Vertical AI agents could be 10x bigger than SaaS

Executive overview

SaaS produced 300+ unicorns over 20 years by replacing boxed software with cloud delivery. Vertical AI agents are poised to do the same to SaaS — but the market is larger, because agents replace not just the software but the people who operate it.

The key structural reason SaaS fragmented into hundreds of vertical companies applies equally to AI agents: no single player can be deep enough in every domain. Enterprises have already been trained to trust vertical point solutions, shortening the adoption curve.

The companies that win will combine the software and the workforce into a single product — eating both the SaaS line and the payroll line.

Why SaaS fragmented — and why AI agents will too

• Ajax (XML HTTP request, 2004) was the technical unlock for SaaS, just as LLMs are for AI agents.
• Three buckets emerged from the SaaS era: obvious consumer ideas (incumbents won), surprise consumer ideas (startups won), and B2B verticals (300+ unicorns, startups won).
• No "Microsoft of SaaS" exists because each vertical requires deep domain expertise — Google never built a Gusto competitor.
• The same logic applies to AI agents: no single platform can serve every vertical at the depth required.
• Enterprises have already built the muscle to trust vertical startups over broad legacy platforms.

Why vertical AI agents can be 10x bigger than SaaS

• SaaS replaced software costs; AI agents replace software costs and headcount — the much larger line item.
• A QA agent (e.g. Momentic) doesn't make QA teams faster — it eliminates the QA team entirely, removing the internal political friction that hobbled prior SaaS tools.
• The sell moves up the org chart: go to engineering or the CEO, bypass the team being replaced.
• Companies that never built a legacy team can scale on agents from day one.
• Coase's theory of the firm still applies — firms will specialise, but the outer limit of what one person can manage expands dramatically with AI tooling.

Real examples from the current batch

• Outset — LLM-powered surveys replacing Qualtrics; language-native insight extraction.
• Momentic — AI QA agent; pitch is "no QA team needed", not "faster QA team".
• A Pryora — full technical and recruiter screen automation; removes need to build a recruiting function.
• Kappa.AI — developer support chatbot trained on docs, videos, and chat history; DevRel teams shrink.
• PowerHelp — AI customer support; most competitors use zero-shot prompting and can't replace real teams; true replacement requires complex workflow software.
• Salient — AI voice agent for auto-lending debt collection; sold top-down to banks; replaces high-churn call-centre roles.
• SweetSpot — AI agent to bid on government contracts; discovered by watching a friend manually refresh procurement pages.
• Medical billing for dental clinics — founder spent a day at his mother's practice; identified claims processing as the target.

How to find the right vertical

• Look for boring, repetitive admin work — that is the common thread across every successful vertical AI agent.
• Direct domain experience or proximity to the workflow dramatically shortens discovery time.
• Target roles where the buyer is not the person being replaced — sell top-down to avoid internal sabotage.
• Early traction is faster than ever: enterprises are already conditioned to adopt vertical solutions without waiting for a general-purpose platform first.

Voice agents as a fast-moving subcategory

• Latency and realism crossed the threshold for realistic voice AI within roughly the past six months.
• Infrastructure platforms like VAPI enable deployment within hours.
• Key open question: can voice-infra companies raise the ceiling fast enough to retain customers as OpenAI releases competing voice APIs?
• The pattern mirrors early SaaS: early winners built on raw LLM APIs; the moat comes from depth of integration and domain-specific evals.

The horizontal vs. vertical tension

• Rippling is the counterexample: Parker Conrad recruits founders to build vertical products inside a horizontal HR/IT platform, treating shared infrastructure as the moat.
• 100+ founders now run individual SaaS verticals within Rippling; each product hits multi-million ARR at launch.
• LLMs extend the "Dunbar limit" for managers — AI that reads, summarises, and communicates at scale lets a small leadership team operate a much larger organisation.
• The bull case: AI tools expand the effective span of control, enabling firms to grow larger before hitting efficiency limits — not just replace workers.