Interview, Fireside Chat
Vertical AI Agents Could Be 10X Bigger Than SaaS
Y CombinatorGary, Jared Harge, Diana, Mark Mandelbaum, Aaron Cannon, Mike, Brett Taylor, Parker Conrad, Matt McGinnis, Salient, VAPI, Rippling, Speedy Brand, OpenAI, Claude, Melanie Warrick, Nico, A Priori, Capital.ai, PowerHelp, Giga ML, Zepto, Mementic, Rainforest QA, Triplebyte, Outset, Vector Shift, Mark Benioff, Paul Graham, Travis, Jake Heller, Flo Cravello, Tia, Hashi Roginio, Mark Mirchandani, Francesc Campoy Flores
- The pace of AI advancement has accelerated such that "every three months" sees significant improvements, progressing from basic LLM wrappers to "full-on vertical AI agents" capable of replacing entire enterprise teams and functions.
- Jared Harge projects that $300 billion+ companies will emerge solely within the vertical AI agent category, drawing a parallel to the SaaS boom of the last two decades.
- Historically, over 40% of VC dollars in the last 20 years flowed to SaaS companies, resulting in over 300 SaaS unicorns, a number surpassing any other category.
- The SaaS boom was technically catalyzed in 2004 by the introduction of "XML HTTP requests" (Ajax), which enabled rich internet applications and shifted software from desktop CD-ROMs to browser-based models.
- Silicon Valley's 20-year history produced three distinct types of winners:
- Obvious consumer products (e.g., email, chat): 100% of value accrued to incumbents (Google, Facebook, Amazon); zero startups won.
- Non-obvious consumer products (e.g., Uber, Airbnb): Incumbents failed to compete due to the "innovator's dilemma" and regulatory risks, allowing startups to win.
- B2B SaaS: Approximately 300 unicorns emerged because no single "Microsoft of SaaS" could effectively cover every vertical and domain nuance.
- Jared Harge argues that vertical AI agents are a continuation of the SaaS trend where incumbents (Google, Apple) will likely win obvious general-purpose AI tools, but startups will dominate specialized verticals due to the difficulty of mastering complex, obscure domain knowledge.
- Enterprises have historically paid more on payroll than software; vertical AI agents can disrupt this by replacing both the software and the human labor required to operate it, potentially creating companies 10x larger than the SaaS predecessors they disrupt.
- A key strategic shift for new AI startups is avoiding the "consumer replacement" narrative (selling to the team being replaced) and instead targeting top-down decisions (CEOs or Engineering VPs) to bypass employee friction.
- Specific examples of vertical AI agents replacing entire teams include:
- Mementic: Replaces entire QA teams rather than just making them more efficient.
- Nico (A Priori): Automates the full stack of recruiting, including technical screening and initial outreach.
- Capital.ai: Reduces DevRel and support teams by ingesting documentation and chat history to provide technical support.
- Salient: Automates debt collection in auto lending, replacing low-wage call center workers with high-accuracy AI voice agents.
- GigaML: Handles 30,000 daily customer support tickets for specific marketplace clients like Zepto, replacing teams of 1,000 people.
- Founders are advised to identify "boring, butter-passing jobs" (repetitive administrative tasks) by observing real-world workflows, such as a founder's mother processing dental billing or a friend refreshing government contract websites.
- The market is moving from general-purpose "wrapper" apps (2023) to specialized, multi-step agents with complex evals and specific domain training sets.
- Jared Harge anticipates a more competitive foundation model landscape, moving away from OpenAI's monopoly to include contenders like Claude, creating a "fertile marketplace ecosystem" for consumers and founders.
- Diana Hargreaves notes that while general-purpose AI voice assistants face stiff competition from incumbents, specialized AI voice agents for specific verticals (e.g., debt collection, recruiting) are seeing faster enterprise traction.
- Mark Mandelbaum points to Rippling as a counter-example to the vertical model, where Parker Conrad is building a horizontal platform by recruiting founders to build specific verticals (e.g., HR, IT) on a shared infrastructure, aiming to become a $100B+ company.
- AI agents may extend the "span of control" for leaders beyond the Dunbar limit (150 people) by autonomously summarizing interactions with thousands of employees, effectively allowing one manager to oversee a much larger organization.
- Companies are currently navigating the transition of AI voice capabilities; while latency and realism have improved drastically in the last six months, platforms must build "moats" beyond basic APIs to retain customers against direct competition from model providers like OpenAI.
- Startups are encouraged to leverage their own domain experience or relationships (e.g., a founder working with a dentist parent) to identify high-value vertical opportunities rather than attempting to build generic tools.
- The "box software" to "SaaS" to "AI Agent" evolution suggests a pattern where early adoption focuses on general utility, followed by a surge of highly specialized vertical solutions that offer superior user experiences and efficiency compared to broad legacy platforms.