sf-observations

#sf-observations

@AI_Whisper_X: Greg Isenberg spent 48 hours in San Francisco and wrote down 21 observations 1、Selling data is lucrative: He heard many stories about data brokers selling corporate data to labs, making around $2 million a week. He can't tell whether this is the beginning or a bubble, and is researching which verticals are still open. 2、A soul-searching question: Someone at dinner asked "A new model ships next month — does your company get better or worse?" Half the people couldn't answer. He thinks it's the best question for evaluating AI startups right now. 3、Revenue without retention: The benchmark people talk about now is $1B in revenue within 12 months (the new unicorn standard), but almost nobody talks about retention. He predicts we'll see lots of AI companies with huge revenue and terrible retention. 4、It's clear we're in the middle of a reinvention of the computer. Models are the CPU, the harness is the operating system, skills and plugins are the apps. This framework makes it much easier to spot the gaps. 5、We may be just 12 months away from an agent that watches you work for a week and then takes over your job without any instructions. Screen recording + agent memory + local models — that combination is already possible today. (Creepy) 6、Agent harnesses feel like the new generation of GPT wrappers, but the good kind, especially vertical ones. The best teams he talked to treat models as interchangeable parts. One founder demoed how he routes different tasks to different models: one for aesthetics and visuals, a cheap one for routine grunt work, and another specialized in text. 7、He's never seen this much reverse engineering and cloning. People are decompiling software and rebuilding it with agents. Someone showed him a suite of about 10 apps, each modeled after an original worth over $5B, all open-source clones he built. Everyone is still trying to figure out what it means when software becomes this easy to copy. 8、His guess is value shifts to data. If software can be rebuilt quickly, the thing worth paying for in an old SaaS company is the customer data and workflows accumulated over years. Some SaaS companies are actually stealth data companies. 9、He feels a significant portion of legacy software won't survive into the agent era — especially the ones designed around "making humans click through a bunch of pages." 10、Every company he talked to wants to consolidate internal data in one place and expose it to agents via MCP. Then they look at the data and discover: a total mess. Five versions of the same customer, a pile of dead Slack channels, unsearchable PDFs. Cleaning all that up feels like a massive consulting opportunity! 11、Building-block companies will win: Many founders are working on the primitives agents need — APIs, data layers, and tools. Some companies have almost no UI at all because their primary user is the agent. 12、Labs aren't short on training data, but they can't get real examples of agents failing inside enterprise environments. Whoever has those logs has something valuable. 13、Someone building a large consumer product said the next battleground is memory. The model that remembers the most about you wins — being the smartest doesn't matter as much as people think. 14、Astra is the first model to make people believe AI can do CAD, and people are starting to wonder what "vibe manufacturing" looks like. A robotics founder said: the hardest problem in AI right now is deployment, not research. Their robots work on their own, call a human when stuck, and every call becomes training data. 15、Cybersecurity companies are having their moment — it's genuinely cool to be doing this now, especially post-incident cleanup. Every agent with access to your systems is a new door into your company, and many of those doors are wide open. 16、Dead is worth more than alive: A failed company's five years of Slack messages, emails, and code can be licensed for money. He wouldn't be surprised if founders factor that into their shutdown timing. 17、People are starting to talk about how the "agent internet" gets monetized. It feels very early, but when the most active users of the internet become agents, what kind of startup should you build? 18、Agent insurance is coming. Once agents make expensive mistakes, insurers will demand audits before writing policies. 19、Several people said the people they most want to hire are those with taste who can write evals. Models are getting better at generating ideas — the hard part is judging which ideas are good. 20、The new founder trajectory: Take VC money → use investors' money to make yourself an influencer → company fails → founder becomes a full-time influencer and bootstraps other products with their own money. 21、Traffic is brutally hard to get: Founders are sponsoring buses, going wild on billboards, and hiring famous DJs for parties — all just to be seen. It has that dot-com bubble smell. Attention is the truly scarce resource! The growth lead who went viral with a UGC strategy for Wispr Flow is a perfect example. This is the recipe everyone wants, and maybe 50 people in the world know how to scale this kind of business efficiently in this era. Every time he visits San Francisco, he comes back energized. He feels incredibly lucky to experience such an incredible journey at this golden moment for startups. Much of the time, it almost feels unreal. Great read! Sharing with everyone! (He noted these came from conversations with a group of billionaires, data lab people, AI founders, and executives)

X AI KOLs Timeline ↗ · 2026-10-15 Cached

Greg Isenberg shares 21 observations from 48 hours in San Francisco spent talking with billionaires, AI founders, and executives, covering data brokerage, agent harnesses, software cloning, model memory, cybersecurity, and the scarcity of attention as key AI startup trends.

0 favorites 0 likes

sf-observations

Submit Feedback