Bret Taylor co-founded Sierra, which builds customer-facing AI agents, and chairs the board of OpenAI. Earlier in his career he was chief technology officer of Facebook and co-chief executive of Salesforce. At Sequoia Capital’s AI Ascent event in 2025 he sat down on stage with Sequoia’s Ravi Gupta, and Sequoia published the conversation on 8 May 2025 as part of its Training Data series. They covered how founders grow, Sierra’s thesis, how the AI market is dividing, outcome-based pricing, why business models trip up incumbents, and how to sell to large companies.
Growing out of your strengths
Taylor opened with a lesson from his time at Facebook. Founders, he said, tend to be single-issue voters who fix every problem with their own speciality, when growth needs attention to sales, product and competition in turn. Sheryl Sandberg once told him to stop doing the work himself and instead hold his team to his standard. He had been shaping the job around what he was good at. Once he reframed it around what mattered most each day, he found more satisfaction in the impact.
The agent as the front door
Sierra’s hypothesis, he explained, is that within about five years most digital interactions with a company will run through an AI agent, following the path of websites and then mobile apps, and that each company will need one main branded agent. He described these agents as an expression of the brand, much as a website is today. He named ADT and SiriusXM as customers, with examples such as a homeowner whose alarm has stopped working and a subscriber whose promotional price has expired.
He divided the AI market into three layers. Foundation models are capital-hungry and likely to consolidate into a few players with thin margins and great scale, rather like cloud infrastructure. Tool makers, the pickaxe sellers, face pressure as model companies move into their territory. Applied AI is the layer he is most positive about. He sees agents as the new form of software-as-a-service: a customer buys an agent that does a job. Harvey in legal work and Sierra in customer experience were his examples.
Many enterprises that licensed models after ChatGPT have seen little value, he said, and now feel buyer’s remorse. Vertical agents also open larger markets, because they can take on work that currently relies on expensive expert labour. He noted that today’s largest enterprise software firms sit at roughly $200 billion to $300 billion in market value, and predicted that the first trillion-dollar applied enterprise software company would be built on selling outcomes instead of productivity.
Paying only for results
Sierra uses outcome-based pricing. Its median customer pays a pre-agreed rate each time the agent resolves an issue on its own, and nothing when the case is handed to a person. Taylor placed this as the next step after perpetual licences and SaaS subscriptions: paying for a finished job, much as a business pays commission on a sale.
The model also changes how the vendor behaves, he argued. Enterprise software firms often sell a product and step back while integrators spend six to twelve months deploying it. Charging for outcomes keeps the vendor accountable for whether the product actually works.
Why incumbents struggle
For established firms, he said, the business model is the harder hurdle. He pointed to Microsoft’s move to Azure and Adobe’s switch to recurring revenue as difficult transitions, and to chief executives who lost their jobs during shifts of that kind. Start-ups are not tied to an old model.
“Closing a technology gap in your product is hard, but not impossible.”
Changing the business model, by his account, is much harder. Incumbents can still use AI to their advantage, for example by changing the cost structure of contact centres that cost hundreds of millions a year and putting the savings into lower prices or growth. He compared the moment to the web era, when Walmart adapted and Blockbuster did not.
He also reflected on how much easier it has become to start a company. At FriendFeed his team ran its own servers, and a single power cable fault took the site down. Cloud infrastructure, open source and modern finance tools now remove most of that overhead, so founders should spend their time on what sets them apart.
Selling to large companies
Executives hear from many AI vendors every day, and he said most of the pitches sound alike. His advice was to ask questions, listen closely and then link the product’s value to the problems the customer has just described. Before meetings, vendors should research prospects in depth, using deep research tools.
Answering questions from the audience, he backed vertical agents built for specific workflows, citing telecom customer service, the different needs of commercial and residential banking, insurance claims and queries about health benefit statements. He was sceptical of horizontal platforms, though he allowed that LangChain might succeed as open source. On pricing, he said top-line impact matters most, with cost savings still counting, and expected savings to shrink over time as agents are compared with other agents.
What it means for leaders
- Customer experience leaders weighing agent vendors can ask for outcome-based terms, since paying per resolved case moves delivery risk back to the supplier.
- If an agent becomes the main way customers reach a company, its tone and accuracy turn into brand questions, and marketing should help design it.
- Taylor puts the real barrier for incumbents in the business model. Our reading is that pricing, revenue recognition and sales commission plans need review before the technology roadmap does.
- Agents tied to one workflow look like the safer first purchase, while broad platforms leave more of the work to the buyer.
