AI Agents Go Mainstream: OpenAI, Meta, and Open Source All Ship

If 2025 was the year of agent demos, this summer is the year of agent products. OpenAI announced ChatGPT Work, an agent offering aimed at knowledge work that is planned to merge into the main ChatGPT experience — against the backdrop of ChatGPT reportedly reaching about one billion weekly active users. Google and Anthropic are likewise putting task-running agents in front of broader user bases.

The developer side is moving just as fast. Meta released Muse Code, a terminal-based coding agent built for repository-level work, and Prime Intellect open-sourced Prime Agent under the MIT license — an autonomous coding agent that keeps persistent context across steps and posted a 95.5% score on the ARC-AGI-3 benchmark, according to the announcement.

What we tell clients: the barrier to adopting agents is no longer the model — it is your own plumbing. Agents are only as useful as the APIs, permissions, and audit trails you can wire them into. Companies that invested in clean internal APIs are deploying agents in weeks; companies with tangled legacy integrations are still stuck in pilot mode. If you want an honest assessment of which camp you are in, that is literally our job.

Sources: Radical Data Science AI news briefs, Build Fast with AI roundup.