The week AI stopped chatting and started working

Editor's note

There is a quiet shift running through this week's stories, and it is about output. For two years we have been training ourselves to chat with AI. This week the headlines are about AI producing things, dashboards, documents, research direction, voice that sounds like a person rather than a hold queue. OpenAI is pushing its models into labs and mathematics departments, Anthropic is turning conversations into deliverables, and Google is doing what Google does, nudging us back to the fundamentals of analytics prompts and URL hygiene. If your 2027 planning assumes AI is a drafting tool, the gap between that assumption and what your competitors are shipping is widening fast.

Agents are now the majority of enterprise AI use in Europe

OpenAI's latest Economic Research report says more than half of AI use among its EU enterprise and business customers now involves agents. Small business users of these tools have grown eightfold across the region. The report frames Europe's shift towards agentic AI as an adoption story with an obvious next chapter, turning broad usage into deeper usage.

What it means for you: Breadth of adoption is no longer a differentiator in Europe, depth is. If your teams are using agents for one-off tasks rather than wiring them into repeatable workflows like reporting, lead qualification or content production, you are in the shallow half of that statistic. Audit where agents sit in your funnel, not just whether people have access.

Claude collapses the gap between chat and deliverable

Anthropic has removed the chat-or-cowork choice. You can ask Claude to analyse your spending, then ask for the same thing as a doc, a dashboard or a design, all inside one conversation. The commands are simple, /doc, /dashboard or /design.

What it means for you: This compresses the analysis to artefact cycle that eats so much of a marketing team's week. The practical win is in reporting, where the journey from data question to stakeholder-ready dashboard has always involved a handoff. Try it on your next monthly performance review and measure the hours saved before you buy another reporting tool.

OpenAI goes deeper into research, with Rosalind and $5m for mathematics

Two moves in the same week. OpenAI is expanding access to GPT-Rosalind for academic research through a new Rosalind Academic Plan, aimed at helping scientific labs understand biology, explore ideas and decide what to test next, with sponsored access for eligible labs at US universities. Separately, OpenAI is contributing $5 million to a new Simons Foundation initiative so mathematicians can examine, understand and build on AI-produced results, including results OpenAI recently published. The Simons Foundation is committing $15 million overall across partnerships with OpenAI and Anthropic, funding workshops, conferences and special programs, plus model and compute access for supported researchers.

What it means for you: This is not a marketing story on the surface, but it tells you where credibility is being built. The labs are investing in verification and human understanding of machine output, which is exactly the posture buyers will expect from you. If you publish AI-assisted research, data or thought leadership, show your working. The provenance of a claim is becoming part of the product.

Voice gets less awkward with GPT-Live-1.1

OpenAI shipped GPT-Live-1.1 to Voice in ChatGPT, Codex and dots, promising new voices, better accents and less of that "checking..." filler. New voices arrive in French, German, Hindi, Mexican Spanish, Korean and Japanese. It rolled out to all users immediately, with API access coming soon.

What it means for you: Latency and accent quality were the two things keeping voice out of serious customer-facing deployments. With the API on the way, start scoping where voice actually fits in your business, support triage, multilingual onboarding, in-store or in-app assistance. The language list is also a hint about where conversational discovery is scaling first, so if you sell into those markets, get your localised content in order.

Google's housekeeping: Ask Advisor prompts, Cape Town, and a rubber chicken

Three smaller items worth your attention. Google published a Help Center guide on getting more from Ask Advisor in Google Analytics, with prompting examples and step-by-step setup, built around the point that the quality of your question determines the quality of the answer, whether you are troubleshooting a traffic dip or hunting growth opportunities. Search Central Live returns to Cape Town on 11 November, with free seats available by application before 19 October. And Google Search Central launched episode one of The Non-Canonical Cabinet, covering what can legally go in a domain name, internationalised domains and Punycode, emoji in URL paths and percent-encoding, whether keyword-stuffed URLs win rankings, and the hyphens versus underscores question, now settled.

What it means for you: Prompting is becoming a genuine analytics skill, so get your team trained on Ask Advisor rather than letting it sit unused in the interface. If you have anyone in South Africa, the Cape Town deadline is tight, apply this week. And forward the URL episode to whoever on your team keeps proposing emoji domains, it will save you a conversation.

Looking ahead

The teams that win the next quarter will be the ones who stop asking what AI can tell them and start asking what it can hand over, finished.