Pretxt emerged from stealth on Nov. 17, 2026 with a $5M seed round led by Deciens Capital to supervise autonomous AI agents at runtime in enterprises.
Key Takeaways
- Pretxt raised a $5M seed round led by Deciens Capital; valuation was not disclosed.
- Howdy Partners, Service Provider Capital and NSV Wolf Capital also joined the round.
- The product is in early access and targets financial services, insurance and healthcare.
Lead
Pretxt, a startup based in Aspen, Colorado, with a presence in Dublin, Ireland, announced on Nov. 17, 2026 that it has raised $5M in seed funding and left stealth. Deciens Capital led the round. Howdy Partners, Service Provider Capital and NSV Wolf Capital participated. The company did not disclose a valuation.
Pretxt sells what it calls AI supervision, a runtime layer that watches autonomous agents as they act and corrects them when they stray from company rules. The platform is in early access. The company was co-founded earlier this year by security veterans Brian Soby and Aaron Costello.
Who Is Behind Pretxt?
Brian Soby is CEO and Aaron Costello is CTO, and both come from enterprise security. Soby held senior security roles at Salesforce and Mitre and founded the SaaS security company AppOmni. Costello met Soby at AppOmni and later found significant vulnerabilities in AI agents at Salesforce and ServiceNow.
That pedigree shapes the pitch. The founders have spent years examining how agents with broad permissions can be pushed outside their intended scope, and they are now selling the control layer they say was missing.
What Does the Pretxt Platform Actually Do?
Pretxt evaluates an agent's behavior against operating standards written in plain language, and it does so in milliseconds. The company describes the engine as near-deterministic and says it does not use a large language model as the judge. It claims the approach runs 60 times faster than LLM-as-a-judge methods.
The system steers rather than kills. When an agent attempts something outside policy, Pretxt intercepts the action and redirects the agent toward permitted behavior. Costello gave the example of a medical practice's customer service agent that books appointments but must never offer medical advice. Pretxt would catch the attempt and redirect it.
Customers can define policies, track approvals, roll back versions and escalate edge cases to human reviewers. Costello says the people setting rules should not need deep technical knowledge of every system an agent touches. The company reports that in tests on the Zapier AutomationBench, policy violations fell from roughly 20% to zero. That figure comes from Pretxt and has not been independently verified.
Why Does a $5M Seed Round Matter for AI Agent Oversight?
The round is modest, and the size says something about where this category sits. A $5M seed funds a small engineering team and early design partners, not a platform build-out. With no valuation disclosed, there is no public benchmark for what investors think a runtime supervision layer is worth.
The investor mix points to a security-first reading of the problem. Deciens Capital is known for early bets in security and infrastructure software, and Service Provider Capital's inclusion suggests a channel angle. Neither signals a broad market bet on agent governance as a standalone budget line.
The harder question is whether supervision stays a separate product. Cloud providers, identity vendors and agent-framework makers all have reasons to bundle policy enforcement into their own stacks. Pretxt will need to show that an independent layer catches failures that built-in guardrails miss.
Who Else Is Competing in AI Agent Governance?
Pretxt enters a field that already includes startups such as Outerlimit and Capsule Security, along with security incumbents extending existing products to cover agents. The category is young enough that vocabulary is unsettled. Vendors variously describe the same function as guardrails, runtime security, governance or, in Pretxt's case, supervision.
Pretxt's differentiation rests on two claims: speed without a model in the loop, and policies that non-engineers can write. Both are testable in production. Neither is proven by a benchmark run on a third-party automation suite.
What Comes Next for Pretxt?
The near-term job is converting early access into paying deployments. Pretxt is aiming at regulated sectors, including financial services, insurance, healthcare, life sciences and customer support, where an agent error carries compliance cost and buyers already have audit processes.
Customer names, pricing and headcount remain undisclosed. A follow-on round would likely depend on named enterprise deployments and evidence that the zero-violation result holds outside a benchmark.
Outlook
Pretxt launched with $5M from Deciens Capital and three other backers, a founding team with deep SaaS security experience and a product that intercepts and redirects agent behavior in real time. The early-access stage, undisclosed valuation and self-reported test results leave the commercial case open. The next signals to watch are customer deployments in regulated industries and whether larger platform vendors absorb the function.



