Y CombinatorVertical AI Agents Could Be 10X Bigger Than SaaS
CHAPTERS
- 0:00 – 1:07
Vertical AI agents are arriving faster than anyone expects
The hosts set the stage: LLM capabilities are improving every few months, pushing the conversation from simple chatbots to “vertical AI agents” that can replace whole teams. They also note how model competition is increasing, which should accelerate innovation and startup opportunity.
- •Rapid, compounding improvement in LLM capabilities
- •Shift from simple outputs to agents that execute workflows
- •Foundation models becoming more competitive beyond OpenAI
- •Thesis: this unlock will reshape company functions and staffing
- 1:07 – 2:30
The big bet: vertical AI agents could create $300B+ companies
Jared argues founders are underestimating how large vertical AI agent companies can become. He frames the opportunity as potentially producing multiple $300B+ outcomes within this single category.
- •Vertical AI agents are underestimated by founders
- •YC has already funded many early examples
- •Claim: category can produce multiple $300B+ companies
- •Setup for an analogy to the SaaS boom
- 2:30 – 4:22
Why SaaS exploded: the AJAX/XMLHttpRequest unlock and cloud delivery
Jared traces SaaS’s inflection point to AJAX (XMLHttpRequest), which enabled rich web apps and made browser-based software viable. The discussion highlights how technical UX improvements can catalyze entire business categories.
- •AJAX made web apps feel like desktop apps
- •Catalyzed products like Gmail and Google Maps
- •Shift from CD-ROM installs to web/mobile delivery
- •Viaweb as an early proto-SaaS that was ahead of its time
- 4:22 – 9:06
Three post-platform value paths: incumbents win ‘obvious,’ startups win ‘surprising,’ SaaS wins by volume
They categorize big outcomes from the SaaS era into: obvious consumer migrations (won by incumbents), unexpected consumer breakthroughs (won by startups), and B2B SaaS (hundreds of winners). The key observation is that B2B SaaS produced the most unicorns by count.
- •Obvious consumer categories tended to accrue to incumbents (Google/Facebook/Amazon)
- •Startups won in unpredictable categories (Uber, Airbnb, DoorDash, Coinbase)
- •B2B SaaS produced ~300 unicorns, dominating by number of outcomes
- •Structural reason: no single ‘Microsoft of SaaS’ for every vertical
- 9:06 – 11:36
Why big companies avoided B2B SaaS: domain depth, regulation, and bad ‘kitchen sink’ UX
The group unpacks why incumbents didn’t build countless vertical SaaS products: it requires deep domain obsession, navigating messy edge cases (like payroll), and building great UX for specific users. Broad suites often become jack-of-all-trades with poor experiences.
- •Vertical SaaS needs deep domain expertise (e.g., payroll nuance)
- •Too many niche problems for one big company to prioritize
- •Legacy suites (Oracle/SAP/NetSuite) trade breadth for weak UX
- •Vertical products can be 10x better for specific end users
- 11:36 – 16:25
How LLMs may change company building: fewer hires, more leverage, ‘10-person unicorns’
They explore how AI could break the traditional scaling relationship between revenue and headcount. The conversation shifts to how founders might prioritize hiring LLM-savvy engineers to automate bottlenecks instead of building large operational teams.
- •Traditional unicorns often require 500–2,000+ employees by $100–$200M ARR
- •Hiring strategy may shift toward engineers who automate growth bottlenecks
- •Concept of extremely small teams running huge businesses (e.g., 10 employees)
- •Harj’s Triplebyte story: smart engineers can create outsized leverage (now amplified by LLMs)
- 16:25 – 20:28
Core thesis: every SaaS unicorn maps to a vertical AI agent that bundles software + labor
Jared states the central analogy: vertical AI agents are to SaaS what SaaS was to box software. The agent version doesn’t just provide tools—it performs the work, collapsing “software + team” into a single product.
