← All stories
Agentic AI Security & GovernanceAutonomous AgentsFounder StoryPortland Startups

30 Days In: Building the Control Layer for Agentic AI

An inside look at Auspere's first 30 days: the governance gap in agentic AI, the data behind it, and how we're building the control layer to close it.

Auspere · September 8, 2026 · 4 min read

Where We Started

I'm Loren, CEO and cofounder of Auspere. Thirty days ago I joined my technical cofounder full time to bring what he'd spent the last eighteen months building in stealth out into the world: security and governance for agentic AI. Before this, I spent fifteen years in B2B SaaS sales leadership, working at other Portland startups along the way (Cloudability and GreenBits which have since exited to other companies). Most recently, I helped Docker scale from $12M to over $150M in ARR, first as an individual contributor and then as the SMB Sales leader who built and scaled a team to a $1M quarterly run rate. I've joined Auspere to do the same thing here: get something real in front of the people who need it.

The last 30 days have been full of the kinds of milestones that only founders really appreciate. We set up the company, launched the site, and put some roots down in Portland, Oregon. There's a particular kind of hope in that, for what comes next.

What matters more than any of those steps is what's starting to happen now.

We're already talking with people who run autonomous agents in production. It's early. But we know who belongs in this pilot: teams with multiple agents already sprawling across their organization, and security folks who want to work alongside us on implementation, not just sit through a demo. That's the design partner relationship we're building.

This isn't a dashboard bolted onto the outside of your agent stack, and it isn't another layer of visibility that just tells you what already happened after the fact. It's a control layer, a gateway viewing the path of every request an agent makes. 

What makes it different is that it's canonical across every stage of that path, from the moment something is asked to the moment it's proven. One architecture, the same chain of custody, start to finish. That continuity is what's unique about how we've built this.

The foundation is ready for its next phase. We're looking for the right group of design partners to help shape it: teams who want to pressure test the platform in real environments and give direct feedback on its ability to secure autonomous agents and produce auditable, exportable records of their behavior.

What's Next

The reality is most teams experimenting with agentic AI today don't have a clear answer to some very basic questions.

What exactly is this agent allowed to do?

What did it actually do?

And how do you prove that to someone else when it matters?

Those answers can't come from the same system that's taking the action. They require an independent control layer, one that sits outside the model, enforces policy in real time, and produces records you can actually rely on when something goes wrong. That's the layer we're building.

The Governance Gap, By The Numbers

If your team is actively working with AI agents, or even seriously evaluating them, you're already on a path where "we'll figure out governance later" stops being a viable strategy. This isn't a hunch. The data backs it up. 83% of organizations now have more AI agents than human users, but only 21% of them actually govern those agents, according to JumpCloud research published last month.

McKinsey's most recent State of AI survey found organizations getting the most value from AI managed agent risk more, not less, than everyone else: 45% of these high performers are actively mitigating unauthorized or unintended agent action, compared to 33% of other organizations. The same survey found that 32% of companies have already skipped buying a tool because they could build it in house with agentic coding tools, quietly adding more custom, ungoverned agents outside of any procurement review that might have caught them.

Even the frontier labs are running into this.  OpenAI and Anthropic each disclosed a cybersecurity incident this summer involving an AI agent that went rogue, and neither has published what actually happened or what stopped it. That's the coverage gap in one sentence: real incidents, no public evidence trail.

And this is solvable. A recent academic paper on policy enforcement for agentic systems found that continuous, per action authorization stopped or corrected 94.8% of rule breaking actions while still completing 86.9% of legitimate ones. Most teams don't have that layer in place yet.

We're working with design partners who want to close that gap early, before it becomes a blocker, or worse, a liability. If that's you, we should talk. Not to walk through a deck, but to put this into a real environment and see how it holds up under pressure.

From the bottom of our hearts, thank you for taking time to read about our story and what we’re building at Auspere. We care deeply about the future of technology and are committed to making our world a better place.

L.R.

Keep exploring ideas for a more accountable AI future.

Back to the journal →