Voice, Speech, and Conversational AI

Stateful SaaS clones for testing autonomous software.

Agent reliability needs safe mutable test environments; Archal turns SaaS side effects into repeatable CI checks before agents reach production.

Automates job matching, tailoring, applications, and reply tracking.

A YC-backed applicant-side agent can win if ATS coverage, profile memory, and application outcomes make high-volume job search cheaper and less manual.

AI control layer for existing industrial SCADA systems

The bet is that brownfield factories adopt AI fastest through an overlay that speaks SCADA before it tries to replace control systems.

Jo is a private personal AI that knows a user’s Mac context.

Jo’s wedge is a private, Mac-native memory layer with a dedicated cloud machine per user, turning personal context into user lock-in.

Routes stuck AI agents to verified human experts.

If agents keep failing on judgment-heavy tasks, Humwork can own the paid human handoff layer and the corpus of resolved agent failures.

Automates purchase-order entry for manufacturers.

Smartbase is taking a narrow manufacturing wedge where messy POs, ERP friction, and labor cost create a clear automation budget.

Programmable DNA drug platform for genetically defined cancer targets.

A reusable DNA-sensing kill-switch molecule would turn genetic cancer drivers into programmable drug targets if delivery and specificity work.

AI-native agency that sources, screens, and hands off interview-ready candidates.

Prism is testing whether AI can compress agency labor into a lower-fee recruiting service while owning the search and screening workflow.

Atlas turns care-agency back office records into approved operational actions.

Atlas gives a one-person YC company a sharp wedge into regulated care operations where admin labor, billing errors, and audit risk are constant.

AI primary care that reads records and drafts physician-reviewed care.

Clara pairs Circle Medical operating experience with an AI-first care model where the scarce asset is full-chart context plus clinician review efficiency.

Adaptive AI coaching for endurance athletes with messy lives.

Imperfect’s wedge is adaptive coaching for serious amateurs, with founder fit from Garmin data tooling and consumer fintech product building.

AI gets hardware compliance done in weeks, not months.

Fuchsia turns hardware certification into a software-like workflow at the moment lab rules and supply chains are getting harder to manage.

Enterprise AI voice and chat agents for B2C support and sales.

Cignara turns enterprise contact centers into revenue surfaces by pairing governed voice agents with buyer-specific knowledge graphs and founder-led enterprise AI experience.

AI point-defense turret for close-range drone threats.

Passive acoustic cueing plus visual tracking gives 9 Mothers a narrow, high-urgency wedge in close-range counter-drone defense.

Agent-native observability that turns logs into incident answers.

The bet is that coding-agent adoption makes dashboard-first observability too slow, giving Sazabi a timing edge around logs-first, agent-native incident response.

Turn messy operational data into AI-built workflows, apps, and ML pipelines.

Netter compresses data engineering and applied ML implementation into a chat-led operations platform for companies that lack large internal data teams.

Orchestrates parallel coding agents across isolated workspaces.

Superset turns coding-agent adoption into an orchestration layer, with timing, open-source distribution, and review workflow as the wedge.

Developer-native consent infrastructure for web teams.

Inth turns consent from a legal dashboard into developer infrastructure, using open source adoption as distribution into a slow CMP category.

Finds and tracks obscure regulatory changes across jurisdictions.

Regbase's edge is source discovery: it targets obscure public-sector records that generic legal AI and search miss.

Brand OS for agentic marketing creative.

If agents become marketing operators, Bloom can own the persistent brand context they call before creating customer-facing work.

Builds return satellites for in-space manufacturing payloads.

The bet is that downmass, not launch, becomes the scarce utility for in-space manufacturing, and Dispatch can own that return layer early.

Automates wholesale specialty insurance placement for retail brokers

Hedge turns the manual E&S wholesale layer into software-mediated placement, with founders who have built AI brokerage and AI ops systems.

Fuses counter-UAS sensors and effectors into an open defense OS.

ParallaxOS is a YC-backed bet that counter-drone defense needs neutral software, not another closed hardware stack.

Turns therapy sessions into notes, plans, claims, and client context.

Klarify has a narrow wedge into therapist workflow, then expands into the admin and clinical context layer around each client.

AI agents that execute manufacturing back-office workflows.

