Been building this with @MrRazzi17. AI made social engineering cheap, fast, and scarily good. Your employees get phished. Your AI agents get phished. We test both continuously. @cimentoai is live. 🚀

Terrible time to be working for controlling micromanagers

Welcome to @Cloudflare's #BirthdayWeek. As is our tradition, @zatlyn and I penned our annual Founders' Letter to kick the week off: blog.cloudflare.com/cloudfla…

I made a techno remix out of a song I recorded with an acoustic guitar in 2011, and asked Opus 5.5 for a music video with a JavaScript-forward art style. (It’s DDR optimized. Anyone else here plays DDR and is interested in steps?) 🧵

Agent maxxing.

After 15+ years at Cloudflare, I am taking some time off to recharge and look for my next adventure. Please apply if you'd be a great fit to lead an amazing team of Trust & Safety professionals who are helping to make the Internet better every day. linkedin.com/jobs/view/44685…

Same thing as we always do: make the best tech available for everyone for free.

Got similar answer from an Anthropic researcher at their first Build Day back in 2024

Qwen-Image-2.1 is out. One thing @wu_chenfei shared on Rednote that didn’t make the official blog: Qwen also released 2 9B prompt rewriters, 1 for generation and 1 for editing. That’s new for Qwen Image, and a big deal if you want to see or tweak how prompts get rewritten.

13 years ago I thought "I should write a blog post about the economics of open source" but the outcome was too depressing. After noodling on it for a decade, I have finally finished the post. I wasn't expecting to end up at "Microsoft should send corporations a huge bill" but hey, a lot has changed. seldo.com/posts/nobody-pays-…

As COO of a fast-growing company, I spend a lot of time noticing what stopped scaling. Last month it was Slack. Put a few hundred people in a channel and Slack stops being a conversation and starts being noise. So here is how Warp communicates, in 4 pillars: 1/ Threads, always. Reply in the thread, not the channel. One topic per thread, and if the conversation shifts you start a new message. A channel should be readable by someone who joined it five minutes ago. 2/ Communicate concisely. Ayush wrote our Bottom Line Up Front doc in December and it comes down to one thing. The ask or the conclusion goes in the first line, context after. Batch your thoughts into one message instead of five short ones back to back. Easier to read, easier to search, less noise for the people it wasn't for. 3/ Channels over DMs. If the answer could help someone else now or later, it goes in a channel. DMs are for private or sensitive things only. A customer issue debugged in a DM helps exactly two people. Three months later someone solves it again from scratch. 4/ Close the loop. If you're tagged and don't react, the sender has to assume you didn't see it. 👀 is seen and on it. ✅ is resolved. 👍 is an FYI that needs nothing. Banter is exempt. None of that needs threading or batching. Public channels let people self-serve instead of asking the same question twice. That runs both ways: before you ask, search. The answer is often already sitting in a channel, and we protect deep work here, so an avoidable ping that pulls someone out of focus is a cost the whole team pays. And at an AI-native company, your channels are also what your agents read. A decision buried in a DM is invisible to every person and every agent who needs it later. As far as the company is concerned, it never happened. We're 70 people today and will be a few hundred soon. If you want to join us as we scale, we have 20 open roles right now. Link in the comments to apply.

I’m excited to finally announce the newest edition my Stanford course 𝗧𝗵𝗲 𝗠𝗼𝗱𝗲𝗿𝗻 𝗦𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗿. It has been 9 months in the making. Last November, with the release of Claude Opus 4.5, coding agents experienced a step function improvement in capability. We all felt it. The LLMs were more powerful, could reason for longer, solve harder tasks. This year’s iteration of my course reflects the 2026 metamorphosis of software engineering. My core belief is simple: AI-native developers of the LLM era are going to become the most important members of any software organization. I have designed my course to train this next generation of engineers. 𝗪𝗵𝗮𝘁’𝘀 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝘁 𝘁𝗵𝗶𝘀 𝘁𝗶𝗺𝗲 𝗮𝗿𝗼𝘂𝗻𝗱 First, 85% of my Fall 2025 class material is being thrown out. The Fall 2026 syllabus reflects the core capabilities AI-native engineers must have: agent skills, advanced context engineering, MCP portals, agent-ready codebase principles, agentic code review, security, parallelizing background agents, software factories, and more. Second, I am going to teach my students how to have software taste. Every student will be required to ship pull requests to production-grade, real-world codebases. The course is collaborating with the top open-source AI repos who will offer support and mentorship to students on how to meaningfully contribute to their projects. This has never been done before in any university course so I am incredibly grateful to our OSS Partners: @browserbase, @HeyGen, @CopilotKit, @semgrep, @OpenHandsDev, @milvusio, @marimo_io, Pi, @crewAIInc, @warpdotdev, @vercel, @cmux, @arizeai, @UnslothAI, and @anyscalecompute. 𝗪𝗵𝗮𝘁’𝘀 𝘀𝘁𝗮𝘆𝗶𝗻𝗴 𝘁𝗵𝗲 𝘀𝗮𝗺𝗲 I’m fortunate to again have AI software engineering leaders and founders as guest speakers to share their learnings from building top coding agent products. Thank you to @leerob from @cursor_ai, @bcherny of @claudeai code, @EnoReyes of @FactoryAI, @silasalberti of @cognition, @0xine of @semgrep, Rajesh Bhatia of @Cloudflare , @amasad of @Replit, and @eladgil. All resources will be available online. All classes will be available to the public. 9/22 on Stanford campus. See you in class.  themodernsoftware.dev/

Almost half of US unicorns were founded by first-generation immigrants. The chart is from my earlier research on the topic (I should update it — and will), but the point stands: immigrants have an outsized impact on US innovation. At the time, 65 countries had produced at least one US unicorn founder. I'm sure the count is higher now.

First ever @cimentoai hackathon is underway! Hand selected builders here in Utah.

First CMU Agents class done today w/ @dan_fried, looking forward to getting the video processed and uploaded soon!

@cimentoai is hosting our first-ever Hackathon this Friday, and we have a few spots left. Humans get. Nobody's built it. So we're hosting a hackathon on the problem. Aug 28, SLC, 9 to 9. Capped room, approval required. Credits, prizes, and it's how we hire: luma.com/ojcetfo0

😊 I also just uploaded a new version of the lecture notes. cs229.stanford.edu/main_note… (it's still, and perhaps will always be work-in-progress, given the field is moving fast. Any mistakes are mine :) and any feedback is more than welcome :).

@cimentoai hosted one of the best security conversations I've had this year: no slides, no vendors, thirty practitioners, and Joe Sullivan (@joesu11ivan), former CISO of Meta, Uber, and Cloudflare. We need more Joes.

My conclusions from the end of the post

Here’s my review of Qwen 3.8 27B - I can't remember the last time I've had this much fun playing with a local model that runs on my own computers simonwillison.net/2026/Aug/1…

DeepSeek Harness was just released with MIT license. The current 0.1.0 version is a developer preview, and may still have many rough edges. Feedback is welcome! DeepSeek Harness 已经以MIT协议开源发布。现在的0.1.0版本是一个面向Harness开发者的预览版，还很不完善。恳请大家多提提宝贵意见。 github.com/deepseek-ai/deeps… deepseek.com/harness/