AI Builders Digest

Anthropic merges Claude Cowork and chat into one Claude, adds Docs and Slides

Cowork and chat are now a single Claude that figures out what a task needs. You can ask a quick question or hand off a report due at noon, and it keeps working after you close your laptop. Claude Docs and Claude Slides are new today, and Claude Design now works inside conversations, all three in beta on paid plans, with direct editing, presenting, and PowerPoint or PDF export. Tasks can be scheduled (a report every Monday, say), and by default Claude asks before acting unless you let it run; rollout hits Pro and Max over the coming weeks, then Team and Free, while Enterprise admins get at least 30 days' notice.

• #products
• #agents

Levie: agent adoption is bimodal, coding took off and everything else is early

Aaron Levie expects 100X more agent adoption than we've seen so far, because outside coding it has barely started. Even within coding, he sees a small share of developers running background agents in parallel while most still work with agents 1:1. The rest of knowledge work lags because this isn't a chat rollout: workflows have to be rebuilt, data wired up in new ways, and accountability, liability, governance, compliance, and security all rethought.

• #agents
• #enterprise

Sam Altman: treating AI as a religious authority is a real safety issue

Sam Altman says he is very uncomfortable with people ascribing religious force to AI models or surrendering human judgment to them, and he calls it a real safety issue. The risk he names here is human deference to models, not model behavior.

• #safety

VAST Data's Renen Hallak: AI storage demand keeps blowing past forecasts

Demand is outrunning every plan. One smaller AI cloud forecast 500 petabytes over three years, then came back a quarter later asking for an extra two exabytes, and Hallak expects requests in the double-digit exabytes next. The limits are physical (land, power, chips, fabs), and some AI clouds are sold out for the next year and a half. His bigger bet is that enterprises will own their intelligence as fine-tuned weights: VAST's new confidential computing launch with NVIDIA lets them run a model builder's encrypted weights on-prem, so neither side sees the other's secrets. "Sometimes it scares me as to how much demand is there."

• #infrastructure
• #hardware
• #enterprise

Madhu Guru: AI adoption is a product problem, and products look like cockpits

Madhu Guru argues that adoption is mostly stuck on product: even the 2% who pay for AI use it shallowly. He blames interfaces that look like airplane cockpits with 100 levers: connectors, permissions, model selection, token usage. He expects AI products to look materially different within 12 months as product teams make them easier to use.

• #products
• #adoption

Thibault Sottiaux says his team is locking in on simplification

Responding to clear feedback that users want things simpler, Thibault Sottiaux says the only work underway now is simplification, efficiency that buys more usage, groundbreaking features, or new models. He also hit inbox zero for the first time in his life by making it a goal for his "dot". It batch-deleted unneeded email categories, labeled mail by type of work, and walked him through replies while pulling in context in the background.

• #products
• #agents

Guillermo Rauch: security becomes verification engineering plus token allocation

Guillermo Rauch predicts security will become a bigger function at software companies. He defines it as verification engineering ("my code is probably memory-safe") plus capital allocation ("what surface should I throw most tokens at?"). For startups it cuts both ways: as AI adversaries grow more sophisticated, a 2-person-and-a-dog company is harder to trust, but the dire state of global cybersecurity leaves vast areas that small teams can disrupt, and he is excited to invest there. His bolder take: AI will make everything free, including itself, and free energy will be the last domino to fall.

• #security
• #startups

Ryo Lu: AI is accelerating software's convergence to the mean

Ryo Lu argues that when everyone copies what works and optimizes for measurable rewards, products converge into polished, interchangeable sameness. AI makes that loop frictionless and produces "a tidal wave of mid." Access to every reference isn't the same as having a perspective, so he wants builders to treat software as an artistic medium that expresses beliefs about how life should feel. His closing question: "what is the point of everyone being able to make something, if we all end up making the same thing?"

• #design
• #products

Aditya Agarwal: AI gives everyone unlimited time with the best expert

Aditya Agarwal pushes back on the idea that wealth buys unlimited access to top experts: the world's leading cancer doctor will still see you for 20-30 minutes no matter how rich you are. In his view, AI does two things at once. It democratizes access and makes the available time unlimited for every individual, which he calls the practical effect of intelligence too cheap to meter.

• #healthcare
• #products

Nikunj Kothari: personal agents pose an innovator's dilemma for Amazon

Nikunj Kothari sees personal agents as inevitable, and that creates an innovator's dilemma for incumbents. Amazon is protecting its ads cash cow and its brand, while DoorDash, though smaller, is slowly opening the gates with CLIs and ordering by text. He also argues that tokenmaxxing for its own sake isn't useful: do it for a while to explore the edges and maybe find the right purpose.

• #agents
• #commerce