ERP replacement is a hard sell, so Korso’s wedge is an agent layer that makes legacy manufacturing systems act together.

Programmatic email infrastructure for autonomous agents.

If agents need durable inboxes, Primitive can own the email control plane before incumbents repackage transactional email APIs for agents.

AI researches and executes hedge fund trades.

If autonomous financial LLMs produce durable out-of-sample alpha, Standard Signal turns model performance into fund economics rather than seat-based software revenue.

Turns pull requests into AI-structured review chapters.

Stage has a timely wedge: AI-generated code is increasing review load, and the product makes human review faster without replacing reviewer judgment.

Builds verifier-grounded training data for frontier AI labs

Verifier-grounded data is a timely wedge into frontier lab budgets as model gains shift from web scale to correctness.

Builds fund-specific AI command centers for asset managers.

WithAI is a bet that hedge funds will buy AI as private workflow infrastructure, with founder-market fit and early client pull supporting a high-ACV wedge.

Runs coding agents in real dev environments for engineering teams.

Replicas turns coding agents from chat tools into production workflow infrastructure by giving them real environments, CI loops, and team entry points.

Parlay RFQ and settlement layer for prediction markets.

The bet is that parlays become a native trading primitive for prediction markets, with Totalis owning RFQ liquidity and Solana settlement between venues.

AI takeoffs and bid outputs for concrete subcontractors.

Rudus is a focused bet that trade-specific takeoff depth can beat horizontal AI estimating tools in high-value subcontractor workflows.

Simulates how AI agents find and use software products.

The pitch is early agent-distribution infrastructure: teams will need to test how agents choose, use, and fail on their products.

Lets teams turn product requests into agent-built pull requests.

Lightsprint is betting the next coding-agent wedge is shared product planning and review, not another single-player IDE assistant.

AI scientists for semiconductor materials discovery.

The bet is that a closed loop between agentic design and lab validation can compress semiconductor materials discovery into a venture-scale platform.

Scores engineers by observing real AI-assisted coding work.

Saffron is a bet that engineering hiring will move from coding puzzles to audited, AI-assisted work trials inside real repos.

Send and automate trusted iMessage conversations at scale.

Chert turns iMessage into a programmable GTM and support channel at a moment when email, LinkedIn, and A2P SMS are crowded or constrained.

Networked RF test equipment that works without driver installs.

The pitch is modern test equipment for RF teams tired of legacy driver stacks and Windows-bound lab workflows.

AI-native broker for industrial machinery and supplier sourcing.

Andustry is turning industrial sourcing into a success-fee transaction layer, with AI reducing broker labor before the purchase closes.

Robots ask for help, then learn from each intervention.

A credible wedge if human-in-the-loop robotics turns early reliability gaps into training data and better RaaS margins.

AI scheduling agent for waste-fleet dispatch and route planning.

Dayjob has a narrow wedge in waste logistics, where small route-efficiency gains create clear ROI and ERP workflows are old enough to leave room for an AI planner.

One agentic operating layer for company work.

YC timing plus a founder with retrieval and LLM eval depth gives this a credible shot at the company operating-agent layer.

AI agents that run short-term rental operations.

Trellis has a strong vertical wedge: STR operators already run fragmented stacks, and Trellis is building the agent layer that can act across them.

Protects model weights for secure AI deployment.

The bet is that model distribution needs runtime-level IP protection before proprietary weights can safely leave a vendor perimeter.

Builds low-cost strike drones for mass production.

Tenet is a volume-first defense bet: if drone warfare rewards cheap, repeatable production, its factory-led design can matter more than feature depth.

API for rights-cleared real-world AI training data.

Hub.xyz is a bet that model quality will depend on scarce real-world human data, with supply, provenance, and QA becoming the scarce layer.

Manages trade-contractor material workflows from request to invoice.

A trade-contractor system of record with invoice and supplier-document data that can compound across the long tail of construction purchasing.

Installs observability and opens PRs for production bugs.

Self-installing OpenTelemetry gives Superlog the entry point to own incident context before coding agents attempt fixes.

Agents operate owned e-commerce brands.

The wager is that owned brands turn agent execution into a reusable operating record, letting Light Anchor test commerce workflows faster than SaaS sellers.

