Everyone’s building AI agents. But a clear pattern is emerging in who’s getting results, and it comes down to one thing: depth.

The vertical thesis

Y Combinator’s Lightcone podcast made a bold claim: vertical AI agents could be 10x bigger than SaaS. The logic? SaaS made tools accessible and created over 300 unicorns. Vertical agents go further: they replace both the tool and the operational cost of running it, by making entire workflows intelligent.

That sounds ambitious, but vertical agents are already revolutionising entire industries, one workflow at a time. In sales, YC-backed Origami Agents hit $50K in monthly recurring revenue within eight weeks by doing one specific thing well: researching prospects so human reps can focus on closing. In biopharma, a vertical AI agent helped one enterprise cut marketing agency spend by 20-30% while reducing content localisation from two months to a single day.

The common thread? These agents succeed because they go deep into one domain, not wide across many.

Not all agents are built for this

That distinction matters. Over 40% of agentic AI projects won't reach production by 2027, and the pattern is consistent: horizontal agents that try to cover everything are stalling, while vertical agents built on domain-specific intelligence are delivering.

The reason is simple. Only 1% of enterprise leaders say their organisations have reached true AI maturity, and the gap often comes down to knowledge depth. Shallow knowledge creates generic output. Deep, proprietary knowledge creates better decisions, stronger relevance, and more value over time.

IBM’s Watson is a useful reminder. One of the most publicised AI setbacks in healthcare, it struggled because broad AI could not fit real workflows, local practices, and decision contexts. Vertical agents matter because they are built around exactly those realities.

Playing the long game

If the vertical thesis holds, three things follow for enterprise leaders:

• Your next competitive moat is your knowledge base, not your tool stack. When every team has access to the same AI, the differentiator becomes the proprietary data you feed it.
• Think workflows, not features. The real opportunity is asking which processes could be redesigned around AI entirely, built from the ground up rather than patched on top.
• Compounding beats scaling. Systems that learn from every sales call, campaign, and customer objection create an advantage that widens over time. A competitor can adopt the same tools tomorrow, but they can't replicate the intelligence you've been building.