$950M Says Agentic AI Is Infrastructure Now. Open Source Is Ready to Prove It.

Sierra's $950M raise signals enterprise agentic AI has crossed from experiment to infrastructure. Mistral's cloud coding agents and OpenHands are already positioning for what comes next.

The money has spoken. Sierra, Bret Taylor's AI agent startup, raised $950M at a $15.8B valuation according to CNBC — and that number isn't speculative. Enterprise customers are already deploying agents at scale and paying for it. The practical question has shifted from "should we build agents" to "which infrastructure do we trust."

The proprietary layer is moving fast. Mistral AI launched cloud-based remote coding agents powered by its Medium 3.5 model (as of May 2026), shifting from local execution to distributed cloud infrastructure in direct competition with OpenAI. The move adds a credible European alternative to a space with meaningful data-sovereignty implications — particularly for enterprise buyers operating under GDPR or sector-specific compliance regimes that make routing code through US infrastructure legally complicated.

On the open-source side, OpenHands is gaining visibility as a model-agnostic platform for coding agents — SDK, CLI, and web GUI in one package. Its SDK lets developers swap out the chain-of-thought strategy without rebuilding surrounding infrastructure, which is the kind of flexibility proprietary platforms structurally can't match. Meanwhile, a small but telling signal: Simon Willison observed that Bun's experimental PORTING.md reads like documentation written for coding agents rather than human contributors — a sign that agentic code contribution is already reshaping how serious open-source projects think about their own internal documentation.

For builders, the orchestration layer decision is the one to make now. LangGraph's graph-based state model handles complex multi-agent flows where simple loops fail — conditional re-entry, self-critique cycles, branching logic that needs persistent state. OpenHands provides the modular SDK layer for coding-specific automation. These aren't competing choices; teams building serious agent workflows will likely reach for both. For buyers evaluating enterprise vendors, Sierra's valuation signals durable demand — the real due diligence question is whether a managed platform's convenience outweighs what open-source control provides over time. For non-technical leaders, the analog that holds: this is roughly where cloud infrastructure sat in 2012. The categories are clear; the cost of waiting to pick a stack is rising faster than the cost of picking one.

The bottom line: Evaluate OpenHands and LangGraph this month if your team hasn't chosen an agent orchestration stack — enterprise lock-in costs in this space are already becoming real.

Sources