Gergely Orosz · 1 min read · AI & Product
The Pulse: ‘Tokenmaxxing’ as a weird new trend
Read the original on newsletter.pragmaticengineer.com →
Tokenmaxxing is the weird new trend where developers at Meta, Microsoft, Salesforce and other large firms burn tokens (and money) to inflate AI usage metrics and hit internal targets. The piece also highlights tight AI-budget dynamics, noting Anthropic ending enterprise subsidies and Uber burning through its 2026 AI token budget in three months, with per-engineer AI budgets likely to spread soon. It rounds out with related shifts in open-source, AI threat discourse, and tooling from Cal.com, Calendly, and Vercel as part of the broader AI ecosystem.
Key Takeaways
- Tokenmaxxing is the practice of burning tokens (and money) to inflate AI usage metrics to meet targets.
- Enterprise AI budgets are tightening and evolving, with Anthropic halting subsidies and Uber exhausting its 2026 AI token budget in just 3 months; per-engineer AI budgets are expected to roll out across more companies.
- Open-source vs AI-threat dynamics are shaping business decisions, as seen with Cal.com going closed source and Calendly-like moves, alongside tooling shifts like Vercel open-sourcing agent factories and governance signals in the Linux kernel.
- These trends illustrate a broader AI ecosystem shift in Big Tech and startups, spanning budgeting, usage incentives, and governance/ tooling.
Excerpt
The Pulse is a series covering events, insights, and trends within Big Tech and startups. Notice an interesting event or trend? Hit reply and share it with me. Today, we cover: Tokenmaxxing: weird new trend. At Meta, Microsoft, Salesforce and other large companies, devs are purposefully burning tokens (and money!) to inflate their AI usage and hit AI usage metrics which they treat as targets. Are coding AI-agent subsidies doomed? At the same time as Anthropic stopped subsidizing enterprise plans, Uber managed to burn through its entire 2026 AI token budget in just 3 months. I expect per-engineer AI budgets…