved from money to memory. Asked how payments company handles context across teams, answer was much less settled. Different groups are trying different memory setups, some Slack-native delegation in mix.
That order matters. Companies are learning how to meter machine intel before figuring out how its residue becomes institutional knowledge. One team spends, learns & moves on. Another may start same journey again b/c context never traveled w/ work.
Usage came first, then bill. Now come controls. Controls rarely stay accounting devices for long. They become habits. Daily allowance creates one tempo, monthly ceiling another. Engineers begin to feel shape of budget w/o checking it.
Every org putting a meter around intel is making a small design choice about how work happens. Invoice may still say dollars, but underneath sits a theory of time, scarcity & value. That theory finds its way into which questions get asked, which jobs wait, which tasks go to agents & which ones earn the spend.
Maybe that is why conversation began w/ Onam. An old story starts w/ something small enough to seem easily bounded, then suddenly operating at scale of worlds. A unit of intel can look just as unassuming, right up until work starts reorganizing around it.