AI Agents Are Now Running Actual Businesses

AI Agents Are Now Running Actual Businesses

Virtuals is reporting $481M in "agentic GDP" — real revenue, real jobs completed, recurring agent-to-agent business. This isn't chatbot usage stats dressed up as an economy. It's agents transacting with other agents, completing work, and generating repeat business without a human closing the loop each time.

The shift here is from "AI assists a person" to "AI is the economic actor." That's a genuinely new category, and the number — while still small in absolute terms — is the kind of early signal that mattered in previous platform shifts (think API call volume in the early cloud days). The reaction on X has been notably bullish, treating this as confirmation that agents are moving past demo-ware into something with a P&L.

Worth watching whether this is concentrated in a few niches (data processing, simple transactions) or actually broadening. Either way, "agent as mini-enterprise" is a mental model worth adopting now, before it's obvious in hindsight.

Glean's Waldo Bets the Real Gains Are in Routing, Not Bigger Models

Glean launched Waldo, a system that routes tasks to the right model or tool rather than defaulting to the biggest one available. Results: 25% fewer tokens, half the latency, no quality loss. The thesis — that the context layer is a bigger lever than raw model capability — is exactly the argument the "bigger model wins" crowd doesn't want to hear.

This matters because most teams are still burning budget throwing frontier models at tasks a smaller, cheaper model could handle just as well. Waldo is a bet that orchestration — knowing what to call when — is where the margin actually lives, not in chasing the next parameter count.

Tech Twitter's reaction has been pragmatic rather than hype-driven: efficiency wins don't trend the way new model releases do, but they're the difference between an AI product with real unit economics and one that's subsidized by VC money. Expect more infrastructure plays like this one as compute costs stay a genuine constraint.

Berlin Cements Itself as Europe's AI Capital

German AI startups are having a record H1 2026, with VC funding surging as part of a quarter where AI swallowed 81% of all venture funding globally [2]. Berlin specifically now hosts hundreds of AI-focused startups backed by billions in capital, positioning it as Europe's clearest AI hub [1], and "big VC deals are back" for German startups more broadly after a quiet couple of years [3].

This is a useful counter-narrative to the usual "Europe is behind" framing. Capital is concentrating somewhere, and it's concentrating in Berlin — not London, not Paris. For Nordic founders and operators, that's a signal about where partnerships, talent competition, and acquirers are going to come from over the next 18 months.

UK's AI Ambitions Are Colliding With Its Water Supply

The UK water industry is publicly pushing back on the government's AI growth plans, warning that datacenter buildout has largely ignored water demand — a resource constraint that doesn't show up in most AI infrastructure conversations, which tend to fixate on power and chips.

It's a reminder that the physical constraints on AI scaling are wider and weirder than the compute story everyone's used to telling. If you're planning datacenter-dependent products or partnerships in the UK, this is a policy risk worth tracking now, before it becomes a permitting bottleneck.