TLDR 2026-09-29
SpaceX reaches orbit 🚀, Meta poaches MongoDB CEO 💼, unread code 👨💻
Box, swyx, Cloudflare, and Claire Vo are init(). Are you? (Sponsor)
If you're building with AI, clear your schedule for October 7.
WorkOS is hosting, a one-day conference at SFJAZZ Center during SF Tech Week.
Hear from Evan Conrad (SF Compute), Aaron Levie (Box), Howie Liu (Hyperagent), swyx, Rita Kozlov (Cloudflare), Claire Vo, and more on the decisions, tradeoffs, and lessons behind the AI products they're shipping today.
Join the people building what's next.
Register for init() →
Meta launches enterprise AI platform, hires MongoDB CEO to lead new initiative (2 minute read)
Meta's new enterprise AI platform is an initiative aimed at expanding the company's AI offerings to businesses and corporate customers. Chirantan 'CJ' Desai, former CEO of MongoDB, is leading the initiative. The Meta Enterprise Platform will focus on turning the company's AI stack into products and services that companies can deploy for their own businesses. MongoDB has appointed Dev Ittycheria as interim chief executive while the board searches for a permanent replacement.
OpenAI Scraps Release of New AI Model Over Safety Concerns (6 minute read)
OpenAI is scrapping the release of GPT-6.1 Astra due to safety concerns. The company had aimed to launch the model in October, but the model performed poorly on tests measuring alignment compared to its predecessor. GPT-6.1 Astra showed higher levels of deception, and it would push ahead on tasks without asking users for permission. It would also reach for external tools and services even if it might be unsafe. OpenAI says it is working to investigate a range of agent security incidents and address the safety issues underneath them.
🚀
Science & Futuristic Technology
Airbound: As We May Move (130 minute read)
Airbound's first drone, the TRT, weighs in at 1.5 kilograms. The eVTOL can carry a 1-kilogram payload and can land and launch from anywhere, making logistics possible where infrastructure doesn't exist. The company's goal is to move mass more cheaply than anyone else. Its main focus is cost per kilogram-kilometer, and everything is secondary to that.
SpaceX's Starship Makes It to Orbit for the First Time, but Returns to Earth Early (8 minute read)
SpaceX's Starship circled Earth on Monday. The spacecraft had to return to Earth after less than two orbits instead of the planned six because of engine problems that occurred during liftoff. The Starship spacecraft splashed down with a fiery explosion in the North Pacific near Hawaii before noon. Reaching orbit is a key milestone towards SpaceX's goal of eventually sending people to Mars as well as NASA's plans to send astronauts back to the moon.
💻
Programming, Design & Data Science
Stop asking users to test your AI app for you (Sponsor)
Offline evals let you validate before you ship, and
this Datadog guide shows you how. Learn to build annotated test datasets and trace multi-agent workflows from end-to-end during experimentation, just like you would in prod. Then, prevent model drift by aligning offline tests with prod.
Get the frameworkThe Code Nobody Reads (24 minute read)
Know what's worth reading, and on the code you don't read, build checks you actually trust. Good code is a love letter to the next developer. The next developer is now an agent, so the letter looks different now. You need to pass on code that agents can change without breaking something it can't see. The craft has now moved to the parts that decide whether the code deserves to be trusted.
Anthropic has launched Claude Sonnet 5.5 (4 minute read)
Claude Sonnet 5.5 scores just 2 points behind Opus 5.5 (max) on the Artificial Analysis Intelligence Index. However, it uses around 60% more tokens to reach that performance. The model is priced at $2/$10 per million tokens of input/output, matching GPT-6 Sol. Sonnet 5.5 still lags on factual knowledge as it is a smaller class model than Opus 5.5, but it has a lower hallucination rate.
Apps, Agents, and Aggregation (17 minute read)
The new prize in technology is to be the only interface users need for everything, with agents using computers on users' behalf and generating whatever UI is needed. For people who are using agents, this is obvious. Apps are just a vehicle for people to accomplish something or entertain themselves. Once people can get straight to the job to be done, it will seem odd that they did it any other way.
OpenAI Understands Something Important and Rare (17 minute read)
OpenAI understands that the purpose of a business is to create a customer. The company is really good at creating new kinds of customers. It has a well-established track record of being able to find the simple, obvious-in-hindsight breakthroughs that unlock new kinds of consumption behavior. This, along with its durable distribution strategy, will help it become the winner in the AI race.
In the office full time? Make lunch worth the commute (Sponsor)
DoorDash's pre-populated meal calendar serves up a wide selection of restaurants so your team isn't stuck with the same options on repeat and is excited to come into the office.
See how it works
Build for the Agentic Internet at Cloudflare Connect 2026, Oct. 19-21 in San Francisco: (Sponsor)
100+ sessions, five tracks, top speakers, hands-on labs, and practical AI, security, and networking guidance.
Register NowApplied mathematics has met the machine before (16 minute read)
Each time a machine has become better than humans at something done by hand, applied mathematics absorbed it and grew.
We're building multiplayer AI. Here's what we've learned so far (12 minute read)
Multiplayer AI requires shared context to save teams time, energy, and mistakes.
Introducing Forge: the open source pipeline for generating SDKs, CLIs, docs, and more (7 minute read)
Forge is an open source, pluggable generation pipeline that can create SDKs, CLIs, docs, and libraries.
The road to the agentic browser: A Kitesurf update (8 minute read)
Kitesurf is a browser that runs entirely on Cloudflare Workers.
Deployment Gap (5 minute read)
Robotics takes a lot of time to develop, so the industry needs patient capital.
Nvidia Releases Software It Says Can Prevent AI Agents From Going Rogue (2 minute read)
The Nvidia Open Agent Safety Platform has tools that let developers continuously monitor and govern AI behavior to ensure agents follow the rules.
Get the most interesting stories in startups, tech, and programming delivered in a free daily email.
Join 1,600,000 readers for
one daily email