ACCELERATED · WINTER 202484% confidenceObserved October 2, 2026Source: YC tagRadar flagged this 18 months early
The count
4.10×
5/1334 → 15/976; +1.16 percentage points of formation share.
W15W21W23W25S26F26
Open questions naming this formation — 4
OPEN
Human-in-the-loop escalation and handoff infrastructure
Berry, Yuma AI, and Toma all automate customer-facing conversations that will inevitably fail edge cases, implying demand for seamless escalation-to-human tooling as a distinct infrastructure layer rather than each startup building it themselves.
NEXT INVESTIGATION
Check back around August 2027 for new YC batches, to see whether independent evidence for "Human-in-the-loop escalation and handoff infrastructure" has appeared.
OPEN
Agent tool-orchestration and permissioning layer
OneGrep's DevOps-focused conversational AI and the general rise of agents interacting with internal systems suggests a need for tooling that manages what agents are allowed to call, similar to how Vocode is open-source infrastructure for building agents rather than an end-user product.
NEXT INVESTIGATION
Check back around June 2027 for new YC batches, to see whether independent evidence for "Agent tool-orchestration and permissioning layer" has appeared.
OPEN
Vertical-specific conversational agent wrappers
Toma (automotive) shows conversational AI specializing by industry rather than staying horizontal like kapa.ai or Yuma AI; this narrow vertical focus pattern is likely to replicate into other high-call-volume verticals such as healthcare or real estate.
NEXT INVESTIGATION
Check back around May 2027 for new YC batches, to see whether independent evidence for "Vertical-specific conversational agent wrappers" has appeared.
OPEN
Voice agent QA and evaluation infrastructure
Companies like Vocode, Marr Labs, and Toma are building voice-based conversational agents for real business calls (sales, support, automotive), and Hyperbound already exists to simulate sales calls for training reps — suggesting a parallel need to test and score AI-driven calls before deployment rather than just human reps.
NEXT INVESTIGATION
Check back around May 2027 for new YC batches, to see whether independent evidence for "Voice agent QA and evaluation infrastructure" has appeared.
we are making AI feel more human starting with voice. ai applied research lab training a novel end-to-end model & productizing what makes AI more personal, persistent, and human. frontier AI, like Claude and ChatGPT, are weak on memory, personality, relationship, and human-like interaction...that's our domain.
OneGrep is a DevOps Agent that helps engineers automate manual devops tasks. It works across integrations that teams have to search, correlate, and operate automated tasks like runbook automation during incidents.
OneGrep specializes across devops tools to help teams release code faster and more reliably so that engineers can spend more time building rather than being trained in bespoke tools like Observability, Ci/Cd, or Infra.
Achintya and Brian are technical co-founders who have deep domain expertise in building and operating systems at scale at Coinbase, Disney, and Amazon. Achintya led the Search Team at Disney enabling DisneyPlus to scale to 70M users in two weeks. Brian was a founding engineer on the team that built the Incident Management Platform at Amazon that helps 30k DAUs in Amazon resolve critical incidents every day.
https://www.onegrep.dev
Open is an AI-powered, all-in-one customer support and communication platform — used in production by:
Print.com, Mollie.com, Viva.com, More.com, Orderchamp.com, OneCarNow.com, and other major enterprise companies.
We specialize in solving and automating complex support cases across email, web, phone, SMS, and WhatsApp.
We’re a research-based company currently working on releasing Support-Bench, the world’s most comprehensive AI customer support benchmark.
Our team consists of +3 programming language builders, a nationwide cybersecurity competition winner at age 16, and a dentist.
At Toma, we are building the AI software suite for underserved industries, starting with Automotive.
We started as a voice AI infrastructure company and still are our own biggest customer to this day. Our product empowers hundreds of businesses with a world-class AI receptionist and customer support platform. Toma’s AI agents engage in thousands of customer interactions every day and generate millions of dollars worth of revenue.
Why Automotive first? It is a sleeping giant contributing trillions of dollars to the GDP. The industry and its processes are ready for disruption. Toma is bringing AI to all operations in the industry while working hand-in-hand with some of the biggest names, such as the Car Dealership Guy, and public companies like Lithia Motors and AutoNatio. Toma is backed by the best investors in Silicon Valley, including Andreessen Horowitz and Y Combinator.
Zep assembles the right context from chat history, business data, and user behavior so agents are personalized, accurate, and fast. Our open source project Graphiti hit 20k GitHub stars in under 12 months. Sub-200ms retrieval, SOC 2 Type 2/HIPAA certified, used by teams from startups to Fortune 500s.
Terrakotta is a power-dialer that lets sales reps leave AI-generated voicemails. Our first customers are Commercial Real Estate brokerages that use Terrakotta to clone their brokers voices, dial prospects and when no one answers, leave an A/B tested, personalized voicemail that gets a callback.
Marr Labs is making AI-voice agents that are indistinguishable from humans. Han and Dave previously created speech tech company Vlingo that powered the first Siri app and was acquired for $225M.
Now they are taking voice to the next level by making AI-voice agents that can handle virtually any phone-based job, from answering calls to schedule appointments to call center sales and support jobs—all at a fraction of the cost of a human agent.
In the US alone, businesses handle 17 billion calls a year at an average cost of $5 per call–an $85 billion addressable market.
kapa.ai makes it easy for technical companies to build AI support and onboarding bots for their users. Teams at +150 leading startups and enterprises incl. OpenAI, Mixpanel, Mapbox, Docker, Next.js and Prisma use kapa to level up their developer experience and reduce support.
We enable companies to use their existing technical knowledge sources incl. docs, tutorials, chat logs, and GitHub issues to generate AI bots that answers developer questions automatically. More than 750k developers have access to kapa.ai via website widgets, Slack/Discord bots, API integrations, or via Zendesk.
Hyperbound is the leading enterprise-first Revenue Activation Platform, pioneering a new category focused on turning insights into revenue-driving actions. Hyperbound AI analyzes thousands of sales calls to identify the behaviors that lead to closed-won deals, then transforms those insights into realistic, highly customizable human-like roleplays with AI buyers that drive behavior change at scale.
Trusted by enterprises like IBM, LinkedIn, Monday.com, Bloomberg, Vanta, Autodesk, and G2, Hyperbound helps revenue teams ramp 50% faster, define what good looks like, and scale top performance across thousands of reps.
Interested in trying it out today? Start a free demo simulation at https://app.hyperbound.ai/buyers and book time with our team to learn more.
Surface Labs helps B2B companies capture, qualify, and convert inbound leads through intelligent contact forms. We replace tools like HubSpot Forms, Zapier, and Chili Piper with a single platform that embeds on any site with one line of code.
Our AI agents handle lead capture, enrichment, routing, follow-up, and CRO testing — automating what used to require full marketing ops teams. Customers see a 30% lift in conversions in under 2 weeks, and our partial response capture feature collects 50% more emails from form drop-offs.
Inside Conversational AI
AI Assistant
A primary internal company segment within Conversational AI, derived deterministically from the shared yc tag “AI Assistant”. This is a classification of current window members, not a causal layer or a separate time trend.
Other / unclassified
Measured-window members of Conversational AI that could not be assigned to a narrower shared YC tag or industry without forcing a classification.
Sales
A primary internal company segment within Conversational AI, derived deterministically from the shared yc tag “Sales”. This is a classification of current window members, not a causal layer or a separate time trend.
Operations
A primary internal company segment within Conversational AI, derived deterministically from the shared yc industry “Operations”. This is a classification of current window members, not a causal layer or a separate time trend.
Outside corroboration — does anyone else see this?