Leaked code names OpenAI's always-on coworker: Aeon

Two scoreboards went public on Tuesday and a line of leaked code arrived the same week. Together they say something OpenAI would rather not hear: in the personal-agent race it started, it is now the one catching up.

OpenAI is developing features aimed directly at SpaceXAI's Grok Bot, and leaked ChatGPT client code suggests the always-on coworker it is building may be called Aeon. The Information reported that OpenAI is working on features to compete more directly with Grok Bot — the AI teammate that signs into your apps and works from a persistent cloud computer — and that company leadership has discussed a dedicated personal assistant to answer Meta's Muse. The awkward part is how OpenAI intends to get there: not by building a new product, but by repackaging agent capability it already ships inside ChatGPT and Codex. The same report notes ChatGPT's growth has slowed, and that the 1 billion weekly actives OpenAI originally targeted for last year were only reached this summer.

Then came the code. Tibor Blaho, the engineer who has spent years pulling apart ChatGPT builds, posted a screenshot from the client showing an aeonId field listed alongside accountUserId — a second identity inside the product, not a user's — with a memberAeonIds field recurring across two consecutive Android releases. A test note in the same build describes protecting "aeon-style behaviour," where the capabilities available to the system are re-captured at every sampling step instead of once per user turn. That is the plumbing a multi-step agent needs and a chat turn does not. OpenAI has not commented, and none of this is a product announcement; treat it as a codename and a few field names, not a launch date.

Why it matters: the difference between the three products is not intelligence, it is where the agent lives and what it is allowed to hold. Grok Bot's bet is a cloud computer that keeps your logins and browser state; Muse's bet is the app you already open every day; OpenAI's execution layer — ChatGPT Work, cloud runs that continue after you close the laptop, the Codex harness — has existed for months. What it lacks is a named teammate you delegate to and a story about permission, which is precisely the gap competitors are monetizing. Reporting from 36Kr adds that OpenAI hired OpenClaw's author, Peter Steinberger, in February to build its next-generation personal agent. The technology was never the missing piece. The trust conversation is.

Grok Bot passed 418,000 users as of September 14, up 24 percent from the week before, according to a presentation Bloomberg viewed at an event in London. SpaceXAI launched the agent on August 11 with Cursor, positioning it as an always-on team rather than a single assistant: every Bot under one username shares a persistent cloud computer holding files, browser sessions and logins, and several Bots can split work between them. We covered that launch in August — SpaceXAI launches Grok Bot, a cross-platform AI agent app — and the first-month number is the first hard read on whether handing over the keys is a mainstream instinct or an early-adopter habit. At 24 percent weekly growth it is still climbing; against ChatGPT's reported 1 billion weekly actives it is also still a rounding error.

Meta's Muse took the No. 1 spot on Apple's US free iPhone chart on September 18, ten days after launch, with more than 730,000 US downloads counted by Sensor Tower. That overtakes ChatGPT, which had held the top of that chart, and it lands a second blow in the same week: Muse is the consumer agent that competes for the app you already open, while Grok Bot competes for your credentials. Our September 10 read on Muse's climb — Meta's agent Muse hits No. 2 on the US App Store — slowly — questioned whether chart position would hold. It held, and it moved up.

What to watch: whether OpenAI ships a named agent before year-end, and whether Grok Bot's weekly growth rate survives past the beta crowd it recruited in August.

Would you hand an always-on agent your logins if it meant never opening your inbox again? Tell us in the comments.

Akhetonics says its all-optical CPU reaches a customer in 2026

Light-based computing keeps promising more than it delivers — but one Munich startup has just put a date on its bet, and the interview laying it out is doing the rounds on Hacker News this week. Akhetonics says it will deploy its first commercial machine with a major customer by the end of 2026, with several more planned for 2027. The company, founded by Michael Kissner and Leonardo Del Bino, is building a computer where data enters as light, is switched as light, and circulates through memory

The Week in AI — October 5–11, 2026

Every big claim this week turned out to rest on fine print more interesting than the headline: revenue only the company reporting it can define, safety tests sandboxed while the product keeps the web, and a Pentagon phase-out nobody would confirm until reporters kept asking. The week's top 5 1. OpenAI's revenue was $20 billion below the numbers everyone quoted — and the gap was definitional. The Financial Times reported Thursday that OpenAI's annualized revenue runs roughly $20 billion unde

Drone strike shuts a third Yandex data center, taking YandexGPT offline

Russia's largest tech company is learning what the AI era's infrastructure war looks like from the receiving end — three data centers in four days, and with them much of the cloud layer Russian businesses run on. A Ukrainian drone strike knocked out Yandex's data center in Vladimir early Sunday morning, the third of the company's facilities hit since October 8. The site — reported at roughly 50 MW and designed for about 2,880 server racks — stopped operating completely after the attack, Yandex

Agent teams cost up to 5x more, barely score higher

The multi-agent hype train hit a benchmark this weekend — and the grid and the trucking regulators had quiet weeks of their own. Vals AI put agent teams head-to-head with single agents on its Vibe Code Bench, and the teams cost between 1.8 and 5.1 times more for almost no extra quality. The evals company ran GPT-6 Sol and Claude Opus 5.5 solo and in teams across 50 apps at two reasoning efforts; out of four comparisons, only one was statistically significant — Sol at medium effort, where the t

Read more

Akhetonics says its all-optical CPU reaches a customer in 2026

The Week in AI — October 5–11, 2026

Drone strike shuts a third Yandex data center, taking YandexGPT offline

Agent teams cost up to 5x more, barely score higher