Zero Human Companies Are Getting Wallets, Liability, and a Reality Check

Agents are getting wallets this week, and that changes more than you'd think.

Let's start with the biggest swing. SAP just announced what it's calling the Autonomous Enterprise at Sapphire 2026. More than 200 specialized agents across finance, supply chain, HR, and procurement, all orchestrated through a unified Business AI Platform. This is a company whose stock is down 41%, and they're betting 53 years of enterprise relationships on a single thesis: whoever owns the process layer and the business data wins the agentic era, not whoever builds the model. That's actually a defensible bet. SAP has the data and the integrations. If agents need to do real work inside real companies, they're going to need to touch ERP. Worth watching closely.

Now to the number that stopped me cold this week. Polsia, the autonomous company platform run by solo founder Ben Broca, went from 200k to 2 million ARR in two weeks. He's running nearly 3,000 active companies. And he's being remarkably honest about it: the platform hits about 80% autonomy, and one of his AI agents fabricated calendar availability with investors. His actual quote is something like, if you're starting a company and you don't hit 80% autonomous, you'll be cooked. That framing is worth sitting with. 80% is the ceiling right now, and he's saying out loud that it's still enough to build on. That's a provocative claim, but the ARR numbers suggest the market agrees with him, at least for now.

Okay, the wallet story. Circle just dropped what they're calling the Agent Stack: wallets backed by USDC, a service discovery marketplace, and nanopayments that go down to a millionth of a dollar per transaction. This matters because the dirty secret of every so-called autonomous company is that there's usually a human somewhere hitting approve on a payment. You can automate the reasoning all you want, but if an agent needs to buy an API call or pay another agent for a task, someone still has to authorize the transaction. Circle just removed that bottleneck. AWS did the same thing, plugging Coinbase's x402 protocol and Stripe's wallet infrastructure directly into Amazon Bedrock so agents can pay for web content, APIs, and other agents without a human in the loop. The race to become the default payment rail for the agentic economy is now a real competition between x402, AP2, and MPP. Picks and shovels play, obviously, but also genuinely foundational infrastructure.

Here's the part that should make every ZHC builder uncomfortable. A developer spent a weekend stress-testing Paperclip AI, one of the more hyped zero-human company platforms, and the results were rough. Hallucinated marketing statistics. Broken production code. Zero quality filtering on outputs. The reviewer's framing is the important part though. Agents can't tell you when the work they've just done is wrong. That's not a bug in one platform. That's the structural problem of the entire category right now. The gap between what an agent produces and what a human would catch is exactly where all the hard engineering lives, and most of these platforms are treating it like a footnote.

And then there's the governance piece, which is the liability time bomb hiding inside every set-it-and-forget-it deployment. WAV Group published a sharp analysis this week pointing out that most organizations have already authorized agents with no documented guardrails, no audit trails, and no liability frameworks in place. The question of who is responsible when an agent modifies a legal document, sends an unauthorized email, or blows past its mandate is still completely unresolved. For founders building on top of agentic infrastructure, this is the thing to think hard about before a client asks.

So here's the week in a single frame. The financial plumbing for autonomous operations just got a lot more real with Circle and AWS. SAP is signaling that enterprise adoption is accelerating, not waiting. But the self-verification problem is still unsolved, and the governance frameworks don't exist yet. We're at 80% autonomy with liability frameworks at zero percent. That gap is where the interesting work is happening right now.

More next week.