Startups & Funding

Pretxt raises $5M to supervise AI agents before they act

Pretxt emerged from stealth with $5 million in seed funding. Its system steers AI agents toward approved processes and cut policy violations to zero on Zapier's AutomationBench.

Pretxt raises USD $5 million to supervise AI agents
Pretxt raises USD $5 million to supervise AI agentsgeralt / pixabay
By Rebecca Stone5 min read

Why it matters

  • Pretxt emerged from stealth on November 20, 2026, with $5 million in seed funding.
  • Deciens Capital, Howdy Partners, NSV Wolf Capital and Service Provider Capital backed the round.
  • Pretxt says its system runs 60 times faster than alternatives that use an LLM as a judge.
  • On Zapier's AutomationBench, Pretxt says policy-violating actions fell from nearly one in five to zero.
  • Pretxt entered early access without disclosing a customer count.

Pretxt emerged from stealth on November 20, 2026, with $5 million in seed funding for a system that supervises AI agents while they carry out tasks and intervenes before an action reaches a customer or system.

Deciens Capital, Howdy Partners, NSV Wolf Capital and Service Provider Capital backed the round. Pretxt says it is introducing a new category of AI supervision.

Pretxt says businesses are already testing AI agents for customer support, financial services, marketing and other operational work. It argues existing security, governance and observability tools do not fully establish whether an agent is following an organisation's rules and procedures as it works.

Pretxt places the launch amid a broader debate over control once agents move beyond drafting text and begin acting inside enterprise software. It identifies issuing refunds, verifying customer details, processing claims and making commitments as examples of consequential work.

What does Pretxt's supervision system do?

Pretxt says its system evaluates agent behaviour in the context of a task and steers agents back on course when they deviate from approved processes. It can perform that intervention without human review and without using a large language model to judge each step.

According to Pretxt, that approach makes the system 60 times faster than alternatives that rely on an LLM as a judge. The company says it also avoids added latency, cost and inconsistent judgments.

Pretxt does not present the product as a replacement for security controls. It describes its layer as a check on whether agents follow company rules during execution and says it intervenes before an action reaches a customer or system.

How did Pretxt test its system?

In testing on Zapier's AutomationBench, which Pretxt described as representing real-world support workflows, the company says it reduced agent actions that violated company policy from nearly one in five to zero. It did not disclose customer numbers.

Pretxt says businesses define expected agent behaviour in plain language and can test supervision plans against their own examples before approving changes. The system also analyses behaviour over time to suggest refinements, subject to human review.

Pretxt did not disclose customer numbers, leaving the benchmark without a disclosed customer scale against which buyers could judge it. Both the 60-fold speed claim and the benchmark result originate with Pretxt, so enterprise buyers would need to test those claims in their own workflows before treating them as procurement benchmarks.

What gap does Pretxt say existing tools leave?

Brian Soby, Pretxt co-founder and chief executive officer, said:

You don't need a prompt injection attack for an AI agent to do something it shouldn't,

Security remains essential, but the next question is whether enterprises can trust agents to carry out real work according to their standards.

You can't supervise a workforce of AI agents with hope, a rulebook and reviews after the fact.

AI needs supervision while it acts.

That's the category Pretxt is here to establish,

Soby's position defines Pretxt's category claim: enterprise trust depends on whether agents act to company standards during work, not merely whether they resist an attack or receive a later review.

Chris Steffen, Vice President of Research at Enterprise Management Associates, framed the issue as one of accountability.

Enterprises have invested heavily in tools that ask whether an agent is secure, compliant or observable.

Far fewer can answer a more basic but critical question: Is the agent correctly doing the job it was given?

That's important because organizations that deploy AI own their outcomes, no matter who's asking: regulators, insurers or customers.

Most tools can only report what went wrong after the fact, when it's too late.

Pretxt intervenes while the outcome can still change to get agents back on track, creating a new category for AI supervision the market has been missing at the moment enterprises need it most,

Steffen's accountability framing matters because deploying organisations remain responsible to regulators, insurers and customers even when an agent performs the task. The distinction is between observing a failure and intervening before the outcome becomes final.

Pretxt argues that failures can happen even without an external attack. Its core question is not only whether an AI system is secure, but whether it performs a task in line with business expectations and regulatory obligations.

The distinction becomes material when an agent acts inside business systems. Pretxt says a procedural deviation can affect customers or the company even when no attacker caused it.

Why are investors backing runtime supervision?

Investors in the round said Pretxt targets a key constraint on wider AI deployment inside large organisations. As companies hand more tasks to agents, those investors say, teams face a trade-off between automation and control, especially in regulated sectors.

Nate Pacer, Partner at NSV Wolf Capital, said:

The ROI on AI agents is limited by how much real work a company is willing to hand them.

In our view, the entire AI revolution hinges on precisely this point.

Pretxt expands that limit.

When every action is held to an organization's own standards as it happens, teams can move agents from simple tasks to the work that drives revenue and carries real risk.

That's a large and growing market, and Pretxt is defining it,

Vishal Rana, Partner at Deciens Capital, said the investment case rests on runtime oversight becoming standard infrastructure:

AI agents are moving into workflows where mistakes have real consequences.

We invested in Pretxt because we believe supervision at runtime will become a foundational layer of enterprise AI infrastructure.

The opportunity is not just to identify a bad outcome after it happens, but to help organizations change the outcome while the work is still underway,

Taken together, the investors back Pretxt as infrastructure, not as another agent. Their thesis is that control over procedural actions will decide whether organisations delegate higher-value and higher-risk work.

Who is building Pretxt, and where will it start?

Pretxt was founded by Brian Soby and Aaron Costello, both of whom previously worked at AppOmni. The company says its wider team has experience at Salesforce, MITRE, SAP, Deloitte, VMware, HackerOne and AppOmni.

Its initial target sectors include:

  • Financial services
  • Healthcare
  • Insurance
  • Pharmaceuticals
  • Customer support

Pretxt says errors in those sectors can create financial, operational or compliance problems. The product is available through an early access programme.

Pretxt's early access programme must now show whether the reported zero policy violations on AutomationBench holds as regulated enterprises delegate work that carries financial and compliance consequences.

Source: itbrief.com.au

Share this article:

More from Rebecca Stone

Rebecca Stone

Show full bio

Correspondent covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.

802 articles

Related articles

  1. Uptiq launches GovernAI to govern AI agents in banking at no extra cost
  2. Australian firm opens the books on five months of AI agent controls
  3. Shepherd Brings Version Control to AI Agent Oversight
  4. OpenAI to Acquire Promptfoo and Build Red-Teaming Into Frontier
  5. Microsoft’s AI Transparency Initiative Targets Enterprise Agents

« Previous articleNext article »