Boris Cherny

“#20 𝕏 Boris Cherny says he has Claude direct-message people on Slack, monitor their responses, and create a chart with the results.”

#20 𝕏 Boris Cherny says he has Claude direct-message people on Slack, monitor their responses, and create a chart with the results.

“He recommended running `/usage` for a token-by-token breakdown, noting that excessive usage is most often caused by a runaway loop or inefficient skill.”

#10 𝕏 Boris Cherny said there were no changes on their end and usage is monitored carefully. He recommended running `/usage` for a token-by-token breakdown, noting that excessive usage is most often caused by a runaway loop or inefficient skill.

“Boris Cherny announced Claude mods, which let users customize how Claude works and looks through prompting. Mods can also be shared as plugins for others to try.”

#2 𝕏 Boris Cherny announced Claude mods, which let users customize how Claude works and looks through prompting. Mods can also be shared as plugins for others to try.

“Boris Cherny shared a blog post explaining how the claude.ai login experience and Desktop app became faster in the last few weeks, with techniques for engineers speeding up their own apps.”

#8 𝕏 Boris Cherny shared a blog post explaining how the claude.ai login experience and Desktop app became faster in the last few weeks, with techniques for engineers speeding up their own apps. #9 𝕏 Boris Cherny commented that Claude appears to model tricky state machines or race-prone code, identify counter-examples as suspected bugs, reproduce them, and fix the code. He clarified that this targets the hardest parts rather than formally verifying the entire codebase.

“#5 𝕏 Boris Cherny demonstrated using Opus 5.5 to formally verify the Claude Agent SDK with Lean, saying a couple short prompts produced 16 PRs addressing bugs and race conditions.”

GenAI PM Daily September 23, 2026 GenAI PM Daily 🎧 Listen to this brief 3 min listen Today's top 20 insights for PM Builders, ranked by relevance from Blogs, X, and YouTube. OpenAI News announces GPT-6 Sol and Luna #1 📝 OpenAI News Introducing GPT-6 Sol and Luna - Introduces two new GPT-6 variants, Sol and Luna, highlighting their capabilities and intended use cases. The announcement describes distinctions between the models and how organizations can adopt them. Also covered by: @Cognition , @OpenAI , @Sam Altman , @There's An AI For That #2 𝕏 Claude Opus 5.5 is available today. Also covered by: @Thariq , @Mike Krieger , @Anthropic , @Aravind Srinivas , @Cursor , @v0 , @Claude , @There's An AI For That #3 📝 OpenAI News Better prompt caching for GPT-6 - Announces improvements to prompt caching for GPT-6 to reduce latency and cost for repeat prompts, improving developer and product performance. #4 𝕏 Aravind Srinivas shared research on a post-training approach that teaches the Perplexity Computer agent from real user sessions by combining rejection sampling fine-tuning with hint-guided self-distillation. #5 𝕏 Boris Cherny demonstrated using Opus 5.5 to formally verify the Claude Agent SDK with Lean, saying a couple short prompts produced 16 PRs addressing bugs and race conditions.

“Anthropic announces Claude Mods for Claude Code #1 𝕏 Boris Cherny announced that Anthropic’s Claude Mods are landing now, highlighting a Tetris-in-Claude mod.”

