Future Technology[AI Weekly] The week your AI assistant got a wallet
Three of this week's biggest stories have almost nothing to do with smarter models. They're about what AI is now allowed to do on its own: spend your money, help design its own successor, and act without a human in the loop. The capability race is quietly turning into a permission race.
The Big 3
Your AI assistant can now spend your money
Coinbase launched Coinbase for Agents, which lets assistants like ChatGPT and Claude actually move money: trade crypto, pay for services, buy access to paid data, and eventually shop on your behalf. For years the advice was never to hand your card details to a chatbot. That advice just expired.
Why this matters: if you use an AI assistant for anything transactional, the bottleneck stops being whether it can decide and becomes how much you trust it with a live wallet. Spending limits are about to be the most important setting you configure.
Read more → (4 minute read)
The EU moved its biggest AI deadlines, and banned nudifiers
The EU's Digital Omnibus package pushes the heaviest high risk obligations back: standalone high risk systems now apply from 2 December 2027, and high risk AI inside regulated products from 2 August 2028. It also widens lighter compliance to firms with up to 750 staff and 150 million euro revenue, and from 2 December 2026 it explicitly bans nudifier apps that generate intimate images without consent. Transparency rules still land in August 2026.
Why this matters: if you build AI products for the EU, you just got breathing room on the expensive high risk paperwork, but the August transparency rules (labelling AI output, disclosing chatbots) still apply this summer.
Read more → (5 minute read)
Anthropic is letting its AI help build the next AI
Anthropic confirmed it's leaning into recursive self improvement: using its own models to help design their successors. Internal benchmarks claim a typical engineer now ships roughly eight times more code than a couple of years ago. Read that as ambitious framing rather than gospel, but the direction is the story. The people building frontier AI are increasingly using frontier AI to do it.
Why this matters: faster internal loops mean releases keep arriving quicker than anyone can properly evaluate them. If your roadmap assumes a stable model every 12 months, it's already out of date.
Read more → (4 minute read)
Quick Hits
China’s open models keep getting harder to ignore. Zhipu AI released GLM-5.2 under a permissive MIT licence, a large mixture of experts model with a usable one million token context window. For teams that want serious capability without a closed API, the self hosted option just got stronger. Read more → (3 minute read)
Microsoft’s "Autopilots" want their own logins. Microsoft introduced a new agent category called Autopilots, led by Microsoft Scout: an always on autonomous agent with its own governed identity inside your org. Handy, and also a reminder that who has access now includes software that never sleeps. Read more → (3 minute read)
Agents are racking up bills, so here’s the meter. Databricks launched Unity AI Gateway to cap and monitor AI spend across providers, with spending limits and runaway usage protection. The unglamorous plumbing of the agent era: someone has to watch the bill. Read more → (3 minute read)
ChatGPT slipped below half the market. ChatGPT’s share of the global AI assistant market fell to 46.4%, its first time under 50%, with Gemini at 27.7% and Claude at 10.3%. Still the leader, just no longer the whole conversation. Read more → (2 minute read)
Tool of the Week
What it is: a TypeScript and Python toolkit for building your own AI agents that call tools, run MCP servers, and spawn sub agents.
Who it's for: if you've outgrown copy pasting prompts and want to wire an assistant into your actual systems, this is the cleanest on ramp.
Honest caveat: this is a developer tool. If you don't write code, start with a no code agent builder instead. There's no affiliate programme either, so that link earns us nothing.
Read more → (3 minute read)
One More Thing
We spent a decade teaching people never to give their card details to a chatbot. This week we started handing them the wallet and the keys. The technology is ready. The open question for the rest of 2026 is whether our habits, and our spending limits, catch up in time.
Forward this to someone who'd find it useful. Or don't. But you should.
Free weekly AI briefing: futuretechnologyhq.com/newsletter
Some links in this newsletter may be affiliate links. We only recommend products we genuinely think are worth your time.