AI agents retrieve, redact, and draft FOIA responses for agencies.

GovGuard is betting that AI-generated FOIA volume turns public-records work from ticket tracking into archive automation.

AI sales agents that quote, qualify, and hand off insurance buyers.

Kinro turns regulated insurance distribution into an AI agent workflow, backed by founders with DeepMind, Zoox, and insurtech operating experience.

AI health companion for chronic illness patients.

Juno has early pull in a neglected chronic illness market where daily use can create personal health memory and paid consumer retention.

Automates servicing workflows for commercial lenders.

Zolvo attacks the servicing headcount tax in factoring and ABL by sitting on top of legacy loan systems instead of replacing them.

AI reservation ops for restaurants.

The wedge is missed-call capture, but the asset is restaurant-owned booking and guest data that can feed retention, table yield, and direct demand.

Deterministic developer AI for structured backend tasks.

Interfaze packages specialized small models behind an OpenAI-compatible API for tasks where variance, latency, and cost matter more than chat fluency.

Turns enterprise documents into agent-run operational workflows.

Qomplement is a YC-stage bet on document-heavy back offices replacing manual copy-paste work with agentic workflow execution.

Global business finance for non-U.S. companies.

HEVN is attacking the gap left by U.S.-centric Brex, Ramp, and Mercury with a crypto-native team and a wedge into global company banking.

Turns event engagement into CRM pipeline.

A sharp wedge is event signal capture, where offline interactions become structured CRM context and pipeline attribution for marketing and sales teams.

Maximizes data-center compute with chillerless dew-point cooling.

Madrone attacks the power and water penalty of AI data centers at the facility layer, replacing chillers with dew-point cooling that could raise compute per grid permit.

Agent-led marketplace for startup hiring introductions.

Standout can price like a recruiter if its agents turn private talent intent into high-quality hires with software-like labor ratios.

Predicts B2B SaaS churn risk from account usage signals.

A YC pivot from Wudpecker gives Userlens founder-market fit and a wedge into CS teams that need account-level churn signals without rebuilding product analytics.

Turns website traffic into customers with AI-run CRO experiments.

Sherpa compresses CRO from agency-and-tool workflow into a closed-loop agent, with early Lindy proof and founder ties to growth engineering.

Salesgraph turns revenue-system context into deal-specific sales guidance.

Salesgraph turns scattered deal traces into a Slack-native operating layer for enterprise reps, with founder-market fit from building the first version inside Mintlify.

Agent-native marketplace for hiring people to do physical-world tasks.

RentAHuman is a timely bet that agent software will need a trusted human execution layer for work that still requires presence, proof, and judgment.

Turn one marketing brief into hundreds of persona-specific social campaigns.

The bet is that performance marketing spend moves toward managed AI creative systems that can test far more personas than in-house teams can produce manually.

Automates PI case workups from intake to demand.

Andco targets the upstream PI bottleneck, using agents to gather the records and claim context that downstream legal AI needs.

Agent validation infrastructure with production-like service twins.

Arga turns agent reliability from prompt QA into infrastructure, with service twins and CI workflows that can become a required layer for AI-native engineering teams.

Local-first customer memory for enterprise deal execution

Drip could become the account memory layer for founder-led B2B teams before the CRM becomes the system of truth.

Voice agents that plug into legacy enterprise telephony.

The bet is telephony-native voice AI for enterprises that cannot replace CUCM, PBX, and SIP infrastructure.

Turns limited robot videos into synthetic training data for physical AI.

Synphony’s wedge is scarce real-world manipulation data: start in strawberries, then sell synthetic datasets and evals to robotics labs.

AI-native managed billing for independent medical practices.

Taiga packages AI coding and denial work as a managed service, giving small practices a revenue-cycle team without hiring billers or operating software.

Voice-first maintenance assistant for manufacturing technicians.

Pairio can wedge into industrial maintenance by turning technician voice, photos, and repair history into a plant-specific knowledge base before incumbents modernize UX.

Semantic browser runtime for AI agents.

The bet is that semantic site graphs become the control plane for agents that need authenticated web apps to work in production.

Portfolio-aware AI research desk for buy-side PMs.

A team that ran its own quant fund is productizing the research desk it wished existed for PMs who cannot build an internal AI and data stack.

