Ninety-Two Companies In, One Shape Emerging
Hey there, investors and innovators!
Fall 2026 just crossed 92 publicly launched companies, up from the 67 we had when we last wrote...this batch is moving fast. With a much fuller picture now available, a few clearer patterns have emerged across the batch as a whole, beyond what any five companies alone could show. This edition keeps our five favorite picks from this week's drops (scroll to see GodHands, Veeza AI, antimattr, The Subvocal Company, and Collar) but widens the trends section to reflect what the broader 92-company set is actually telling us about where this batch is headed.
What the Full Batch Shows
- Infrastructure for AI agents is the single biggest sub-theme in the batch. GodHands is far from alone; companies like Maritime, Nodus, Ascii, Herdr, and Workers IO are all building different layers of the same stack: isolated runtimes, persistent sandboxes, workspace managers, and reliability layers that let AI agents actually run in production rather than just demo well. If S26 was about proving agents could reason, F26 looks like the batch figuring out how to host, monitor, and keep them running.
- Physical and embodied AI now has its own dedicated infrastructure layer, not just applications. Beyond any single robot or drone company, this batch includes tools purpose-built to feed the physical-AI pipeline itself. For example, wearables capturing human motion as training data, labs converting raw instrument signals into usable datasets, and sensors translating equipment behavior into structured intelligence. Physical AI is starting to get the same "picks and shovels" treatment that language models got a few years ago.
- Vertical AI agents are getting narrower, not broader. Rather than general-purpose assistants, this batch is full of agents built for one specific professional workflow end to end: legal intake, hiring, insurance servicing, government contract sourcing, private credit origination, visa paperwork, deal sourcing. The pattern favors founders who've picked one workflow inside one regulated or service-heavy industry and gone deep, rather than building a horizontal tool for many industries at once.
- Simulation and physical-world modeling are compressing timeframes across engineering and product work. Foundation models that replace physics simulation, systems that predict how real users will behave before a product ships, and labs teaching AI to read raw physical instrumentation are all attacking the same bottleneck from different angles: the wait between building something and knowing whether it works.
- The interface itself keeps moving off the screen and into the body. antimattr, The Subvocal Company, Sentient OS, and Hemlock represent a real cluster now, not just an interesting coincidence. Voice, sub-vocal muscle signal, on-device personal context, and physical force-sensing are all being treated as legitimate new input methods for both humans and the AI systems learning from them.
- Batch-wide context: With 92 companies now visible, F26 is tracking a broadly similar AI concentration to prior 2026 batches, with B2B still the largest single category. However, the texture of AI is shifting from "smarter model" plays toward infrastructure, physical deployment, and narrow vertical automation. As more of the batch launches, we'd expect these five patterns to sharpen rather than dilute.
Eight Capital tracks every YC batch, cohort to cohort. Reach out if you'd like the full running list of this batch's launches or a deeper cut on any theme above.
Fund Manager Picks
GodHands
We reconstruct how employees actually work, so nothing breaks
Why it's interesting
GodHands is an enterprise software startup building a knowledge layer for AI-native software engineering teams. The platform automatically captures context created during development and reconstructs any employee’s work from the systems where it was performed, enabling seamless role handovers and preserving critical knowledge. By reducing the time and cost associated with employee transitions, GodHands aims to prevent operational disruptions and ensure business continuity. The company was founded in San Francisco in 2024 by Sai Vamshi Batchu and Aditya Bommakanti and is a member of Y Combinator’s Fall 2026 batch. Its mission is to enable complete role handover in one day by reconstructing employee work and dependencies from their machines and applications, ensuring nothing breaks when employees leave or roles change.
Veeza AI
AI travel agent for people with weak passports, starting with visas.
Why it's interesting
Veeza AI is an AI-powered travel agent focused on simplifying the visa application process for individuals with weak passports, particularly those from the Global South. Founded by Aly Moursy and part of Y Combinator’s F26 batch, the company leverages artificial intelligence to automate and streamline the preparation of tourist visa applications, including document verification, appointment booking, and real-time updates. Veeza AI offers its services both online and through a network of 40+ retail stores, and is trusted by employees at major firms such as Deloitte, PwC, and Amazon. The company’s mission is “to fix the broken visa process plaguing residents of the Global South by leveraging AI to simplify and streamline visa applications.”
antimattr
voice-first hardware interface for human-agent interaction
Why it's interesting
antimattr is developing a voice-first hardware interface for human-agent interaction, comprising a wearable ring and a pair of earphones. The ring is designed to record and summarize conversations, capture thoughts on the go, and control digital agents, while the earphones leverage proprietary voice isolation technology to block out background noise for clear calls and accurate dictation. The company is part of Y Combinator’s Fall 2026 batch, with a registered office in Bangalore and operational presence in the San Francisco Bay Area. antimattr’s first product, Project Mnemosyne, is scheduled to ship in Q4 2026. The founding team includes Sridipto Ghosh (CEO) and Sirsho C (CTO), both repeat founders with prior experience scaling a bootstrapped SaaS business.
The Subvocal Company
Talk to your computer without talking
Why it's interesting
The Subvocal Company is developing a discreet, under-chin wearable device that translates silent, subvocalized speech into text and computer commands in real time. Founded in 2026 and based in San Francisco, the company is a Y Combinator F26 cohort member. The founding team consists of Arnav Kumar (CTO) and Aarush Yadav (CEO), both with computer science backgrounds from Georgia Tech. The company’s mission is “to create a new fundamental way humans interact with computers by enabling silent speech communication that is private, fast, and imperceptible.” The device is designed for seamless integration with productivity tools such as Slack, Gmail, and Calendar, emphasizing privacy, intentional activation, and robust biosignal decoding.
Collar
AI for Investor Relations
Why it's interesting
Collar is an AI-powered SaaS platform designed to automate and streamline investor relations (IR) workflows for public companies, with a particular focus on lean IR teams. Founded by Kazel Koh En Yang (CEO) and Jamie Yau (COO), Collar participated in Y Combinator’s F26 batch. The company’s platform automates earnings preparation, peer monitoring, and investor targeting, enabling IR professionals to focus on relationship-building and strategic communication. Collar claims global coverage, mapping over 12,000 institutions and portfolio managers across the US, Europe, APAC, and the Middle East. The company’s stated mission is to deliver institutional-quality IR insights and targeting at a fraction of the current cost, empowering under-resourced teams to compete more effectively.
