Skip to content
0MESubscribe
SaaS17 Nov 20262 min read

Pretxt Raises $5M to Police Autonomous AI Agents in Real Time

Startup Pretxt AI launched with $5 million in seed funding to monitor enterprise AI agents at runtime and stop policy violations before mistakes happen.

Listen to this storyAbout 3 min, read by your device

Autonomous artificial intelligence agents are taking on more complex business tasks, such as modifying databases and processing customer refunds. However, these systems often skip workflow steps, hallucinate answers, or make unauthorized decisions. Startup Pretxt AI emerged with $5 million in seed funding to fix these risks through a continuous oversight platform.

Deciens Capital led the funding round, with backing from Howdy Partners, Service Provider Capital, and NSV Wolf Capital. Pretxt was founded by Chief Executive Brian Soby and Chief Technology Officer Aaron Costello, who previously worked together at AppOmni.

Intervening at Runtime

Pretxt monitors autonomous agentsAutonomous agentsAutonomous agents are software programs designed to make decisions and perform tasks independently. Open in glossary while they operate, checking their behavior against corporate rules and policies. Traditional oversight tools often block agents completely or flag errors after the fact. Other approaches rely on a secondary large language modelLarge language modelAn AI program trained on huge amounts of text so it can read and write like a person. Open in glossary to evaluate the primary model, which adds latencyLatencyThe delay between a request being sent and a response being received. Open in glossary and computational expense.

Instead of using another model as a judge, Pretxt relies on a near-deterministic engine. The company claims this system evaluates agent actions in milliseconds, making it up to 60 times faster than model-based judges. In tests using Zapier’s AutomationBench benchmarkBenchmarkA standard test used to compare AI models or chips. Open in glossary, the software cut policy violations from roughly 20% down to zero.

Human Oversight for Edge Cases

The platform lets administrators set rules in plain language without needing to dissect the underlying models. The software continuously tracks approvals, lets teams roll back plans, and halts tasks when an agent encounters unexpected edge casesEdge casesUnusual or extreme situations that happen outside normal operating conditions. Open in glossary.

  • In financial services, the software enforces borrower verification and directs complicated insurance claims to human workers.
  • In healthcare, it stops automated agents from exceeding regulated guidelines.
  • In customer service, the system blocks agents from handing out unapproved refunds or discounts.

Pretxt has launched its platform in early access, focusing primarily on highly regulated sectors where errors can cause legal or financial damage.

How does this story land?

One tap per reader.

Quick poll

Can enterprise AI agents be trusted with financial tasks without human approval?

One tap. No sign-up.

Ask this story

Still have a question?

Answers use only what is in this story. They are written by AI and can be wrong.

Sources

  1. SiliconANGLE, With $5M in funding, Pretxt launches runtime oversight layer to supervise AI agents in real time
Up next in SaaSTalkdesk Targets Pharma to Speed Up Patient Care Instead of Just Cutting Call Costs2 min read

Keep reading

All SaaS stories