Every time a user reports a fake listing or a suspicious message, a clock starts ticking. A human investigator has to pull data from a dozen different logs, cross-reference policies, and make a high-stakes call, often with a quota of hundreds of cases per day. It is a job defined by cognitive overload and burnout, and Suhas Manangi spent years inside the teams that do it at Snap, Airbnb, and Amazon. His new company, Precognition Labs, is betting that the only way to scale this work is to stop asking humans to do most of it.
Founded in 2025, the Seattle-based startup is building what it calls AI Agentic Systems to automate the investigative workflows for Trust & Safety, fraud, and compliance teams. In early pilots with online marketplaces, the company claims its tools handled roughly 90% of the investigation process, from evidence gathering to a recommended decision, while maintaining the team's expected accuracy [Ascent Valley, Unknown]. The remaining 10%, presumably the most ambiguous or severe cases, are flagged for human review. It is a classic automation play, but applied to one of the most sensitive and legally fraught corners of a digital platform.
The wedge after the detection
The landscape of online safety tools is crowded with companies that detect bad behavior: content moderation models, fraud scoring APIs, and risk signals. Precognition Labs is aiming for the step that comes next. After a detection model flags an account or a transaction, someone still has to investigate it. This is the manual, time-sucking bottleneck where policies meet messy reality.
The company's initial product suite, all bearing vaguely prescient names, tackles different parts of this post-detection workflow.
- PreCog Dash is designed to connect disparate data sources,user logs, transaction records, chat histories,into a single case file for an investigator or an AI agent to review [Preqin, 2026].
- PreCog Arthur focuses on real-time quality assurance, presumably checking the consistency and accuracy of decisions made by either automated systems or human reviewers [Preqin, 2026].
- PreCog Agatha allows teams to run controlled A/B tests on their investigative workflows and policy changes, a nod to the fact that platform rules are constantly evolving [Preqin, 2026].
The bet is that by automating the rote parts of evidence synthesis and initial assessment, teams can reallocate their most expensive resource,human judgment,to the cases that truly need it.
Operators from the integrity front lines
The company's credibility rests heavily on its founders' resumes, which read like a who's who of platform integrity at scale. Suhas Manangi, the CEO, is a former Trust & Safety and AI product leader with experience at Snap, Airbnb, Amazon, Lyft, and Microsoft [precognitionlabs.ai, 2026]. His co-founder and CTO, Xiuduan Fang, joined in early 2026 and brings over 17 years of experience building machine learning and enforcement systems at scale, with a background at Google and Snap [marketplacerisk.com, 2026] [LinkedIn, Jan 2026].
This isn't a team of academics theorizing about content policy. They are operators who have lived inside the teams they are now selling to, which explains their focus on pragmatic workflow tools over abstract AI promises. Manangi has been actively building this reputation, appearing as a keynote speaker at industry conferences like Hardwear.io and VIEW Conference in 2026 [hardwear.io, 2026] [VIEW Conference, 2026].
| Founder | Role | Key Prior Experience |
|---|---|---|
| Suhas Manangi | Co-Founder & CEO | Trust & Safety/AI product at Snap, Airbnb, Amazon, Lyft, Microsoft [precognitionlabs.ai, 2026] |
| Xiuduan Fang | Co-Founder & CTO | Machine learning and integrity systems at Google, Snap [marketplacerisk.com, 2026] |
Early traction and the path to paid pilots
Precognition Labs is in its earliest days, having raised a $200,000 pre-seed round from venture builder Antler in late 2025 [Preqin, Unknown]. The company reports having an additional $300,000 in capital committed for a forthcoming round [Ascent Valley, Unknown]. Its current traction is based on successful pilots rather than publicly named enterprise contracts, with a focus on online marketplaces and digital platforms [Ascent Valley, Unknown]. The business model is subscription-based, with annual or monthly fees for access to its tools [Preqin, Unknown].
The company is also hiring, with an open role for a Principal AI/LLM Researcher, signaling an intent to build proprietary depth beyond just integrating off-the-shelf models [shine.com, 2026].
Where the wheels could come off
The risks here are not subtle. Selling into Trust & Safety teams means navigating long enterprise sales cycles, deep integration requirements, and an extreme aversion to risk. A single high-profile automation error,a wrongful ban of a legitimate seller, a missed child safety report,could crater a platform's trust and a vendor's reputation. The 90% automation claim from early pilots is impressive, but it will need to hold up under the scrutiny of a full-scale deployment with more varied and adversarial edge cases.
Furthermore, the competitive landscape, while not named in the company's sources, is implicit. Large platforms like Meta and Google have built vast internal tools for this work. Other AI automation startups are certainly eyeing the same operational pain point. Precognition Labs' answer is its founders' domain expertise: they are not selling generic workflow automation, but a system built by people who know the specific contours of a safety investigator's day.
Their most plausible path is to become the trusted specialist for mid-market platforms and fast-growing marketplaces that lack the resources to build these complex systems in-house but face the same scaling pressures as the giants.
The next twelve months
The immediate milestones are clear: convert pilot engagements into paid contracts, deploy the additional $300,000 in committed capital, and land a first marquee customer willing to be named. The founders' conference circuit appearances suggest a parallel track of building industry credibility to support both sales and fundraising.
A back-of-the-envelope calculation illustrates the operational stakes. If a mid-sized marketplace's safety team of 50 investigators handles 500 cases each per week, that's 1.3 million human decisions annually. Automating 90% of that workflow doesn't just mean reassigning 45 people. It means those 1.17 million decisions are made with consistent, auditable logic, freeing the team to focus on the 130,000 most complex cases. The unit economics shift from pure headcount cost to a blend of software subscription and elevated human expertise.
For Precognition Labs to succeed, it must prove its systems are not just faster, but more reliably accurate than the overburdened human processes they replace. Its true competition isn't another startup. It's the entrenched, manual status quo inside every platform's integrity ops center,a bottleneck that scales linearly with pain.
Sources
- [Ascent Valley, Unknown] Precognition Labs startup profile | https://www.ascentvalley.com/startup/precognition-labs
- [Preqin, 2026] Precognition Labs company profile | https://www.preqin.com/academy/glossary/company/precognition-labs
- [precognitionlabs.ai, 2026] Precognition Labs website | https://www.precognitionlabs.ai
- [marketplacerisk.com, 2026] Xiuduan Fang profile | https://www.marketplacerisk.com/team/xiuduan-fang
- [LinkedIn, Jan 2026] Xiuduan Fang announcement | https://www.linkedin.com/in/xiuduan-fang
- [hardwear.io, 2026] Keynote Fireside Chat at Hardwear.io 2026 | https://hardwear.io/talks/keynote-fireside-chat/
- [VIEW Conference, 2026] Fireside Chat Keynote at VIEW Conference 2026 | https://www.viewconference.it/article/1273/fireside-chat-keynote
- [shine.com, 2026] Principal AI/LLM Researcher job posting | https://www.shine.com/jobs/principal-ai-llm-researcher/precognition-labs/18662082
- [Preqin, Unknown] Precognition Labs pre-seed funding details | https://www.preqin.com/academy/glossary/company/precognition-labs