Analysis
Three companies sold the same pitch this week, in three unrelated categories: Numeral automates sales-tax filing, Ema automates HR/IT/finance workflows, and Bird.com is wiring AI agents directly into messaging and calls. None of them are pitching a smarter chatbot. Every one of them is pitching an agent that does a specific job a human currently does, for less.
That's a different story than the one the AI-agent narrative usually tells, which is about horizontal platforms -- one agent framework flexible enough to handle any enterprise workflow. What's actually shipping and getting funded right now is narrower and, I think, more durable: vertical agents that own one function end to end, priced against the headcount or the incumbent software they replace, not against a subscription seat. Ema's raise, reported by TechCrunch, cited 50-fold revenue growth in two years and roughly 180% net dollar retention -- numbers that look like customers are expanding usage once the agent proves it can do the job, not experimenting and churning.
“The pattern that should worry incumbent SaaS vendors is that this isn't a feature war.”
The pattern that should worry incumbent SaaS vendors is that this isn't a feature war. Salesforce can bolt Agentforce onto its existing platform, and it has, but a startup selling "this agent replaces three FTEs in your AP department" is competing on a completely different axis than "this agent makes your existing software smarter." Headcount replacement is a budget line CFOs already understand and approve against; a smarter-software subscription is a nice-to-have that gets cut first in a downturn.
Room for disagreement: the bear case here is real. Every one of these companies is still small relative to the incumbents they're displacing, none has proven the wedge holds once an Avalara, a Workday, or a Twilio matches the price and ships a comparable agent of their own, and "replaces headcount" is a pitch that works until a customer discovers the agent needs more human oversight than advertised -- Numeral itself keeps a human in the loop for edge cases, which is the correct design choice but also a tacit admission full automation isn't there yet.
What I'd actually diligence on any vertical-agent pitch chasing this thesis: ask for the net-revenue-retention number, not the logo count. Bird.com's own case study -- cutting headcount from 1,000-plus to roughly 120 while growing EBITDA -- is the sharpest version of this story, and it's a company automating itself with the same category of tooling it now sells. If a vertical-agent startup can't point to a comparable before-and-after in its own operations, be skeptical of how confidently it's selling that outcome to customers.