$65 Million Just to Make AI Agents Safe. That Should Tell You Something.

$65 Million to Solve One Problem: Trust

On March 30, Sycamore Labs — a Palo Alto startup led by former Atlassian CTO Sri Viswanath — announced a $65 million seed round to build what it calls the "agentic operating system" for the enterprise. The round was led by Coatue and Lightspeed Venture Partners, with participation from Dell Technologies Capital and 8VC. The angel investor list reads like a Silicon Valley power directory: Databricks CEO Ali Ghodsi, former OpenAI Chief Scientist Bob McGrew, Intel CEO Lip Bu-Tan, and AI researcher François Chollet.

Sycamore is not building another AI model. It is not building another chatbot framework. It is building the governance, monitoring, and permission infrastructure that enterprises need before they will let AI agents touch production systems.

Sixty-five million dollars. In a seed round. Just to solve the trust problem.

That number should tell every business owner something important about where AI agents actually stand in 2026. The capability is here. The models work. What is missing — and what the largest venture firms in the world are betting hundreds of millions to solve — is the layer between "the agent can do the work" and "we trust the agent to do the work without breaking something."

Sycamore's pitch makes this explicit. Their platform uses a tiered trust system where agents "earn" autonomy by proving reliability over time. New agents are heavily monitored. As they demonstrate consistent results, they get more freedom. The system captures institutional knowledge as agents interact with company workflows, and it enforces governance policies structurally — not through prompts, not through hopes, but through infrastructure.

This is the right architecture. And the fact that it takes $65 million and the former CTO of a $50 billion enterprise software company to build it tells you exactly how hard this problem is.

The Experimentation Trap

The same week Sycamore announced its round, new research compiled by Salt Creative from LinkedIn, Salesforce, and HubSpot data painted a stark picture of where small businesses actually stand on AI adoption.

The headline: 57% of U.S. small businesses using AI are still in experimentation mode.

Not production. Not systematic deployment. Experimentation. They are testing tools, running pilots, trying things out. More than half of the businesses that have adopted AI at all have not moved past the "let's see if this works" phase.

The research uncovered other numbers that should be uncomfortable:

• 67% of businesses report their data is not properly structured for AI use
• 70% of employers provide no AI training whatsoever
• 76% of marketers using AI apply it exclusively to content creation and copywriting

That last number is revealing. Three out of four businesses using AI for marketing are using it to generate blog posts and social media copy. Content generation is the shallowest possible AI use case. It is the equivalent of buying a CNC machine and using it as a paperweight. The tool can handle scheduling, customer communication, reporting, operations coordination, and workflow automation. Most businesses are using it to write Instagram captions.

The data gap is even more telling. Two-thirds of businesses say their data is not ready for AI. That is not a technology problem. An AI agent can connect to virtually any data source with the right integration work. The real problem is that nobody has done the integration work. Nobody has mapped the business's data into a format an agent can act on. Nobody has decided which systems the agent should access, what permissions it should have, or what guardrails should exist around its behavior.

This is the governance problem. The same problem Sycamore just raised $65 million to solve for enterprises. Except enterprises have IT teams and security budgets and compliance departments to build governance frameworks internally, even if slowly. Small businesses have none of that. They have an owner who is already working 60 hours a week, a handful of tools they barely have time to configure, and an AI vendor who sold them on "set it up in 20 minutes."

Why the Enterprise Solution Will Not Trickle Down

When investors pour $65 million into Sycamore, they are betting on enterprise contracts. Fortune 500 companies. Organizations with thousands of employees and AI budgets measured in millions per quarter. That is where Sycamore's "agentic operating system" will deploy first.

The Forbes Tech Council published a piece on March 30 arguing that AI is narrowing the gap between small and large organizations — not just in access to technology, but in access to knowledge, standards, and analysis. That is true at the capability layer. A small business can access the same Claude or GPT models that a Fortune 500 company uses. The models do not charge more because the business is smaller.

But the governance layer is an entirely different story. Sycamore's product is being built for organizations with dedicated IT teams who can configure agent policies, monitor dashboards, and adjust trust tiers. The pricing, the complexity, the integration requirements — all of it will be calibrated for enterprise buyers.

This is not a criticism of Sycamore. They are solving a real problem for their target market. But their target market is not a 12-person accounting firm or a regional restaurant group or a mobile food vendor. Their target market is companies that already have a Chief Information Security Officer.

