Madhu Guru, Senior Director of AI at Meta
Madhu Guru, Senior Director of AI at Meta, argues that product experience is now a major constraint on AI adoption, beyond model capability alone. He says only about 2% of people pay for AI today, and even their usage remains shallow. He attributes this to AI products that feel like airplane cockpits, with too many controls for connectors, permissions, model selection, and token usage. Guru says product teams are beginning to simplify these experiences. He expects AI products to look materially different and become easier to use over the next 12 months.
Aaron Levie, CEO of Box
Aaron Levie says AI agent adoption is sharply uneven: coding and adjacent tasks have taken off, while other knowledge work remains early. Even in software development, adoption varies widely: only a small share of developers run background agents on projects in parallel, while many still work one-on-one with an agent. He argues that bringing agents into knowledge work usually requires rebuilding workflows, not merely adding a chat tool. Companies also need to reconnect data, establish accountability for decisions, and adapt governance, compliance, and security practices. These requirements help explain why agent deployment outside coding is still at an early stage. Levie says the current picture is far from mature, and adoption beyond the few categories already seeing traction could grow 100-fold from today's level.
Thibault Sottiaux, Codex and ChatGPT lead at OpenAI
Thibault Sottiaux shared how he used dot to clear his inbox: it was the first time he had deliberately set inbox zero as a goal, and reaching it felt great. He says dot deleted email categories he did not need to keep, validating them in batches, then identified work-related labels and applied them across messages. For emails that needed replies, dot looked for relevant context in the background and walked him through them one by one. In another post, he said the team's current focus is simplification, greater efficiency for more usage, and breakthrough features or new models. He says user feedback is clear: people want products to be simpler, and the team is working on it.
Nikunj Kothari, Partner at FPV Ventures
Nikunj Kothari sees the rise of personal agents as a new test of competition between platforms. He observes that Amazon is protecting its advertising business and brand, while the smaller DoorDash is beginning to open access through CLIs and allow orders by text. In his view, whether platforms expose their capabilities will shape whether personal agents can complete transactions for users. In another post, he cautions against tokenmaxxing for its own sake; brief experimentation can help reveal a useful purpose, but simply consuming more tokens is not valuable. Together, the posts point to a practical test for agents: whether they can complete real tasks and whether platforms let them reach the capabilities they need.
Ryo Lu, who has designed for Cursor, Notion, and Stripe and built startups
Ryo Lu argues that products tend to converge when teams rely too heavily on market metrics and competitors. AI can absorb a vast range of examples quickly, he says, but seeing more references is not the same as developing a point of view. Personal experience, the details someone cares about, and repeated experimentation are what produce a perspective others can feel. He believes software can express ideas about life just as music, painting, or film can, through choices about attention, what users can change, and whether a task feels like paperwork or play. When teams keep adding features based only on markets, workflows, and rivals, products become more alike and harder to distinguish. His conclusion is that faster tools make deciding what deserves to exist more important; speed cannot make that judgment for people.
Aditya Agarwal, General Partner at SPC
Aditya Agarwal argues that great wealth does not necessarily buy unlimited access to leading doctors, lawyers, or creative professionals. He gives the example that even a wealthy cancer patient may still get only 20 to 30 minutes with a top cancer doctor. He describes AI's difference as both widening access to expertise and making the amount of time available to each person nearly unlimited. He calls this the practical effect of intelligence becoming too cheap to meter. His point is that AI's value is not only greater intelligence, but also extending professional time to more people.
Guillermo Rauch, CEO of Vercel
Guillermo Rauch expects security to become a larger function inside software companies. He describes security as verification engineering, such as checking whether code is likely to be memory-safe, as well as capital allocation: deciding which attack surfaces deserve the most tokens. For startups, more sophisticated AI adversaries and rising public concern make it harder for a two-person team to earn trust. At the same time, he says the poor state of global cybersecurity and growing reliance on centralized systems leave substantial room for small teams to innovate. Rauch says he is excited to invest in those opportunities.