Sri Viswanath spent years watching enterprises build AI agent pilots that worked beautifully in demos and fell apart in production. The agents did not fail technically. They failed politically: no one owned the policy layer, no one could explain why the same agent handled one workflow correctly and catastrophically misbehaved in another, and trust was assumed rather than engineered.
That observation is the core premise of Sycamore, the company Viswanath launched after leaving his role as a partner at Coatue. Sycamore has raised $65 million in seed funding, co-led by Coatue and Lightspeed Venture Partners, to build what it calls an operating system for autonomous enterprise AI — not a chatbot wrapper, but a governance and orchestration platform that treats trust as a designed feature rather than a retroactive patch. The angel list reads like a survey of enterprise infrastructure gravity: Bob McGrew, former chief research officer at OpenAI; Ali Ghodsi, CEO of Databricks; Lip-Bu Tan, CEO of Intel; Frederic Kerrest, co-founder of Okta; and François Chollet, co-founder of Ndea and a widely-cited AI researcher. Additional institutional backing came from Abstract Ventures, Dell Technologies Capital, 8VC, Fellows Fund, and E14 Fund. TechCrunch reported additional angels including Bipul Sinha, CEO of Rubrik; Mike Knoop, co-founder of Zapier and Ndea; and Jerry Tworek, former research scientist at OpenAI.
"We've gotten pretty far with AI agents in siloed environments," Viswanath said in the company's founding blog post. "They work fine in demos. But they lack a centralized portal that allows humans to make sure they are following company policies, staying within security boundaries, and learning from their mistakes." The $65 million is meant to close that gap, funding engineering hiring, enterprise deployments, and R&D on trust architectures, memory systems, and multi-agent coordination.
The operational gravity problem
Thomas Laffont, co-founder of Coatue, told SiliconANGLE that enterprises have already gotten far with AI agents in siloed environments and run impressive demos. But those agents lack a centralized portal that lets humans enforce company policies, stay within security boundaries, and learn from mistakes. Viswanath calls this the "operational gravity" problem — the point at which agents are doing enough real work that you need a control plane. Without one, the agent layer becomes a liability rather than an asset.
Sycamore's answer is a three-layer architecture that treats governance as structural, not additive. The first layer is the Agent Operating System — the kernel that watches every decision an agent makes, tracks what is working and what is not, and continuously rewrites code on the fly with continuous integration and continuous deployment (CI/CD) discipline applied to agent behavior rather than just application code. The second is the Adaptive Factory: a natural-language-driven engine that generates production-ready apps, integrations, and agents from intent rather than specifications. The third is Collective Intelligence, which connects the agent layer to organizational knowledge so agents do not repeat mistakes or ignore institutional context.
The design philosophy is embedded in what Sycamore calls the Progressive Trust System: agents earn autonomy through demonstrated reliability, moving from observation to action with every operation isolated, auditable, and governed from the start. This is architecturally different from most agent frameworks, which ship capability first and add guardrails as an afterthought.
The enterprise software diagnosis
Sycamore's company blog lays out a three-part critique of why existing enterprise software fails in the agent era: software decays without continuous maintenance, SaaS has become bloated with point solutions that do not integrate, and the consulting model creates dependency rather than capability. The implicit argument is that an agent-native OS short-circuits all three — agents maintain themselves, the OS integrates across tools, and natural-language intent replaces the consulting dependency.
Laffont called Sycamore a "Big F Idea" — Coatue's term for a market that expands the entire category rather than taking share from incumbents. "Every enterprise system today is built for humans doing the work," Laffont said. "The next generation of enterprise software will be autonomous, continuously learning, and adaptive. Sycamore is building the operating system for that future."
That framing echoes what the hyperscalers are building toward with their own agent runtimes. Microsoft Azure Foundry, AWS Bedrock AgentCore, and Google Cloud's agent development environment all offer lifecycle management for AI agents. Sycamore's bet is that enterprises need a vendor-neutral control plane rather than one tied to a specific model's inference layer.
Who this is for — and the honest skepticism
Sycamore is already working with Fortune 500 enterprises as design partners deploying autonomous AI agents in production, according to its company blog — meaning the architecture is being load-tested against real workloads, not just internal demos. The team includes researchers from Stanford and Cornell and engineers from Meta, Google, and Atlassian. Viswanath himself scaled Atlassian's engineering org from roughly 800 to nearly 4,000 people during his tenure, and argues that governance as a feature — not a constraint — is what allowed that scale.
The competitive set is crowded. OpenAI, Anthropic, Microsoft, AWS, and Google all have agent runtimes with governance features. Startups including Port, Airia, and Isara are attacking the agent governance problem from different angles. The integration challenge — ensuring AI agents work with the legacy systems modern enterprises run on — is where most of these plays will either prove out or stall.
The honest question is whether "agent operating system" is a real category or a repackaging of what the hyperscalers are already building. Viswanath's track record at Atlassian is the strongest argument that Sycamore's approach is differentiated. The $65 million seed buys roughly 18 months to prove it.
What to watch: whether the Fortune 500 design partners go public with deployments, and whether the Progressive Trust System's architectural choices show up in any open-source components or developer-facing tooling. If the governance layer becomes a commodity, Sycamore's bet is that execution and enterprise relationships carry it. If it does not, the hyperscalers will have absorbed the category by default.
