In partnership with
News of the day
1. Coinbase cut its AI spend by nearly half while growing token usage by leveraging multi-model routing, aggressive caching, and default low-cost models. → Read more
2. Vercel acquires Better Auth to provide AI agents with unique identities and scoped permissions, enhancing security and control. → Read more
3. Ant Group's Robbyant open-sources LingBot-Vision, a 1B parameter vision model excelling in dense spatial perception tasks, outperforming larger models. → Read more
4. AI agents are set to revolutionize enterprise observability, taking over root cause analysis from humans within two years, boosting efficiency and proactive problem-solving. → Read more
Our take
Hi Dotikers!
A few days ago, Arthur Mensch reminded us that proprietary models give labs a comfortable window into their customers' data. Coinbase and Vercel have just delivered a very concrete answer to that concern: they simply stopped picking a lab.
Both companies are making the same architectural bet. Rather than building their stack around a single provider, they route work across multiple models. Vercel pushes over a trillion tokens a day through its gateway and has declared exclusive partnerships obsolete. Coinbase, for its part, cut its AI bill nearly in half while usage kept climbing, without imposing a single quota on its engineers.
The recipe comes down to three levers. First, an internal gateway that defaults to cheap open models, GLM 5.2 and Kimi 2.7, hosted in-house so no data ever leaves the company. Second, task-based routing: a frontier model for complex planning, a budget model for execution. Third, aggressive caching that pushed the reuse rate from 5 to 60 percent.
With its 1,200 full-time agents, Coinbase reckons humans no longer have any business choosing the model. A trillion tokens a day, and not an ounce of loyalty.
The message is clear, and it proves Mensch right in a way he probably didn't anticipate: the advantage is shifting from the model to the infrastructure that decides which one to use. Trying to crown the best model is a losing game, since it won't stay the best for long. That said, trading vendor lock-in for orchestration complexity is anything but free: you don't eliminate the hard problem, you relocate it.
Alex.
Scale AI support on AWS, see how July 9
Customer expectations keep rising. Support budgets don't. On July 9, Fin and AWS are hosting a live executive session on how leading enterprises close that gap: scaling AI-powered support while simplifying how they buy it.
You'll see how to resolve an average 76% of conversations with Fin on AWS enterprise-grade infrastructure, procure through AWS Marketplace to put committed cloud spend to work, and turn the Fin and AWS collaboration into lower support costs. Register for the live session to see how.
Meme of the day