GenAI PM Daily September 15, 2026 GenAI PM Daily 🎧 Listen to this brief 3 min listen Today's top 20 insights for PM Builders, ranked by relevance from X, YouTube, and LinkedIn. Anthropic announces Claude Mods for Claude Code #1 𝕏 Boris Cherny announced that Anthropic’s Claude Mods are landing now, highlighting a Tetris-in-Claude mod. The linked anthropics/claude-code issue includes the latest community update, technical details, and additional demos. #2 𝕏 Google DeepMind shared a Google Blog link about WeatherNext 3, noting that advance visibility into rapidly changing weather helps teams balance power grids and match clean energy to everyday consumer demand. Also covered by: @Google DeepMind #3 𝕏 Google AI shared Dreambeans from Google Labs, an opt-in experience that connects information across Gmail, Calendar, Search, Gemini, and Google Photos face grouping to generate customized illustrated stories and relevant next steps. Daily in-app notifications surface new stories, while privacy filters and thumbs-down feedback give users control over personalization. #4 𝕏 NVIDIA announced new SDKs, NIM microservices, playbooks and blueprints for its AI for Media collection at IBC2026. The tools can help verify potentially AI-generated footage, create smoother slow-motion sports replays and translate programming with lip-synced dubbing. #5 𝕏 Aravind Srinivas announced expanded work with NVIDIA to bring fully local, unmetered AI to Microsoft Windows PCs with RTX GPUs using NVIDIA hardware and a Perplexity harness. #6 𝕏 DeepLearning.AI recapped GPT-6 Astra, reporting that it tops the ARC-AGI-3 leaderboard while cutting token costs and ties Claude Fable 5.1 on Artificial Analysis’ Intelligence Index. For builders, it highlights asynchronous tool calls, retained reasoning memory across API calls, and efficient context management for scaling agents. #7 𝕏 NVIDIA AI shared an engineering deep dive on tuning caching, memory, parallelism, decoding and more for the Nemotron 3 Ultra NIM. On four B200 GPUs, these optimizations supported up to 2.5x more concurrent users while maintaining 50 TPS/user. #8 ▶️ Building a Software Factory that actually works (Full Course) Greg Isenberg The software factory uses five or six markdown files to run an agent workflow of isolate, build, prove, and ship: fresh Git work trees from origin main, service-layer code structure, before-and-after evidence, and Greptile PR review loops to a five-out-of-five score. Each feature starts in a fresh Git work tree branched from origin main ; separate agents work in separate copies of the app, and completed changes are merged back after the work tree is finished. The code structure skill directs agents to use a service layer architecture, while evidence-driven testing records a broken before state and a working after state as video or screenshots in the pull request. For a performance change, page-load time was reduced from 815 milliseconds to about 61 milliseconds; Greptile feedback below five out of five sends the agent back through build, prove, and ship until the PR reaches five out of five. #9 𝕏 LlamaIndex 🦙 shared a two-pass, just-in-time OCR workflow for ad-hoc data rooms: LiteParse provides a layout-aware first pass across 50+ formats—reportedly processing a full data room in 32 seconds—while LlamaParse handles only pages needing deeper extraction. The approach aims to avoid the cost and latency of parsing every page upfront. #10 𝕏 Radio lets agents in Claude Code, Codex, or Cursor communicate in real time across machines by pasting one link. Agents get their own channel, where humans can answer questions or redirect work. #11 𝕏 Harrison Chase said Slack is where agents are doing work and that Managed Deep Agents has a tight Slack integration. He shared a tutorial about the integration. #12 𝕏 Guillermo Rauch said agents are only as effective as their proof-checkers, compilers, type systems, and linters, highlighting shadcn/lint as a way to keep agents aligned with design-system rules. He characterized verifiers and skills as the new “frameworks.” #13 𝕏 Santiago said the Cline desktop app is available, open-source, and usable with open-weight models. He argued that agentic coding tools are shifting from terminal interfaces to desktop apps with better affordances. #14 in Peter Yang shared a video showing how to set up the 11 Grok Bots he currently uses: 3 for orchestration, long-term planning, and bot maintenance; 3 for growing his business; and 5 for saving time and money and improving his life. The video also shows the bots talking to one another. #15 𝕏 claire vo 🖤 shared a link to a clairevo.com page about building a cursor-following portrait with Flora and Codex. #16 𝕏 Thariq shared a conversation with Sid and Robert about building Claude Code, covering how much things have changed, the challenge of keeping up with model capabilities, and what they miss about software engineering before AI. #17 in Claire Vo recapped a How I AI episode with John Bai and Peng Zheng about the GrokBot design team’s use of AI agents. Demonstrations include a hyper-personal homepage maintained by Grok Bot, Figma Bro handling design production, and voice memos turned into production code. #18 in Greg Isenberg commented that agent harnesses loop models, provide tool and file access, manage memory, enforce rules, and retain job knowledge so builders can switch models and improve workflows through human corrections. He argues they could support outcome-based pricing across many of 800+ occupations, with OpenAI’s Agents API, Claude’s managed agents, Vercel’s eve, and LangChain making them easier to build. #19 𝕏 Peter Yang recapped Brex CEO Pedro’s view that good AI products pair an agentic loop with tools exposed to a model. As an example, Brex built Jim, an AI recruiter that uses Greenhouse, LinkedIn, and other software to automatically deliver qualified candidates. #20 𝕏 Sam Altman commented that American companies developing increasingly capable AI must act responsibly and that every frontier lab should provide confidence in its conduct. He discussed federal frameworks, Responsible Scaling Policies, safety cases for frontier reinforcement-learning runs, pacing progress, alignment and monitoring, and international coordination. Found this valuable? Share it with another PM - they can subscribe at genaipm.com Unsubscribe • Switch to Weekly

“Boris Cherny requested feedback on an early concept for making Anthropic’s Claude Code more extensible, describing it as “a little crazy, and very exciting.””

Boris Cherny requested feedback on an early concept for making Anthropic’s Claude Code more extensible, describing it as “a little crazy, and very exciting.” More details are available in issue 91870 of the anthropics/claude-code GitHub repository.

“Boris Cherny said his unspecified group uses the same exact Fable and is working to reduce cybersecurity refusals.”

GenAI PM Daily August 25, 2026 GenAI PM Daily 🎧 Listen to this brief 3 min listen Today's top 19 insights for PM Builders, ranked by relevance from Blogs, YouTube, and LinkedIn. GPT-5.6 in Kiro advances developer price-performance #1 📝 OpenAI News Advancing price-performance for developers with GPT‑5.6 in Kiro to improve price-performance for developers, enabling more cost-effective and performant model access for applications. #16 𝕏 Boris Cherny said his unspecified group uses the same exact Fable and is working to reduce cybersecurity refusals.

“He stopped writing code by hand in November 2025 but still codes daily with agents, and said Opus 4.8 was the first model that felt better than him at coding, while stressing that Claude’s code still has bugs and inefficiencies.”

#7 𝕏 Boris Cherny commented that models are advancing through three stages—coding, coding-adjacent engineering, and most computer-based tasks—and judged Claude to have reached stage 1, part of stage 2, and early signs of stage 3 for his work. He stopped writing code by hand in November 2025 but still codes daily with agents, and said Opus 4.8 was the first model that felt better than him at coding, while stressing that Claude’s code still has bugs and inefficiencies. #10 𝕏 Boris Cherny said Opus excels at long-running work and coding but has quirks, with verbosity a major concern and broader fixes a team priority. As a temporary mitigation, users can run `claude /config outputStyle=concise`.

“Boris Cherny recommended running `/usage` for a detailed breakdown, noting that the issue is most often caused by extreme parallelism, a runaway loop, or a very inefficient skill/plugin.”

#14 𝕏 Boris Cherny recommended running `/usage` for a detailed breakdown, noting that the issue is most often caused by extreme parallelism, a runaway loop, or a very inefficient skill/plugin. #16 𝕏 Boris Cherny announced that an unnamed item, which had been worked on with customers for a while, was expected to arrive in the fall. He said Mythos-class models require additional safety measures, enterprises need to meet their own privacy and compliance rules, and customers can own and control their data while Anthropic retains none of it.