What the people actually building AI said today. One page — a 2-min read.
AI security dominated the day, with builders warning that agents are accelerating vulnerability discovery while also offering a way to defend at scale. Replit expanded its agent and government efforts, and Codex shipped a set of practical CLI updates.
Chroma team joins Replit to work on agent memory
Amjad Masad welcomed Jeff and the Chroma team to Replit, where they will work on giving agents the context and memory needed for real work. Replit also announced Replit for Government and said Michael Greenwald, who previously led AWS’s public sector business, is joining the team.
Cloudflare expects an intense stretch of AI driven security work
Dane Knecht said serious vulnerabilities are surfacing far more frequently and expects security to consume much of engineering teams’ time over the next 18 months. Cloudflare plans to expand its red teams and build security checks into development, while keeping human judgment focused on reducing code complexity. As he put it, “if you're not shipping, you're dying.”
Shared agent machines face a hard to eliminate timing attack
Guillermo Rauch described how an agent on a shared machine could signal to another through CPU cache activity that the second agent detects by timing its own work. He said dedicated hardware is the only complete isolation, at significant cost, and that Vercel Sandbox uses practical mitigations alongside monitoring and red teaming.
Codex CLI adds prompt presets and enterprise MCP sign-in
Codex CLI now has reusable prompt presets, inline model and permission settings, and enterprise MCP sign-in. Updates also address long-session resumption and Windows corporate-proxy authentication. Thibault Sottiaux said CLI usage has doubled since DevDay.
Thariq leaves Anthropic after working on Claude Code
Thariq said he has left Anthropic after more than a year working on Claude Code and plans to take a break before starting something new. He credited colleagues and the community for the product’s growth and said he still believes in competing to raise the bar responsibly.
Aaron Levie predicts agent swarms will become a major cyber category
Levie expects agents to review code and infrastructure, test many possible risks, and process security data at a scale human teams cannot match. His case for cyber defense swarms is that additional computing power can directly expand the amount of testing done.
Broad training problems may drive robotics advances
Aditya Agarwal pointed to AI advances that came from solving many problems together instead of training a separate system for each task. He highlighted Sergey Levine’s argument that robotics could benefit from the same approach: make the training problem bigger.
Source data comes from the open-source project follow-builders by zarazhangrui, released under the MIT license. Summaries are generated by an LLM from that project's public feeds, and the summarization prompts are adapted from it. Every item above links to its original source.
Summaries generated automatically. Read the original before relying on any claim.
