The newest zero-human company signals are skewing back toward North America, and that concentration is meaningful. Capital is pricing autonomous software engineering, cloud vendors are packaging agent guardrails into product, agent workbenches are becoming control planes, and voice models are getting good enough to move from demos into live operations.
1. Investments: Cognition Is Pricing the Autonomous Software Factory
On May 27, 2026, Cognition said it had raised more than $1 billion at a $26 billion valuation. The company also said enterprise usage is up more than 10x since the start of the year and that run-rate revenue has reached $492 million.
That is not normal assistant pricing. It is capital treating autonomous software work as a durable control point. The important detail is not just Devin itself. It is the claim that cloud agents have moved from niche to mainstream and that large buyers are willing to operationalize them around security remediation, modernization, and internal engineering throughput.
This extends the pattern we mapped in deployment companies. The stack is no longer only about selling intelligence. It is about owning the workflows where agents repeatedly do high-value work.
2. Frameworks: AWS Is Turning Agent Best Practices into Product
AWS has quietly made one of the clearest framework moves of the month. On May 6, 2026, it launched the Agent Toolkit for AWS, a bundle of managed MCP infrastructure, curated plugins, and more than 40 validated agent skills. Then on May 14, 2026, AWS said its Transform agents were available in Kiro, Claude, Cursor, and Codex.
The strategic shift is that the framework is no longer just a developer library. It is a vendor-defined path for how agents should build, authenticate, observe, and recover inside a real cloud environment. That makes hyperscalers part of the agent harness layer, not only the compute layer.
3. Tooling: Warp Is Splitting Harness Choice from Model Choice
Warp made two product moves that matter together. On May 19, 2026, it said Oz now acts as a multi-harness control plane for Claude Code, Codex, and Warp Agent. A day later, Warp added bring-your-own inference and OpenAI-compatible endpoints.
OpenAI's May 27, 2026 customer story on Warp adds the strongest operating proof so far: nearly 1 million developers on the product, usage across more than 56% of the Fortune 500, and around 90% of Warp's internal pull requests created with agents.
That matters because it separates two layers many teams still conflate. One layer is the harness that plans, edits, and executes. The other is the inference and routing setup behind it. Zero-human companies will want optionality on both.
4. AI Capabilities: Voice Agents Are Getting Closer to Real Frontline Work
On May 7, 2026, OpenAI introduced GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper. OpenAI says the new stack can reason in conversation, translate from 70+ input languages into 13 output languages, and transcribe speech live as someone talks.
The important change is not just lower latency. It is that voice is being pitched as an action surface, not a cosmetic wrapper. A capable realtime stack pushes zero-human company design further into customer support, sales intake, concierge work, and multilingual operations.
That sharpens the thesis in our earlier notes on Sierra and VibeVoice: the voice layer is becoming operational infrastructure.
5. The Geographic Read
Compared with our May 25 and May 23briefings, today's freshest signals are less distributed across Europe and Asia. The newest moves are concentrated in the North American software stack: software agents, cloud guardrails, orchestration surfaces, and voice interfaces.
That is still a global story. It suggests the current bottleneck is not only raw model intelligence. It is packaging agent work into repeatable operating systems that enterprises can actually buy.
6. What Changed Since Our Last Coverage
Our previous briefings argued that the category was widening from models into infrastructure, standards, and domain-specific execution. This week adds more institutional conviction.
Investors are assigning software-engineering-scale multiples to autonomous coding. Cloud vendors are converting best-practice scaffolding into official product layers. Agent workbenches are becoming cross-harness control planes. And voice models are moving closer to directly handling revenue-adjacent and support-adjacent tasks.
That is a stronger operating stack than we had even two weeks ago.
Related: See our prior research on the May 25 briefing, deployment companies, managed agents, AI gateways, Sierra, and VibeVoice.