Edition #15. On Thursday, Anthropic published a notably political policy piece about 2028, with two scenarios for who wins the AI race and what democracies must do now to avoid falling behind. On the same day OpenAI brought Codex into the ChatGPT app for iOS and Android, letting you steer coding tasks and approve steps from your phone. Otherwise it was a quiet news week, so I'm also sharing my takeaways from last Friday, when I spoke at the SV Anatolia alumni event about AI in Dutch SMBs. Let's dive in!
Anthropic sketches two scenarios for 2028, and openly picks a side
On Thursday, Anthropic published a policy piece titled "2028: Two scenarios for global AI leadership". The tone is unlike what we're used to from an AI lab: not about models or benchmarks, but about geopolitics. The core claim: democracies currently hold a 12 to 24 month lead in frontier models, and the next few months will decide whether that holds or China catches up.
The figures they cite are concrete. With current export controls, the US and allies have roughly eleven times more compute than China. Huawei produces about four percent of NVIDIA's aggregate compute in 2026. Anthropic's recommendations follow logically: tighter export controls, better protection of models against distillation attacks by Chinese labs, and actively pushing the US AI stack as the global standard. And in a line that lingers: "AI will soon become powerful enough to be used to repress citizens at unprecedented scale."
What I take from this: Anthropic now openly positions itself as US-aligned, and that has practical consequences for anyone using Claude. Which models can run where, which customers get access, how export rules are tightened. For European companies it raises a question we've put off too long: do we build our processes on a US stack, or do we make sure we have options? A question that belongs in any AI strategy you put together this year.
OpenAI brings Codex to the ChatGPT app: your phone becomes the approver
On the same Thursday, OpenAI shipped a preview of Codex in the ChatGPT app for iOS and Android. Available to all users, free plans included. You can start tasks from your phone, watch real-time screenshots and terminal output, and approve steps while Codex runs on your desktop, Mac mini or server. The four million weekly Codex users now have their workstation in their pocket.
On its own this isn't a big launch. But the pattern underneath is interesting. AI agents are going async: they start, run long, occasionally need an approval, deliver a result. The human shifts from operator to approver. Anthropic has the same model with Claude Code sessions on a remote server, but without a native mobile app. The real question is whether this way of working fits how you work.
What I notice myself since I started running agents on a remote server: the separation between "I'm working" and "the agent is working" disappears. The agent asks a question, you answer from your phone, it keeps going. It takes getting used to, but if the workflow is set up right, your output doubles without you putting in more hours. For teams setting up their first agents, this is the right moment to get the async flow right.
Last Friday I was a guest at SV Anatolia: three takeaways from the room
Last Friday I was a guest speaker at the SV Anatolia alumni event. After a short intro I talked about how I use AI in my own business, where SMBs currently stand with AI, and where I still see major growth opportunities for entrepreneurs and employees in larger organizations. After that I got into a conversation with attendees from different sectors about how they use AI today. Three observations from that evening stayed with me.
1. Everyone is on AI. If you're not, you fall behind.
What stood out: virtually everyone in the room uses AI daily. ChatGPT for email and summaries, Claude for longer pieces, sometimes already a custom GPT or an MCP connection. Five years ago only developers would have been like this. Now it's HR, sales, finance, legal, everywhere. "Do you use AI?" is becoming the wrong question. The right question is "how well do you use it?"
2. Younger employees do the same work faster than older ones.
Several real examples in the room: younger colleagues pick up AI faster and as a result do the same work in less time than older colleagues. That's not a judgment about who's smarter, it's a skills gap you can close with training, as long as you take the time to actively bring along people who didn't roll into AI naturally. For organizations thinking about internal AI training: this is exactly the gap you want to close.
3. Individually everyone is already on AI, but commercially SMBs are still lagging.
This was the most striking observation of the evening. People use ChatGPT at home on their phone to write emails and quote text, and at the office the same processes still run on Excel and email. The gap between what people already do privately and what they do at work is currently the biggest piece of low-hanging fruit in AI for SMBs. Thanks to SV Anatolia for the invitation and the great evening.
Want to work smarter with AI?
Want to work smarter with AI, in your business or just for yourself?
At AI Agent B.V. we help companies and individuals with AI advisory, AI implementation and practical training. So not just nice stories about what AI can eventually do, but concretely helping with how to use AI smarter today in your work, processes and daily operation. From strategy and tooling choices to training, implementation and custom AI agents.
Interested? Schedule a free 30-minute call.
