Browse · People
Person · Latent Space wiki

Bret Taylor

Sierra co-founder and chair of OpenAI’s Foundation and Group PBC boards, with earlier work spanning Google Maps, FriendFeed, Facebook, Quip and Salesforce.

Bret Taylor co-founded Sierra with Clay Bavor to build AI agents for businesses. As of September 9, 2026, he chairs both the OpenAI Foundation and OpenAI Group PBC boards. His Sierra work centers on agents that complete customer tasks through business systems; his OpenAI responsibilities include board oversight and governance. He previously co-created Google Maps, co-founded FriendFeed and Quip, and served as Facebook’s CTO and Salesforce’s co-CEO.[2][4][8][13]

Sierra: choosing the problem and the model strategy

Taylor and Bavor selected digital customer experiences after conversations with CIOs, CTOs and CEOs. In their February 2024 Semafor interview, Taylor described customer service as an immediate demand within a larger ambition: a company’s agent could become as important to its digital identity as its website or app. Bavor characterized their approach as customer-led.[8]

Taylor explained the founders’ decision to use external models rather than pretrain their own. Retrieval could supply changing business information, while fine-tuning and other post-training could adapt behavior. His example was a retailer extending its holiday return period: changing that policy should not require rebuilding a foundation model. He argued that Sierra’s investment should concentrate on differentiated agent systems while benefiting from other companies’ investment in models.[8]

Taylor described a development sequence of proving concepts with powerful models and specializing afterward. His commercial position likewise emphasized the completed task. In February 2024, he said Sierra customers paid when an issue was fully resolved, with no charge when the customer had to move to a human. He later argued that outcome pricing better aligned an agent maker with its customer than charging for seats or tokens.[8][1]

Business systems, supervision and human control

The surrounding software is central to that strategy. Bavor explained that a model cannot know an individual order’s status or return eligibility without access to company records. He described supervisors that review proposed responses and actions, alongside deterministic protections for access to customer and company data.[8]

Sierra agent engineer Natalie Meurer’s July 2024 technical explanation describes an Agent SDK that composes reusable skills, controls the order of API interactions and connects agents to systems of record. Built-in and custom supervisors check behavior. The SDK can also selectively bypass an LLM when a response requires precise regulatory language, combining flexible conversation with deterministic behavior.[9]

Taylor also emphasized the people who shape an agent after deployment. He described an unnamed design partner whose customer operations staff began calling themselves AI architects while testing the agent, reviewing live conversations and changing its behavior. He argued for tools accessible to people without technical backgrounds, drawing inspiration from the Salesforce administrator role. His optimism about new work came with a qualification: automation could reduce traffic to human service workers, and the transition required sensitivity and accountability.[8]

Taylor and Bavor’s July 16, 2026 Horizon announcement extended their focus to goals spanning multiple interactions over days, weeks or months, supported by a context engine and long-horizon planning. They described an unnamed US healthcare provider coordinating specialist referrals among patients, specialists and referring physicians—a task that can require many texts and calls before an appointment is scheduled. The founders retained outcome pricing: “you don’t pay for tokens, you pay for business outcomes delivered.”[10]

OpenAI: crisis mediation and governance

Taylor entered OpenAI’s governance during its November 2023 leadership crisis. In his February 2025 Latent Space interview, he recalled talking with Adam D’Angelo and Sam Altman and acting informally as a mediator. He described seeking a path that would reinstate Altman while commissioning a review of the board’s concerns. His public explanation for becoming involved was that OpenAI’s work had helped inspire Sierra, and he feared the institution might dissolve.[1][8]

As chair of the new initial board, Taylor set priorities of stabilizing OpenAI, broadening board perspectives, commissioning an independent review and strengthening governance. A Taylor–Larry Summers committee selected WilmerHale to conduct the review, as announced on December 8, 2023. In March 2024, Taylor stated the board’s unanimous confidence in Altman and Greg Brockman. OpenAI also announced governance guidelines, a stronger conflict-of-interest policy, an anonymous whistleblower hotline and additional committees.[13][14]

OpenAI’s published review summary attributed the crisis to a breakdown in trust. It said the prior board had acted within its broad discretion, while concluding that Altman’s conduct did not mandate removal. The subsequent endorsement of leadership therefore did not amount to a finding that the prior board had no concerns.[14]

The September 9, 2026 announcement identifies Taylor as chair of both OpenAI boards. The Foundation is the nonprofit that controls Group PBC, the for-profit public benefit corporation. The Foundation holds the rights to appoint and replace Group directors; those are institutional powers. In February 2025, Taylor described safety and access as central board concerns under OpenAI’s mission to ensure that AGI benefits humanity.[4][3][1]

