The Sequence Radar - Issue 944 — Last Week in AI: Last Week in AI: OpenAI Connects the Dots, Gemini Levels Up, and Agents Cash In
TheSequence Jesus Rodriguez ● Covered by 15 sources
OpenAI, Google, Meta, AMD, and investors all pushed harder on agents last week. The new pitch is simple: let software do the boring work, and charge for the privilege.
Based on reporting by TheSequence, Jesus Rodriguez — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
AI spent the week trying to get hired. Not as a toy, and not just as a smarter chat box. The common thread across the biggest announcements was delegation: persistent agents, longer reasoning, enterprise controls, and the infrastructure to make all of that look normal.
OpenAI made the most direct move. At DevDay, it introduced Dots, always-on agents powered by GPT-6 Astra, with their own cloud computer and connections to more than 4,000 apps. They can keep going across ChatGPT, Slack, and Teams, and they roll out to Pro and Business Premium users in eligible markets, with an Enterprise beta once a workspace admin switches it on. OpenAI also pushed GPT-6.1 Sol, which it says comes close to GPT-6 Astra on agentic coding, computer use, and professional work at about one-fifth of Astra’s standard token pricing. Ultrafast mode, ChatGPT Space and Pages, the Agents API computer use feature, Codex Cloud, Security Cloud, plugin extensions with MCP Events, and @ChatGPT in Slack and Teams all fit the same pattern: make the agent persistent, cheaper, and easier to hand real work.
Google answered with Gemini 4 Argon. The headline number is a one-million-token output limit, up from 64,000, which gives the model room for longer stretches of work. Google says it got 77.9% on DeepSWE v1.1, and it is first opening access to trusted cyber defenders through the Fairwind Program. That is a cautious launch, but the message is clear enough. Google wants the model to spend more time thinking, especially on software engineering, enterprise knowledge work, and cybersecurity.
Meta, meanwhile, hired for the enterprise side of the bet. It launched Meta Enterprise Platform and brought in MongoDB chief executive Chirantan “CJ” Desai to run it, reporting to Mark Zuckerberg. AMD also moved upstream, agreeing to buy Fei-Fei Li’s World Labs for about $8.2 billion in stock, because spatial intelligence and 3D simulation are starting to matter to chip strategy, not just research demos. And Instinct raised $1 billion at a $10 billion valuation for a personal agent that handles travel bookings, groceries, and forgotten subscriptions. The money is chasing the same idea everywhere: if software can actually finish tasks, it can start acting like a business.
My take — AI-written commentary, not fact-checked reporting
The loud part of the AI market is still model scores, but the real race is permissions, persistence, and billing. That is a much less glamorous story, which is probably why it is the right one. Everyone wants an agent; almost nobody wants the support ticket when it orders the wrong thing.
Read more about this at: TheSequence