TOP LAUNCHES

Supercharge Codex

• Codex Plugins package reusable Codex workflows into installable bundles. Teams can bundle skills, app integrations, and MCP server setups together, then reuse the same setup across projects instead of rebuilding it every time.

• Agentation turns UI feedback into something coding agents can actually use. Click any element on a page, add a note, and it gives you structured output with selectors, file paths, component hierarchy, styles, and your feedback, so Claude Code, Codex, Cursor, or whatever else is not left guessing which blue button you meant.

• Claude Code auto-fix watches your pull requests in the cloud and responds to CI failures or review comments on its own. It pushes a fix when the change is obvious, asks when it is not, and keeps running even after you close the laptop.

• Linear Agent lives inside Linear and works across the app, comments, Slack, Teams, and mobile. It can pull context from your roadmap and issues, answer questions, create issues from conversations, recommend what to do next, and trigger workflows when new work hits triage. Code intelligence is also on the way, with public beta available now for all teams.

• Magine spins up autonomous browser agents that see the web the way a person does. Instead of falling apart when a button moves, they watch the screen, click around, log in, post, monitor dashboards, and run scheduled browser tasks from a terminal-style UI.

FROM THE FORUMS

Seven days of Claude

Imed ran a seven-day experiment where Claude wrote all the code with no human review, and it went about as calmly as you would expect. It shipped some working features fast, then started breaking unrelated parts of the product, driving up API costs, dropping old data, hardcoding stuff that should have been configurable, and taking the site down with a dependency update. The post is basically a very clean reminder that code generation is not the same thing as judgment. 

WHAT’S HOT

End of video slop?

OpenAI is shutting down Sora, which is a pretty sharp comedown for a product that was pitched like the future of video. It launched as a TikTok-style app for AI clips, shot to the top of the App Store, and six months later it is being wound down.

The story is not really one big dramatic failure. It is a pile of very expensive problems. Video generation costs a lot, user interest dropped off fast, and Sora kept attracting the kind of deepfake, copyright, and moderation headaches that make a flashy demo look a lot less fun in production.

So this reads less like AI video is dead and more like OpenAI deciding this particular experiment was not worth the chips, the mess, or the distraction. Reuters says the company is shifting focus back to core products and broader AI work, while other reporting points to competition from tools like Google’s and Kling’s video models making the whole category harder to justify.

That is the part that matters. Sora was one of the most hyped AI products on the market, and even that was not enough to save it from the boring stuff: cost, retention, and trust. Turns out making jaw-dropping videos is one thing. Building a durable product around them is another.