Investors back regulated AI operators; Garry Tan publishes GBrain; YC profiles 33 collaboration and 44 conversational-AI startups
AI is increasingly being built into operating systems for recruitment, hospitality, payments, auditing and regulated medical businesses, while Garry Tan’s GBrain offers a concrete model for persistent, sourced agent memory. The developments also expose a central constraint: autonomy and scale are advancing alongside demands for licensing, human oversight, provenance and measurable evaluations.
Tuesday, Sep 22, 2026
Tracking: Y Combinator
Geography: San Francisco, Silicon Valley, United States
September Funding Shows AI Moving Into Regulated, Operational Businesses
On September 22, a wave of startup financings showed investors backing AI that operates inside regulated industries and physical systems, rather than only general-purpose models.
Paris medical-device company HighLife raised more than $90 million to expand its transcatheter mitral-valve business in Europe and advance a U.S. pivotal study; London payments company CellPoint added $34 million for AI-driven travel payments.
Early-stage deals extended the pattern.
Belgium’s Spott raised €18.3 million for an AI recruiting operating system, Dallas-based AI Hospitality Group secured $7.5 million to build a hotel-management company around autonomous back-office agents, and Norway’s Benford raised €5 million to become a licensed technology-driven audit firm.
Together, these deals suggest investors are rewarding regulatory permission, human oversight, and operational integration alongside software itself.
Key facts:
HighLife completed more than $90 million (€80 million), co-led by Andera Partners, Sofinnova Partners, Supernova Invest, and Mérieux Equity Partners.
CellPoint added $34 million as it deepened AI in travel payments.
Spott secured €18.3 million for an AI-native recruiting operating system.
AI Hospitality Group raised $7.5 million to build a hotel-management company around autonomous agents.
Benford raised €5 million at pre-seed to become a licensed technology-driven audit firm.
Why it matters: The deals point to a higher bar for AI startups. Model access remains useful, but licensing, clinical validation, data provenance, audit trails, human escalation, and domain expertise can determine whether software is trusted in consequential settings.
They also show more founders pursuing AI-enabled service businesses instead of selling software seats. That may expose companies to larger revenue pools, but it also gives them responsibility for operational outcomes.
The next test will be whether these startups can turn regulatory and technical barriers into scalable businesses without losing the oversight that makes their products viable.
Garry Tan publishes GBrain, an AI memory system for personal and company agents
Y Combinator CEO Garry Tan has published GBrain, a GitHub project he describes as the “production brain” behind his OpenClaw and Hermes AI agents.
The system ingests meetings, emails, social posts, voice calls, and ideas, then stores sourced facts that agents can retrieve, correct, withdraw, and share under access controls.
Tan says his deployment contains 155,795 pages, 24,589 people, 5,340 companies, and 66 autonomous cron jobs.
The project also reflects a broader position he has emphasized in recent writing: AI startups need rigorous evaluations, or repeatable tests of whether outputs are correct and improving, because “evals are emerging as the real moat.
” Together, the developments show Tan moving beyond general startup advice toward a concrete model for persistent, provenance-based AI infrastructure—an area that overlaps with YC’s company-brain startup theme.
Key facts:
Garry Tan identifies himself as YC’s president and CEO and GBrain’s builder.
GBrain’s repository says the system contains 155,795 pages, 24,589 people, and 5,340 companies.
The system runs 66 autonomous cron jobs ingesting meetings, emails, tweets, calls, and ideas.
GBrain supports sourced corrections, withdrawals, shared access, and permission-scoped company use.
Tan says evals are “the real moat for AI startups.”
Why it matters: Tan is both promoting and building the infrastructure he argues AI companies need: systems that retain context, show sources, and can be evaluated continuously.
For YC founders, that points attention away from model access alone and toward proprietary memory, reliable workflows, and controls that make agents usable inside companies.
The immediate question is whether GBrain’s approach gains users or contributors beyond Tan’s own deployment.
Future evidence to watch includes how its permissions perform in real teams, whether provenance reduces incorrect agent outputs, and whether persistent memory produces measurable product advantages.
YC’s September directory update spotlights workplace collaboration and conversational AI startups
Y Combinator’s September 22 directory pages profile 33 collaboration startups and 44 conversational-AI startups, providing a current snapshot of companies it identifies as funded and active.
The entries range from OrgOrg’s internal productivity suite and Mattermost’s secure, self-hosted collaboration platform to Party’s workspace for teams and software agents. The conversational-AI page adds newer companies from the S2026 and F2026 cohorts.
