Swyx
Swyx compares Flow’s role in hardware engineering with Git and GitHub’s role in software. He focuses on aligning thousands of stakeholders in complex projects. Pipelines for cars and rockets are valuable and difficult to reverse, making coordination especially important. He argues that teams will struggle to return to endlessly renamed spreadsheets after using Flow. He also discloses that he is both a friend of the team and a small investor.
OpenAI Codex and ChatGPT product team · Thibault Sottiaux
Thibault Sottiaux announced that users can build and deploy MCP servers directly through ChatGPT, restrict access, or share them publicly. He said demand for GPT-6.1 Sol was exceptionally high across the API and subscriptions, putting ChatGPT and Codex under heavy load. The team added capacity and expected speeds to improve over the following hours, approaching twice the previous day’s speed. He also asked his dot to paint a portrait of him. When the first result was too cartoonish, he asked it to find recent photographs online and try again, which improved the result. His accompanying video showed him navigating the dot’s computer.
AI educator Peter Yang
Peter Yang reacted positively to Gemini 4. He also argued that Google still needed to become more competitive in coding harnesses and personal agents. He specifically named Antigravity on the coding side. For personal agents, he named Spark. His assessment distinguishes the model itself from the coding and personal-assistant products built around it.
OpenAI Codex product team · Nan Yu
About a week after joining OpenAI, Nan Yu was struck by the team’s strong bias toward action. A casual discussion about a possible feature might turn into a pull request before he returned to his desk. A suggestion on Slack to have lunch the following week could produce a calendar invitation within five minutes. That pace has downsides: work can start immediately and plans can change underneath people. But he considers this energy difficult to teach and very hard to recover once lost. It persists even as the company occupies several large buildings in San Francisco.
Anthropic Claude Code developer Thariq
Thariq described concrete problems in turning a game prototype into a playable product. He wants a character to have a large jump, but after trying two or three variants he still has not found a control scheme he likes. He explicitly says the graphics and feel remain prototype quality and are not ready for players. For a real game, he expects to iterate on that character three or four times and include perhaps eight to ten characters. He would like to continue with a team of two or three people. Animation and design already exist, but specialists could improve them substantially.
Google Labs
Google Labs announced that lessons from Opal are carrying forward into Skills in the Gemini App. Opal originally explored how users customize workflows and also informed Gems made by Labs. Skills let users save custom instructions in chat and automate repetitive tasks. As Skills roll out globally, the Opal experiment will close on November 17, 2026. Google thanked users who built mini-apps and tried Gems, saying their usage helped shape the subsequent capabilities.
Vercel CEO Guillermo Rauch
Guillermo Rauch argues that token-aggregation data is often distorted by promotional usage and unverified zero-data-retention claims. He says Vercel AI Gateway reflects real customer usage, with over 400,000 paying customers and thousands of enterprises on the platform, and no markup. The team rejects free-token promotions from providers with questionable reputations rather than maximizing provider count. He says legal, compliance, privacy, and back-office operations receive as much attention as engineering. In a separate post, he introduces Connect and invites service providers to reach more than 20 million developers and the agents they build. As generating code becomes cheaper, he sees connecting external services as a major obstacle. Connect targets apps and agents, aiming to reduce the developer burden and security liability of distributing static credentials to each project.
Box CEO Aaron Levie
Aaron Levie sees a major opportunity in deploying AI throughout the economy. Enterprises need to migrate legacy systems, reorganize data and permissions, connect software to agents, redesign workflows, arrange human review, and maintain evaluations. New models also require these systems to keep evolving. Unlike conventional software that delivers a tool, agents deliver work output inside a process, changing the complexity of implementation and enablement. He expects service providers to specialize by industry, company size, and internal problem. Traditional systems integrators will need to adapt, new firms will emerge, and he sees a particularly favorable moment for forward-deployed engineers and their firms.
