(dictated using voice) These are some random notes about stuff stood out from three podcasts I listened to: 1. Peter Steinberger with YCombinator 2. Peter Steinberger with Lex Friedman 3. Boris Cherny with Lenny Peter Steinberger is the creator of OpenClaw and Boris Cherny is the head of engineering at Claude Code. # Peter's thoughts 1. Peter mentions a really insightful moment where AI implemented a feature that he never put in there. He was using an early version of Open Claw, and before a trip to Marrakech, he set up a WhatsApp integration so that communicate with his AI assistant when the internet was patchy. At some point during the trip, he was hooked and felt like he wanted to send a voice message. He realized he hadn't implemented this feature, but decided to send it anyway. It basically responded to his request that he had asked by a voice. Mind blown. He asked, how the hell did you figure this out. To which the agent replied that it didn't see any message but looked at the head headers to figure out it was a voice message and then looked around for keys to Speech to text the audio via chat GPT's API, and then responded to it. The security implications might be concerning, but it's amazing when something you build has a feature that you never planned for. 2. Peter mentioned that he's not a fan of writing a spec and one-shotting a solution via AI. He says that was the waterfall style of engineering. And also that it doesn't put care into the piece of software. And he doesn't believe in that. He likes to be part of the conversation with the AI building it. 3. Peter says that he can put more attention into the fun parts of programming now and let AI do the tedious or the boring parts of programming. # Boris's thoughts 1. Boris mentions that his workflow involves one third of the time in the terminal, one third in the web UI and one third on his iOS app, and that he always has four to five agents running, doing different things. It seems a lot of the productivity boost comes from the parallelism that agents afford. 2. Boris mentions that 100% of his code is written by AI, but he also reviews 100% of his code. He says he puts out about 20 to 40 pull requests per day which is a crazy number. As someone managing a team, he also uses AI to ping people to fill up statuses and things like that. 3. Boris did mention that some people on his team just like to handwrite C++ because they want to, and there is room for that . 4. Boris said that at some point he was really into the beauty of programming, but eventually he was always interested in getting things done with programming. It felt like his seg-way into TypeScript and types and functional programming, which he even wrote a book about, was a rabbit hole and a distraction from the final utility of programming, which was building things people need. This clearly indicates the type of person he is, and his approach to its programming work 5. Boris said, now he listens a lot more to user feedback and pushes fixes or solutions within five minutes. 6. It was also mentioned that for Claude and Anthropic, the trajectory was always improving the text models, then doing tool use and then doing computer use, which is clicking buttons on a UI, which Claude can do right now. 7. He did say coding is largely solved for the kinds of problems that they face, but it couldn't help make me think that the kinds of problems that they face are really end product things. For example, the Claude desktop app is built using electron, which makes a lot of things easier, but it's also bloated. Why didn't they just do it natively?