Sona8, a four-person San Francisco startup, uses voice interviews to map work processes and track changes after implementation; Speko routes voice agents across tested speech models; Pluto says 27,000-plus professionals and more than 50 companies use its career-search product.
The listings also include robotics, hotel bookings and speech-data businesses, showing conversational interfaces being applied beyond customer support.
Key facts:
YC’s September 22 pages list 33 collaboration startups and 44 conversational-AI startups.
Party, a San Francisco S2024 company, builds a secure workspace for teams and software agents.
Sona8’s four-person F2026 team uses voice agents to map employee workflows.
Pluto says 27,000-plus professionals and 50-plus companies use its career product.
Speko routes voice agents across speech-to-text, language, and text-to-speech models.
Why it matters: The update suggests YC is backing software that sits inside operational workflows, not just chat interfaces.
Collaboration products target internal coordination, while Sona8 and Speko address process discovery and the infrastructure behind voice agents; Pluto applies conversation to recruiting, and Codyco to hotel bookings.
That breadth gives founders multiple commercial entry points, but it also places very different businesses under a single AI label.
For buyers, the practical question is whether these products produce measurable outcomes: faster coordination, better process visibility, more bookings, or more effective hiring.
The next signals to watch are customer growth, repeat usage, and financing; the directory entries provide selected traction claims but no funding amounts or revenue figures.
EWOR names San Francisco leadership team as U.S. applications surge
On Sept. 22, 2026, EWOR named a senior leadership team in San Francisco after U.S. applications rose 237% in twelve months.
The team brings together Charles Ferguson, an early AI investor and former FrontPage seller, Katie Madding, a former Adjust executive, and three-time founder Gigi Brett.
EWOR says fellows can receive up to $600,000 and weekly one-to-one mentorship from experienced unicorn builders without relocating.
It reports that this year’s joiners raised $10 million on average, while 25% of recent joiners reached nine-figure valuations during the past six months. The expansion embeds a global fellowship in the U.S.
market, and the announcement presents EWOR as harder to enter than Y Combinator. Its model combines capital, mentorship, hiring help and customer introductions, making San Francisco a leadership base rather than a relocation requirement.
Key facts:
EWOR’s U.S. applications rose 237% in twelve months, according to the announcement.
Charles Ferguson, Katie Madding and Gigi Brett will lead EWOR’s U.S. operations.
EWOR offers fellows up to $600,000 and weekly one-to-one mentorship.
Founders joining EWOR this year raised $10 million on average.
Twenty-five percent of recent joiners reached nine-figure valuations within six months.
Why it matters: EWOR is testing whether a high-touch fellowship can attract U.S. founders who already have access to Silicon Valley.
The reported application growth suggests demand for bundled mentorship and networks, while the capital and fundraising figures give founders another route alongside established accelerators such as Y Combinator.
Ferguson, Madding and Brett give EWOR local credibility across investing, product leadership and company building.
The next test is whether the San Francisco team can convert application growth into durable companies and repeatable outcomes; the announcement provides early fundraising figures, not evidence of longer-term performance.
Y Combinator-Backed Spott Raises $21 Million for AI Recruitment Software
Spott, a recruitment software startup, announced a $21 million Series A on September 22, with Base10, Y Combinator and Fortino Ventures participating.
The posts conflict on the lead investor: Phillip Chambers said he was leading the round, while a second post credited Balderton with leading it.
Spott’s founders, Lander Degrève, Manu Vanderveeren and Samuel Smeys, built software that semantically indexes call notes, emails and transcripts for recruitment agencies.
Chambers said revenue has increased more than tenfold since January, while agencies using Spott report up to 30% more placements. The company is hiring, expanding its commercial teams and aims to become the recruitment industry’s global ATS and CRM.
Key facts:
Spott announced a $21 million Series A on September 22, 2026.
Base10, Y Combinator and Fortino Ventures participated in the round.
Phillip Chambers said his firm led; another post credited Balderton as lead.
Spott reports revenue rose more than tenfold since January.
Agencies using Spott report up to 30% more placements.
Why it matters: The financing gives Spott room to expand its commercial teams, continue hiring and pursue its ambition to serve recruitment agencies worldwide.
Its product strategy is to replace disconnected CRM, sourcing and spreadsheet workflows with one system that automatically organizes recruitment communications.
The conflicting descriptions of the lead investor should be resolved in future company or investor announcements.
For now, the confirmed signal is a substantial Series A bringing together Balderton or its associated investment team, Base10, Y Combinator and Fortino Ventures around an AI-focused recruitment platform.