The governance gap is not going to close because enterprise tools eventually get cheaper. It is going to close because someone builds governance into the deployment itself — baked into the way the agent is configured, monitored, and operated from day one.

That is a fundamentally different product than what Sycamore is building. Sycamore is a managed layer that sits above the agents and manages them on the enterprise's behalf. What small businesses need instead is a platform where that governance layer is built in from the start — configured by the business itself, not operated by someone else on their behalf.

What "Earning Trust" Actually Looks Like at SMB Scale

Sycamore's tiered trust model is architecturally sound. The idea that an agent starts with limited autonomy and earns more as it proves reliability makes sense. It maps to how you would manage a new employee. You do not hand someone the company credit card on their first day. You give them small tasks, verify the results, and gradually expand their scope.

The problem is that executing a tiered trust model requires someone to do the verifying. Someone has to review the agent's early work. Someone has to decide when it has earned more autonomy. Someone has to monitor for the moment reliability starts to drift — because the Princeton research we covered in our analysis of the Amazon Kiro outage showed that AI reliability improves at one-seventh the rate of capability. Agents get smarter far faster than they get safer.

In an enterprise, that "someone" is a dedicated team. At a small business, that "someone" is the owner — who is also doing sales, managing employees, handling customer issues, and trying to close the books before the end of the month.

This is why the 57% experimentation figure is so persistent. Small businesses try an AI tool. It works for a while. Then something goes sideways — an email goes out wrong, a schedule gets botched, a report has bad numbers — and the owner loses confidence. They do not have the time or expertise to diagnose what went wrong, reconfigure the agent, and set up monitoring to prevent it from happening again. So the agent goes back to being a content generator. Or it gets turned off entirely.

The governance gap is not a technology gap. It is a capacity gap. The business owner does not lack access to AI. They lack the operational infrastructure to use AI safely at any depth beyond surface-level content generation.

The Real Structural Advantage SMBs Are Missing

Here is what gets lost in the conversation about AI governance. The research Salt Creative compiled identified a structural advantage small businesses have over enterprises: the ability to pick a specific use case, implement quickly, and iterate without the procurement cycles, legacy integrations, and stakeholder alignment that slow large organizations down.

That advantage is real. A 10-person business can go from "let's automate appointment scheduling" to "the agent is handling all appointment scheduling" in a matter of weeks. A 10,000-person enterprise might take a year to get through procurement, security review, legal review, and change management.

But speed of implementation without governance is not an advantage. It is a liability. The small business moves fast because nobody is checking whether the agent's permissions are scoped correctly, whether escalation paths exist for edge cases, whether the agent's behavior is being monitored for drift, or whether a single point of failure has been introduced into a critical business process.

The enterprise moves slowly because all of those checks exist — or at least, the organizational awareness that those checks should exist. Sycamore's entire business is built on the premise that even enterprises need better infrastructure for those checks. Small businesses need them even more, and have even less capacity to build them.

The LinkedIn data in the same research found that half of U.S. small business owners said the rise of AI inspired them to consider entrepreneurship as a viable path for the first time. That is a remarkable statistic. AI is lowering the barrier to starting and running a business in ways that are genuinely transformative.

But lowering the barrier to starting a business while leaving the barrier to operating AI safely at its current height creates a predictable outcome: a wave of new businesses that adopt AI tools early, run into governance failures they do not have the capacity to resolve, and either retreat to manual processes or accumulate silent operational risk.

What Governance Looks Like When You Cannot Build It Yourself

The honest answer for most small businesses is that they should not be building AI governance infrastructure. Not because they are incapable, but because it is not a good use of their time. A restaurant owner should not be configuring agent permission boundaries. An accounting firm should not be designing escalation frameworks. A permitting company should not be monitoring model drift.

These are all real operational requirements for running AI agents safely. They are also deeply technical and constantly evolving. Every model update from Anthropic, every platform release from OpenClaw, every shift in API behavior changes the governance surface area. Keeping up with it is a job, not a side task.

This is the gap Associates AI Teammates exists to fill. It is a self-serve platform that builds the governance layer into every Teammate you configure — not as an add-on, not as a premium tier, but as a fundamental part of how a Teammate is set up.

What that looks like in practice:

Permission scoping from day one. Every Teammate you configure is scoped with the minimum access it needs to do its job by default. A scheduling Teammate gets access to the calendar system and the customer database. It does not get access to financial records, email accounts, or internal documents unless you deliberately grant it. Expanding scope is always a specific, visible decision — not a default.