Generates whiteboard explainer videos for AI-native EdTech.

The pitch is fast, deterministic visual explanation infrastructure for EdTech AI, where clarity and generation latency matter more than cinematic video quality.

AI FDEs automate enterprise software implementation.

If implementation becomes a product surface, Lab0 can turn services-heavy SaaS deployment into repeatable agent workflows.

Finds high-intent LinkedIn prospects and automates warm outbound.

The wedge is signal-timed LinkedIn outbound for small teams, with reported YC traction and a founder group that has built and sold SMB SaaS before.

AI food journaling built around reflection, photos, voice, and social retention.

Napkin Math is a bet that AI logging plus a social journal can make food tracking durable for users who reject calorie-counting apps.

AI agents interview employees and map how work gets done.

Ontora compresses consulting-style operational discovery into AI-led interviews and turns the output into reusable company context for automation.

Normalizes prediction-market data for institutional traders.

Prediction markets are becoming tradable macro signals, and Oddpool is building the data plumbing institutions need before they can trade them at scale.

Natural-language image analysis for life-sciences labs.

Alchemy can own the usability layer between wet labs and bioimage analysis if verified workflows become reusable across assays.

Turn agents into lifelike real-time video calls.

YC-backed video-agent infrastructure with a rare founder fit and a visible cost claim: lifelike calls priced as low as $0.06 per minute.

Compact mobile MRI for cancer screening without fixed scanner infrastructure.

Adialante turns MRI access into a per-scan service, with compact RF-encoded scanners that could shift cancer screening into specialty clinics.

Real-time 3D ultrasound for vascular access and procedure guidance.

If DAT makes true 3D ultrasound cheap and portable, Lumius can move point-of-care imaging from expert-only interpretation toward guided procedures.

CAD-to-part automation for mechanical parts.

The bet is that CAD-native automation can compress mechanical iteration cycles from weeks to days in markets where supply chains still run on manual review.

SMS-native AI assistant answers frontline worker questions from company documents.

Sidekick turns SMS into the access layer for frontline knowledge, making AI usable for workers who will not adopt another app.

Mount secures, risk-scores, and insures deployed AI agents.

Mount turns agent security testing into underwriting data, giving insurers and enterprise buyers a way to price autonomous software risk.

Clip-on sensors find commercial building energy waste in real time.

AICE turns energy monitoring from a BMS project into a fast sensor deployment, with savings tied to operational fixes rather than reporting.

Automates physical product compliance for cross-border retailers.

Complir turns product compliance into SKU-level infrastructure for retailers facing faster rule changes, more markets, and larger private-label catalogs.

AI-native customs brokerage for faster importer compliance.

Alchemize turns customs brokerage into a software-speed service, with founder strength in infrastructure and document automation.

Turns trial protocols and source documents into AI-built EDC workflows.

Harbor’s bet is that protocol parsing plus source-document capture can collapse EDC setup and site data-entry labor in regulated trials.

Packages software as fast, portable virtual machines.

A container-like VM runtime with fast boot and portable artifacts could become the default sandbox layer for agentic code execution.

Turns plain-English reward rules into automated payouts.

If incentives keep moving from spreadsheets into programmable workflows, Archer can own the rule layer that decides when money moves.

Self-healing desktop automation for legacy systems.

If AI software needs reliable writeback into legacy systems, Minicor can become the execution layer between new agents and old desktops.

Slack-native AI analyst that answers business questions from connected systems.

Datost’s wedge is accuracy, using business-context retrieval and verification to make Slack a credible analytics surface instead of a chat wrapper.

Voice-first Mac agent that operates the computer by speech.

If Ara makes computer use reliable by voice, it can own a daily Mac workflow before agent platforms become default OS behavior.

Ray-grade cluster execution for quant backtesting at scale.

The bet is that Ray and LinkedIn systems veterans can turn quant backtesting scale into a specialized compute platform before firms rebuild it in-house.

Turns architectural plans into construction cost intelligence.

PLAN0 is pitching the estimate as the entry point to a regional construction-cost data platform for underwriting, procurement timing, and design tradeoffs.

Tests agent workflows across MCP and CLI surfaces.

The pitch is agent experience testing: Armature turns model and harness drift into a monitored reliability layer for agent-facing software.