There's a pattern across today's stories that nobody's saying out loud: AI adoption isn't stalling because the models aren't good enough. It's stalling because the products built on top of them are. Aaron Levie sees it in enterprise. Madhu Guru sees it in the 2% paying for anything. And Thibault Sottiaux, whose posts last week were about what future models would clean up, is now publicly committing to simplification. The models got ahead of the products. Now the products have to catch up.
01
Box CEO Aaron Levie: AI agents are a coding story, and barely anything else
Aaron Levie posted a detailed observation about where AI agent adoption actually stands. Coding and coding-adjacent tasks have taken off. Everything else in knowledge work is still early. Even within coding, he says it's a wide continuum: a small slice of developers running agents in parallel on multiple projects, and everyone else still working 1:1 with an agent at a time. His read on why other domains are behind: the tasks are harder to define, the failure modes are less obvious, and there's no equivalent of "the code compiles" to tell you whether the agent did it right. ---
Why it matters: If you're pitching AI agents to a finance team, an HR department, or a legal team, you're selling into a market where even the most optimistic observer says adoption is "very early." Your customer isn't behind. The product isn't ready. Those are different problems with different solutions.
Vercel CEO Guillermo Rauch: security is about to get a seat at every engineering table
Rauch argued that security is evolving into two distinct disciplines: verification engineering (proving your code does what you think it does) and capital allocation (deciding which attack surfaces deserve the most attention). He sees this as both a structural challenge and an opportunity for startups, noting that the world's growing paranoia about AI-powered adversaries cuts both ways. Enterprises will scrutinize two-person shops. But the overall demand for security competence is enormous. ---
Why it matters: Yesterday's digest covered AI security as a conference topic. Rauch is saying it's becoming a core engineering function, not a compliance checkbox. If your startup doesn't have someone who owns security architecture, you're accumulating liability that will show up in your Series B due diligence.
A major product is locking in simplification as its only priority
Thibault Sottiaux, who works on future models, posted a clear commitment from his team: from here, they're working only on simplification, efficiency, genuinely new features, or new models. The post got nearly 7,000 likes and close to 1,000 replies, which suggests it landed. His framing: user feedback has been consistent, and sometimes you have to stop investing ahead of the curve and actually respond to it. ---
Why it matters: 6,762 likes on a product roadmap post is not normal engagement. That number means a lot of users recognized their own frustration in that announcement. When a team with that kind of reach commits publicly to making things simpler, the pressure to actually deliver on it is real.
Only 2% of potential users are paying for AI, and even they aren't going deep
Madhu Guru made a sharp observation about AI adoption: the paying user base is tiny, and the depth of usage within that group is shallow. His diagnosis is that most AI products look like airplane cockpits, full of model selectors, connectors, permission toggles, and token dashboards that nobody asked for. His prediction: product teams are finally maturing, and the next 12 months will look materially different. ---
Why it matters: If you're building an AI product, your real competition isn't the other AI products. It's the cognitive load your own interface creates. The teams that strip the cockpit down to the three controls users actually need are the ones that will expand that 2% figure.
Peter Yang got scratched by a squirrel while talking to his AI agents
Product builder Peter Yang was walking in the park, talking to his agents, and a squirrel scratched him. He concluded: "nature is telling me something." No further analysis warranted. Rest up, Peter.