- •For each SaaS category, an agent can replace both tool and operator
- •AI agents can disrupt SaaS the way SaaS disrupted installed software
- •Enterprises already value point solutions, enabling faster adoption
- •Debate: early general platforms vs day-one vertical agents
- 20:28 – 21:41
Why vertical AI could be 10x bigger than SaaS: it captures payroll, not just software spend
Diana argues vertical agents expand the market because SaaS mainly sells software while humans still execute workflows. Agents can replace substantial labor costs, making outcomes potentially much larger than the SaaS predecessors.
- •SaaS still requires ops teams to run workflows and input data
- •Agents can absorb much of the labor component (payroll dwarfs software spend)
- •Result: smaller, more efficient companies with fewer humans
- •Possibility that some point solutions remain standalone without broad suites
- 21:41 – 26:32
Real-world vertical agent examples and the ‘don’t sell to the team you’re replacing’ lesson
They share examples across surveys, QA, recruiting, DevRel, and customer support—highlighting that adoption often requires selling top-down. A recurring go-to-market insight: teams threatened by automation may resist or sabotage purchases.
- •Outset: LLM-native surveys/Qualtrics replacement (language-heavy workflow)
- •Momentic: AI QA that can replace QA teams rather than merely augment them
- •Apriora: automating recruiter + technical screens end-to-end
- •Capilot.ai: reduces DevRel/support burden by ingesting docs, videos, chat history
- •PowerHelp: real support automation is harder than ‘demo’ wrappers; market still wide open
- 26:32 – 29:04
Why there will be many winners: hyper-vertical customer support and tailored evals
The hosts argue the market will fragment into specialized agents because real deployments require tailored workflows, test cases, and evaluation sets. This mirrors the SaaS era where customization needs prevented one universal provider from winning everything.
- •Most ‘AI support agents’ are shallow prompting; true replacement needs deep systems
- •Examples like GigaML: specialized to a marketplace’s real ticket load and workflows
- •Hyper-verticalization early; general-purpose comes much later
- •Tailored eval suites and domain constraints create defensibility
- 29:04 – 35:22
Horizontal suites vs vertical agents: Coase’s theory, Rippling’s strategy, and AI-boosted management scale
They debate firm size limits and whether AI increases how much an organization can be effectively managed. Rippling is discussed as a counterexample pushing toward a horizontal platform that can ‘eat’ multiple SaaS verticals by bundling and leveraging a shared go-to-market engine.
- •Coase’s theory: firms grow until coordination costs outweigh benefits
- •AI may extend managerial leverage (‘rocks can read’) and increase firm scale
- •Dunbar number constraints might be stretched by AI summarization and analysis
- •Rippling’s approach: platform + many founder-led verticals; bundling to raise LTV while holding CAC
- 35:22 – 39:49
AI voice calling takes off: debt collection, voice infrastructure, and defensibility questions
They spotlight voice agents as a fast-moving subcategory, including debt collection automation in auto lending. The conversation also addresses platform risk: voice infra makes it easy to start, but companies must raise the ceiling to remain defensible as underlying APIs improve.
- •Salient: AI voice agents automating auto-lending debt collection
- •Call center work: high churn, low-wage ‘butter-passing’ tasks well-suited for automation
- •Voice infra (e.g., Vapi) enables rapid prototyping and adoption
- •Open question: how voice startups defend against commodity underlying APIs (e.g., OpenAI voice)
- 39:49 – 42:12
Founder playbook: pick the right vertical by finding repetitive ‘boring admin’ work you understand
They close with guidance for choosing a vertical: hunt for repetitive, painful administrative tasks and use firsthand exposure to discover workflows. Examples include government contract bidding and dental/medical billing, often found through personal relationships or embedded observation.
- •Look for boring, repetitive admin work with clear ROI for automation
- •Founder-domain proximity matters: experience or close relationships unlock insight
- •Sweet Spot: AI agent for government contract bidding discovered via observing a friend’s job
- •Dental billing agent idea found by spending a day in a founder’s mother’s clinic