Structured escalation, not vague prompts. The platform does not rely on telling a Teammate to "ask for help when unsure" and hoping for the best. You define which high-risk actions route to a human for approval, which customer-facing communications go through review, and which system changes require sign-off. Those safeguards are explicit in the configuration, not left to the agent to reliably self-recognize.

Continuous monitoring for drift built into the platform. Teammates run on infrastructure designed to surface output pattern changes over time — an agent that was 95% accurate on invoice processing in January but slips to 88% in March because upstream data changed or a model update shifted behavior. That visibility is a platform feature, not something you have to build yourself.

Model-agnostic by design. Associates AI Teammates aren't locked to one model provider. When a model updates, you're not stuck waiting on someone else to test and roll it out — you control which model each Teammate runs on and can validate changes on your own timeline.

This is the difference between having an AI tool and having a governed AI platform. The tool gives you capability. The platform gives you the governance primitives an SMB actually needs — oversight, escalation, testing, and operational control — configured by you, without requiring the enterprise governance stack Sycamore is building or a team of specialists to run it for you.

The $65 Million Signal

Venture capital tells you where the market's problems are before the market admits them. When Coatue, Lightspeed, and a roster of Silicon Valley luminaries put $65 million into a seed-stage company whose entire product is AI agent governance, they are making a statement: the governance problem is real, it is unsolved, and it is worth billions.

They are right. But they are solving it for the wrong end of the market — or at least, only one end. The enterprise governance problem will get addressed by Sycamore, by competing platforms, by the hyperscalers who will inevitably build their own versions. Fortune 500 companies will have options. They will have budgets. They will figure it out.

The small business governance problem is a different animal. It requires a different solution: a platform the business owner can actually configure without a dedicated technical team — permission scoping, escalation, and monitoring built in as defaults, not features you have to engineer yourself.

The 57% of small businesses stuck in experimentation mode are not stuck because the AI tools are insufficient. The tools are extraordinary. They are stuck because nobody has built the operational layer that makes those tools safe to run in production on real business processes.

That operational layer is what Associates AI Teammates gives you as a self-serve platform. It is what turns an AI experiment into a business asset.

Frequently Asked Questions

What is Sycamore Labs and why does their funding matter?

Sycamore Labs is a startup founded by former Atlassian CTO Sri Viswanath that raised $65 million in seed funding in March 2026 to build an "agentic operating system" for enterprises. The round was led by Coatue and Lightspeed, with backing from leaders at Databricks, OpenAI, Intel, and Palo Alto Networks. The size of the round — $65 million just for governance infrastructure — signals that the AI industry recognizes agent governance as one of the largest unsolved problems in the market. If the trust and safety layer for AI agents were simple, it would not command this level of investment.

Why are 57% of small businesses stuck in AI experimentation mode?

Research compiled from LinkedIn, Salesforce, and HubSpot data shows that most small businesses using AI have not progressed beyond testing and piloting. The primary barriers are unstructured data (67% of businesses report their data is not ready for AI), lack of training (70% of employers provide no AI training), and absence of governance infrastructure. Small businesses can access the same AI models as enterprises, but they lack the operational capacity to deploy those models safely on real business workflows.

What is the difference between AI governance for enterprises and small businesses?

Enterprise AI governance involves dedicated teams, compliance departments, and platforms like Sycamore that provide monitoring dashboards, trust tiers, and policy enforcement. Small business AI governance requires the same principles — permission scoping, structured escalation, drift monitoring, and model update testing — but delivered as configurable defaults on a self-serve platform instead of a custom build only a dedicated team could maintain.

How does a self-serve platform like Associates AI Teammates handle the governance problem?

Associates AI Teammates builds governance into every Teammate from day one as configurable defaults: minimum-necessary permissions, structural escalation triggers for high-stakes decisions, drift monitoring on agent output over time, and model-agnostic flexibility so you're never locked into testing one vendor's updates on their timeline. The business owner directs the Teammate's work and configures its boundaries; the platform provides the governance and technical infrastructure underneath.

Can small businesses safely deploy AI agents without enterprise-grade governance?

Yes, but only with the right structure. The governance does not need to be enterprise-scale — you do not need a Chief Information Security Officer or a 335-system compliance audit. You do need permission boundaries, human review at critical decision points, monitoring for output quality over time, and someone managing model updates and platform changes. The question is whether you build that structure yourself or partner with a service that provides it. For most small businesses, building it internally requires technical expertise and ongoing time investment that is better spent on the business itself.