Bret Taylor argues AI agents could turn systems of record like ERP and CRM into databases people no longer log into, putting incumbent software value at risk.
Taylor says incumbents' strengths become a strategy tax during platform shifts, while upstarts with best-of-breed products tend to beat best-of-platform suites.
Taylor is an evangelist for outcomes-based pricing, arguing token-based pricing charges for an input that does not correlate with the output customers care about.
Why AI agents may hollow out enterprise software incumbents
Software stocks are sliding on fear that AI agents will hollow out systems of record. Bret Taylor argues platform shifts reward upstarts, because incumbent strengths turn into a strategy tax. Sierra bets that value lies in agents that do specific jobs, priced on outcomes, while agent building itself becomes a commodity.
Takeaways
Software sell-off reflects anxiety, not company verdicts. Taylor says the market is not indicting individual companies, and not every software firm is equally exposed. The biggest risk sits with systems of record such as ERP and CRM, which could become databases nobody logs into once agents do the work.
Incumbents carry a strategy tax in platform shifts. Past shifts such as the browser and smartphone favored best-of-breed upstarts over best-of-platform suites. Incumbent strengths, like on-premises software or perpetual licences, become anchors. Salesforce beat Siebel this way. Most incumbents lack credible AI today, though Taylor expects them to get it.
Sierra wins in complex, regulated industries. Sierra closed $100m in seven quarters and $150m in eight. Over a quarter of customers top $10bn in revenue, and it serves most of the Fortune 20. Taylor expects a culling of undifferentiated rivals. Cigna went live in two months.
Outcomes-based pricing fits agents, not tokens. Taylor calls outcomes pricing the secular business model for agents. Tokens charge for an input unrelated to what customers value. Applied AI means describing the value without mentioning models, like a phone agent solving 80% of calls at a 4.8 CSAT score.
Building agents will commodify; value is in jobs. Every company needed a website in 1997 and will need an agent in 2027. Agent building is a technology that open source tools like LangChain will commodify. Value will sit in agents doing specific jobs, such as antitrust review, financial audits and vendor onboarding.
Early AI adopters will move fastest. Taylor argues companies that learn to work with AI first will move fastest, and says OpenAI's board expected Codex's surge. Rocket applies AI across the whole customer lifecycle, with over a billion dollars of mortgage folders moving each month.
Taylor's predictions
Voice technology such as voice activity detection will probably become a commodity within one or two years.
Regulators will start asking for agents rather than human controls over regulated processes.
A 10-person billion-dollar company will probably appear, but not as the norm, because competition turns efficiency into investment rather than profit.
Much of the economy is physical and will change more slowly than software.
Tap a name to go there · tap a line to see why they are linked
On this episode 2
Show 1
Tasteful, clearly labelled ads in AI products will work, as they did with Google AdWords.