WEEKLY BULLETIN / WEEK 41, 2026
Always-on agents are not coming to Europe, and whoever has an agent built must be able to maintain it
Five developments from the past week in AI assistants and agents, with the source and what it means for you.
01
OpenAI’s always-on agents are not available to European Pro users
OpenAI has been rolling out “dots” since 29 September: agents that keep working on a goal inside ChatGPT. According to OpenAI’s developer forum and Mixed News, Pro users get them “in markets excluding the European Economic Area, Switzerland, and the UK”; only Business Premium gets them in all regions. There is no date for the Netherlands.
What it means for you. If you want an agent in the Netherlands that keeps working while you are away, waiting for the large providers will not get you one. A tailored assistant on a server in Europe does that today.
When ChatGPT, Claude or Copilot is enough, and when it is not (in Dutch)
Source: OpenAI Developer Community, DevDay 2026 announcements and developer resources, 29 September 2026; Mixed News, OpenAI’s new always-on ChatGPT dots exclude Pro users in the EEA, Switzerland and the UK, 30 September 2026; releases.sh, ChatGPT release notes (dots: supported markets), 29 September 2026
02
Gartner: seven in ten companies will abandon agents a vendor built for them
Gartner predicts (29 September) that 70% of large enterprises will abandon AI agents built by vendors’ embedded engineers by 2028. The reason, according to The Register: costs rise and the company cannot adapt the system itself. Gartner advises agreeing on ownership, knowledge transfer and an exit from day one.
What it means for you. If you have an agent built, ask where the settings live and whether you can take them with you as ordinary files. Tailored only works if it stays yours.
Who owns your data, settings and agents, and what happens when you stop? (in Dutch)
Source: The Register, 7 in 10 enterprises expected to abandon vendor-built agentic AI by 2028, 30 September 2026; Gartner, Gartner Predicts 70% of Enterprises Will Abandon Agentic AI Built by Vendor Forward-Deployed Engineering by 2028, 29 September 2026
03
Google releases Gemini 4 Argon, first for security teams, then for paying customers
On 30 September Google announced Gemini 4 Argon, its new top model for programming, legal and financial work and cyber defence. It goes first to a group of security teams in the Fairwind Program, then to paying API customers and AI Ultra subscribers. The introductory price is 2 dollars per million input tokens and 10 dollars for output; after that 4 and 20 dollars.
What it means for you. The best model changes name and price every few weeks. An assistant that picks the right model for each task benefits from that without you having to switch.
What does the technology behind an AI assistant cost per month, apart from the set-up? (in Dutch)
Source: Google, Gemini 4 Argon: our next era of frontier intelligence, 30 September 2026
04
Meta launches Muse for small businesses, in the US and Canada
On 29 September Meta released Muse for Small Business: an agent that works in the background on goals such as finding new customers, connected to Shopify, Stripe, QuickBooks, Canva, Slack and your Facebook and Instagram accounts, among others. Nothing is published, sent or paid without your approval. Available in the US and Canada; the basic version has no charge, with paid plans for those who want more.
What it means for you. Small businesses are now getting agents that work inside their software too, but Europe waits. What you can do today: connect your own software to an assistant that knows your rules.
Connecting what already works well: why a good assistant does not do everything itself (in Dutch)
Source: Meta, The Future Is for Everyone: Muse for Small Business, 29 September 2026
05
Anthropic puts 100 million dollars into people who set up AI inside companies
On 2 October Anthropic announced the Claude Frontier Academy: 100 million dollars to train ten thousand engineers by the end of 2027 who take AI inside companies from idea to working system. The first cohorts run in San Francisco, New York and London, with Accenture, Deloitte and Novo Nordisk among others. Anthropic calls talent “one of the most pressing issues in AI implementation”.
What it means for you. The bottleneck is not the models, but the people who set them up and keep them running. For a single entrepreneur or a small firm, that is a service you choose separately.
Who sets up your AI assistant, and who maintains it when the models change? (in Dutch)
Source: Anthropic, Anthropic invests $100 million to train 10,000 engineers and tackle the enterprise AI talent gap, 2 October 2026
From the knowledge base
Does ChatGPT, Claude or Gemini train on your data? The business versions do not by default and keep data briefly; the consumer versions often do, unless you switch it off. Look not at the brand name but at the plan and the terms you actually use.
Updated
No articles changed this week; the proposed updates await approval.
Curious what this means for you?
It starts with a conversation about your work.