AI summary by TwinMind
Vertical AI Agents Could Be 10X Bigger Than SaaS
Y Combinator
42m 13s
Introduction & AI Progress Overview
- Every 3 months AI capabilities have progressively improved
- Now discussing full vertical AI agents replacing entire teams/functions in enterprises
- Foundation model competition increasing - no longer just OpenAI
- Competition creates fertile marketplace ecosystem for founders and consumers
Historical Context: SaaS Boom Parallels
- Last 20 years: Silicon Valley primarily produced SaaS companies
- Over 40% of venture capital went to SaaS companies in that period
- Produced over 300 SaaS unicorns - more than any other category
- XML HTTP request (Ajax) in 2004 was catalyst for SaaS boom
- Enabled rich internet applications in browsers
- Software shifted from CD-ROM desktop installs to web/mobile access
- Paul Graham invented first SaaS with Viaweb in 1995
- Early SaaS sucked due to page reloads before XML HTTP request
Three Categories of Cloud/Mobile Era Companies
Category 1: Obviously Good Consumer Ideas
- Docs, photos, email, calendar, chat - obvious desktop-to-browser migrations
- Zero startups won these categories
- 100% value flowed to incumbents (Google, Facebook, Amazon)
- Example: Google Docs beat 30+ companies trying to bring Office online
Category 2: Non-Obvious Consumer Ideas
- Uber, Instacart, DoorDash, Coinbase, Airbnb
- Connection between XML HTTP request and these companies not obvious
- Incumbents didn't compete until too late
- Regulatory risks (Travis Kalanick feared prison for Uber)
Category 3: B2B SaaS Companies
- 300+ billion-dollar companies created
- Most successful category by number of logos
- No "Microsoft of SaaS" - structural reasons prevent single dominant player
- Each requires deep domain expertise in specific verticals
- Salesforce was first true SaaS company
- Early skepticism about cloud-based enterprise applications
Why Incumbents Didn't Dominate B2B SaaS
- Too hard for one company to do everything well
- Each B2B SaaS requires deep domain expertise
- Example: Google couldn't build Gusto competitor - no payroll expertise
- Unbundling vs bundling software argument
- Old enterprise software (Oracle, SAP, NetSuite) had poor UX trying to be "jack of all trades"
- Vertical solutions could deliver 10x better experience
- Three software price points: $5/seat (consumer), $500/seat (SMB), $55,000/seat (enterprise)
Current AI/LLM Revolution Parallels
- Similar computing paradigm shift to cloud/mobile
- LLMs enable fundamentally different capabilities
- Same three-category pattern emerging:
- Obviously good consumer AI (general voice assistants) - incumbents likely to win
- Non-obvious consumer AI opportunities
- Vertical AI agents for B2B - biggest opportunity
The $300 Billion Vertical AI Agent Opportunity
- Prediction: $300+ billion companies will emerge in vertical AI agents
- Every SaaS unicorn could have vertical AI agent equivalent
- Key difference: AI agents replace both software AND people
- SaaS still required operations teams to use software
- Vertical AI agents eliminate need for human operators
- Companies spend more on payroll than software - massive TAM expansion
- Potential for 10x larger companies than SaaS predecessors
Current State & Examples
Hiring Challenges Evolution
- Traditional hiring: need buy-in from both engineering and recruiting teams
- Recruiting teams opposed software that threatened their jobs
- AI enables full-stack recruiting replacement
- Example companies: Nico, A Priora - doing full technical and recruiter screens
Customer Support AI Agents
- Appears crowded (100+ companies) but most doing simple zero-shot prompting
- Only 3-4 companies building sophisticated systems to replace full teams
- Less than 1% market penetration cumulatively
- Power Help example: complex workflows beyond simple demos
- Hyper-specialization emerging (Gig ML for Zepto handling 30,000 daily tickets)
Voice AI Applications
- Salient: AI voice calling for auto loan debt collection
- Replaces high-churn call center jobs
- Going live with major banks
- Voice infrastructure companies (Vapi) enabling rapid deployment
- 6 months ago voice wasn't realistic enough - now replacing human callers
Other Vertical Examples
- Outset: LLM-powered surveys competing with Qualtrix
- Much: AI QA testing replacing entire QA teams
- Cap.AI: Developer support chatbots reducing DevRel team sizes
- Sweet Spot: AI bidding on government contracts
- Medical billing AI for dental clinics
Strategic Considerations
Selling Vertical AI Agents
- Must sell high enough in organization to avoid job threat perception
- Teams being replaced will sabotage if they control purchasing
- Often requires CEO sign-off
- Can target companies without existing teams to avoid replacement friction
Technology Infrastructure
- Voice infrastructure rapidly improving
- Question: can platforms maintain differentiation vs direct API usage?
- Need to keep raising ceiling to retain customers
Enterprise Adoption Patterns
- Enterprises already trained on value of point solutions vs broad platforms
- Faster enterprise traction than historical SaaS adoption
- Don't need to reset to general-purpose platforms like early enterprise software
Management & Scale Implications
- AI tools extending manager capabilities beyond traditional limits
- Dunbar number (150 meaningful relationships) potentially expandable
- Example: CEO voice chat calling all 1,500 employees for pulse checks
- LLMs can read, summarize, and potentially think - extending executive reach
- Rippling model: 100+ founders running SaaS verticals within one company
- Counter-argument to pure verticalization - some consolidation opportunities
Advice for Founders
- Look for boring, repetitive administrative work
- Find verticals through personal experience/relationships
- Examples: friend's government contract bidding job, founder's mother's dental practice
- "Dirty, dangerous, and dull" jobs principle from robotics applies to AI agents
- Focus on "boring butter passing jobs"
Current Batch Evolution
- Winter 2023: First LLM apps were basic text generation (blog content, email editing)
- Many early "ChatGPT wrappers" got disrupted by OpenAI releases
- 2024: Full vertical AI agents replacing entire teams
- Foundation model competition increasing (Claude vs OpenAI)
- Boiling frog effect - steady progress every 3 months over 2 years