Y Combinator President and CEO Garry Tan
Garry Tan introduced Capy, emphasizing agents that communicate as peers and coordinate themselves rather than only receiving top-down subtasks. This can expose solutions that were not arranged in advance. He also shared a conversation with U.S. Transportation Secretary Sean Duffy at the Startup Industrial Base event. Duffy described controllers using six screens and paper flight charts, while Air Space Intelligence’s SMART tool predicts weather and routing effects so aircraft speeds and arrival order can be adjusted early. An airport may accommodate 15 aircraft in a 15-minute window while airlines schedule 32, so sharing capacity data can help spread arrivals. Duffy gave an example in which a potential 2.5-hour ground stop might take only 45 minutes. Controllers still make the final decisions; Garry’s emphasis is on augmenting their work with software rather than relying on paper and pencil.
FPV Ventures partner Nikunj Kothari
Nikunj Kothari says he has automated most of his work but still personally sources investments and writes outbound emails. He believes technology does little to replace that part of the job. Meeting founders is another activity he retains, with a strong preference for in-person meetings. He still writes every investment pass note himself and calls it the hardest part of the work. He has no executive assistant and currently has no plans to hire one. In another post, he continues seeking projects with outstanding multiplayer experiences and invites developers to send or reply with a demo.
OpenClaw and OpenAI developer Peter Steinberger
Peter Steinberger finds inter-agent communication increasingly distracting in the chat stream. He changed its visibility in the OC harness to a single line that users can expand. This keeps the details available without displaying the full exchange by default. He expects other products to adopt a similar presentation. In another post, he says waiting on CI and GitHub is slowing his work and prompting him to rethink his workflow.
SPC General Partner Aditya Agarwal
Aditya Agarwal repeatedly asks SPC founders, “What is the maximalist version of this company?” He argues that founders today need to articulate their boldest idea and how it could change the future. Starting with a small ambition can make fundraising harder. It may also make exceptional people harder to recruit and meaningful impact harder to achieve. He calls this AI’s paradox: building products becomes easier while the bar for a great startup rises sharply.
Anthropic Claude team
Claude Founder House is coming to SF Tech Week and Stockholm. The San Francisco events take place October 6–8. The Stockholm event is scheduled for October 14. Sessions include talks, workshops, and office hours with the team, plus opportunities to meet other people building with Claude. The events target founders, CEOs, and builders, and the official posts provide registration links for both cities.
How Sam Altman Uses Dots to Take Back His Time
Key takeaway: Sam Altman values Dots for more than additional output. Persistent agents filter information and advance ideas so he can recover time that interruptions used to consume.
At DevDay, OpenAI CEO Sam Altman described his own use of Dots to Dan Shipper. He said, “They’ve never been able to do more, but there’s never been a higher bar either,” referring to developers. As coding capabilities improve, latent demand grows too. He wants more individuals and small teams to create products rather than expecting one model company to meet every need.
The first use is protecting his mornings. Checking his phone after waking used to fill his most creative hours with problems that accumulated overnight. His dot now judges what is truly urgent and handles other matters, giving him space for deep work or time with his children. Context across applications also matters: when Dan changed a flight, his dot discovered a meeting mentioned in Slack but absent from the calendar and flagged the conflict.
The second is continuous product development in spare moments. Sam sends his dot feature ideas that occur to him on walks; it builds five or six versions for feedback. He described leaving a long voice note before sleep and waking to find not just a build but subsequent steps completed. On another occasion, he spent ten or twenty minutes unsuccessfully searching Slack. His dot continued overnight and eventually found the information inside a screenshot.
Third, documents can keep evolving. Sam still writes new ideas himself to think them through, then asks dots to handle follow-up work. He now uses a changing business summary that draws on Slack and adapts to his interests instead of always sharing one fixed dashboard with everyone. He believes collaboration among several people’s dots can also combine ideas that previously existed only loosely in their separate minds.
Fourth, speed and an open platform can lower friction. Ultrafast’s immediate responses have changed his iterative thinking, and he wants faster inference to become more widely available. Letting people bring their subscription to external apps can also reduce the burden of buying many subscriptions. Sam acknowledges that OpenAI has not yet found a mature App Store equivalent, but believes there are far more good ideas outside its buildings than inside. Reducing friction in creation and collaboration is what he expects to get more people building.