Public views on building AI software

Taylor argues for close integration of product, design and engineering when technological limits are still changing. In February 2025, he explained that requirements for AI products cannot be settled independently of what models can actually do. He favored small teams that understand both customer needs and technical trade-offs, while acknowledging that established business applications can have good reasons to separate the disciplines. He said Sierra built much of its tooling internally and described agents as being in the “jQuery era of agents,” before mature abstractions had emerged.[1]

At that time, Taylor regarded coding assistants as valuable copilots that still required interactive inspection. His longer-term aspiration was an AI-native programming system combining languages, formal verification, testing and automated review so that trust would not depend on a person reading every generated line. He used Rust’s static memory-safety checks to illustrate properties that a compiler can verify. He presented this as a hope for future software development, rather than a firm prediction, and argued that mission-critical systems still need precise specifications.[1]

Taylor’s earlier work provides concrete context for these views: at Google Maps he combined product responsibility with JavaScript development, and at Quip he and Kevin Gibbs reconsidered document collaboration for phones and tablets. Those projects preceded his work on AI agents.[6][1][7]

Earlier career: Google Maps, FriendFeed and Facebook

Taylor co-created Google Maps and was the group product manager who launched it. He credits Where 2 Technologies founders Lars and Jens Rasmussen with setting a native-application quality bar for the browser product. In his later recollection, making a draggable map required progressively loading tiles and working around browser constraints, including limits on simultaneous image downloads from one domain. He also recalled consulting browser engineers about Firefox crashes and implementing JavaScript workarounds for identified bugs.[2][6][1]

Taylor recalls that satellite imagery and Safari support made the XML/XSLT-heavy Maps code slow and fragile. He rewrote the JavaScript front end over a weekend, drawing on his accumulated knowledge as one of its primary authors. He described replacing XML data handling with JavaScript sent from the server and evaluated on the client. His account explicitly corrects the often-repeated description of this as a backend rewrite.[1]

Taylor founded FriendFeed with Paul Buchheit, Jim Norris and Sanjeev Singh in October 2007. On August 10, 2009, Facebook announced an agreement to acquire the sharing service and bring its employees into the company, with all four founders taking senior engineering and product roles.[6]

At Facebook, Taylor contributed to social plugins, the graph API and Open Graph. His June 2010 CTO appointment formalized an unusual remit: advising across technical and product issues while reporting to Mark Zuckerberg, without managing the engineering organization. Zuckerberg’s employee memorandum, supplied to TechCrunch, explained that Taylor would remain focused on Platform while helping shape other areas. In June 2012, Taylor announced plans to start a company with Kevin Gibbs.[15][16]

Quip and the transition through Salesforce

Taylor and Gibbs introduced Quip on July 31, 2013, arguing that the shift to phones and tablets warranted rethinking productivity software. Quip combined documents, discussion and edit history in one thread. Visible diffs showed collaborators what had changed without requiring them to reread the document; screen-adaptive layouts and offline synchronization supported mobile use. Their stated ambition was “the productivity suite for the mobile era.”[7]

Salesforce closed its acquisition of Quip on August 26, 2016. Taylor subsequently served as president and chief product officer, then president and COO from 2019. He became vice chair and co-CEO alongside Marc Benioff on November 30, 2021. A year later, he announced his departure, effective January 31, 2023, saying he had decided to “return to my entrepreneurial roots.” Sierra became his next venture with Bavor, whom he had first met at Google.[11][12][5][2]

Timeline

  1. founding

    Founds FriendFeed with Paul Buchheit, Jim Norris and Sanjeev Singh
    FriendFeed[6]

  2. release

    Introduces Quip with Kevin Gibbs to rethink mobile document collaboration
    Quip[7]

From the Latent Space archive

Original posts and separately reviewed explanations.

Sources

  1. 1
  2. 2
    About SierraAccessed 2026-09-17
  3. 3
    Our StructureAccessed 2026-09-17
  4. 4
  5. 5
  6. 6
  7. 7
    Introducing QuipAccessed 2026-09-17
  8. 8
  9. 9
    Meet the AI agent engineerAccessed 2026-09-17
  10. 10
    The next Horizon in agentsAccessed 2026-09-17
  11. 11
    Quip + Salesforce = Big NewsAccessed 2026-09-17
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16

Model source and teaching reviews passed. Human editorial review pending. This profile describes the sources as of 2026-09-17.

Latent Space

Join Latent Space.

Sign in with Google, or get the newsletter by email.

Loading Google sign-in…

or subscribe by email

Free. Unsubscribe any time.