Bret Taylor, founder of AI Agent startup Sierra, former Co-CEO of Salesforce, and current Chairman of the OpenAI Board of Directors, recently participated in an interview on Lenny's Podcast. Starting from his conclusion that "foundational model startups are a dead end unless you can throw billions of dollars at them like Elon Musk," he discussed how entrepreneurs should precisely position themselves in the current AI wave to find truly unique opportunities.
01 Are Foundation Models a Dead End for Startups, While "Long-Tail Agent Companies" Are the Opportunity?
Bret Taylor's career spans the past two decades of tech waves. In a recent interview, he reflected on his past successes and failures, and based on that, as an entrepreneur, he analyzed the opportunities in the current AI era. He believes the true blue ocean lies in Agents that can deliver business outcomes, with the core business model being "outcome-based pricing," and traditional market rules are being re-examined.
1. Bret Taylor joined Google as a Product Manager in 2003 after graduating from Stanford, where he developed Google Maps.
2. In 2007, he left Google to found the social media company FriendFeed, inventing core mechanisms popular today such as the Newsfeed and the "Like" button. He later sold FriendFeed to Facebook and joined as its CTO.
3. In 2012, he left Facebook to create the document collaboration tool Quip, which he later sold to Salesforce and joined as Co-CEO.
4. In 2023, he left Salesforce to found the Agent startup Sierra, and in the same year, when Sam Altman (after the "boardroom drama") returned as OpenAI CEO, Taylor joined OpenAI as Chairman of the Board.
2. Taylor first deconstructs the AI market into three core tracks: Foundation Models, Toolchain, and Applied AI. He then shared why he leans towards Applied AI, given the current state of each track.
1. For startups, "Foundation Models" represent a "narrow gate" with extremely high capital and technological barriers. This track will ultimately be dominated by a few cloud giants and top-tier laboratories with massive capital. Unless a startup can raise billions of dollars like Elon Musk, there is virtually no room for survival.
2. The "Toolchain" track, while having clear demand, is closely tied to underlying platforms and constantly faces the risk of being integrated into native features by large corporations. Entrepreneurs must continuously answer the sharp question: "Why would customers choose you when a giant releases the same feature?"
3. Taylor believes "Applied AI" is the "wide gate" to a vast market. He is convinced that various Agents will be the ultimate form of AI technology implementation. As the technical barrier to building Agents significantly decreases, any highly repetitive, automatable business process could potentially give rise to a vertical Agent company focused on that specific area.
3. Based on this framework for the AI landscape, Taylor foresees that when orchestrating an Agent becomes as simple as launching a cloud database today, a new ecosystem composed of "long-tail Agent companies" will emerge, potentially even replacing SaaS.
1. Unlike traditional SaaS, which primarily sells "software functions," the core value of an Agent lies in delivering measurable "business outcomes." When evaluating AI products in the future, enterprises will focus directly on: How much cost has this Agent system saved me? How many additional orders has it generated? How much has customer satisfaction improved?
2. This shift towards "Outcome-based Pricing" means that Agent companies naturally have a superior business model to traditional SaaS. They are no longer simple tool providers but partners deeply tied to their clients' business results, thus achieving higher profit margins and customer stickiness.
3. Similar to the attitude towards SaaS products, when enterprises evaluate AI products in the future, users will ultimately care about the value and seamless workflow delivered by the Agent, rather than who provides its underlying technology.
4. After clarifying "what to do?", Taylor pointed out that there are no shortcuts for entrepreneurs when it comes to "how to do it?". There are only three effective models for AI products to enter the market, and entrepreneurs must make clear choices based on product characteristics, rather than blindly following trends.
1. The first category is the "Developer-led" approach. This path is suitable for platform products, infiltrating from the bottom up by winning over engineers' favor, but it is difficult to be effective for business customers who lack dedicated engineering teams.
2. The second category is the "Product-led Growth" (PLG) approach, which requires a high degree of unity between the "user" and the "purchaser," common in small and medium-sized business (SMB) software. Once these two roles separate, this path becomes ineffective.
3. The third category is the "Direct Sales" approach, which targets large enterprise business lines and proceeds through traditional sales processes.
5. Taylor specifically emphasized that in the current AI startup wave, direct sales are making a strong comeback because the end-users (business personnel) and purchasing decision-makers (department heads, IT, or finance departments) for many Agent products are often not the same person.
1. He warned founders, especially those with a technical background, to abandon their prejudice against sales. For most B2B AI companies, building a strong direct sales team is a "decisive move" that must be mastered and is worth excelling at, not an optional extra.
02 Why Is "Outcome-Based Pricing" the True Value Proposition of AI?
1. Bret Taylor believes that to understand the disruptive nature of Agents, one must first return to the essence of business: "Why do companies pay?" His judgment is "outcome > process," and thus, the AI business model will ultimately transition from "per-token billing" to "outcome-based pricing."