Bret Taylor on AI strategy, OpenAI governance, and building enduring companies

Executive overview

Waiting until AI is proven before deploying it means learning only after your competitors have already won. Bret Taylor — Chair of OpenAI's board and founder of Sierra — argues that businesses must accumulate hands-on experience with AI now, treating imperfect experiments as the price of future competitiveness.

The core strategic insight is simple: get at-bats. Deploy AI broadly across your organisation today, accept that not every experiment will work, and build institutional knowledge before the technology matures.

The companies that wait for AI to be "ready" will discover readiness only when a competitor proves it — too late.

Why businesses can't afford to wait

• Don't wait for AI to be perfect; by the time it's proven, a competitor will have proven it first.
• Give employees access to AI tools now — for drafting, analysis, research, and synthesis.
• Back-office and call-centre operations are the lowest-friction starting point: summarising transcripts, synthesising data.
• Every business should have a branded AI agent facing customers; delay means lost learning, not saved risk.
• Demand experimentation from management teams, even knowing many experiments will fail.

How AI is changing software and work

• Software has historically enhanced productivity; AI will instead return the results of doing work.
• Large language models are a new category of software — not a smarter search engine, not a rules engine.
• Hallucination and creativity come from the same underlying property; the two cannot be separated.
• AI increases the performance gap: the best people get an "Iron Man suit" of capabilities; others fall behind.
• The role of the software engineer is shifting from writing code to operating code-generating machines — a more strategic role.
• Professions that resist tool change will be left behind; those with a learning mindset will compound their advantage.

Adapting careers and education for an AI-driven economy

• Education should teach people how to think, not just which tools to use — tools will change, thinking compounds.
• What it means to be a marketer today is almost unrecognisable from 20 years ago; every profession faces this.
• New jobs will emerge that don't yet fit our mental model of "a job" — humans' drive for status and differentiation ensures demand for work.
• Employers must embrace reskilling as a core responsibility, not an optional programme.

OpenAI governance and the board's role

• Taylor joined as board chair during the 2023 governance crisis; his motivation was ensuring OpenAI survived a period when collapse seemed possible.
• OpenAI's mission — AGI that benefits all of humanity — functions as an "inkblot test": optimists focus on access and inclusion, pessimists focus on Terminator-style risk.
• Being a fiduciary to a non-profit with a societal mission means balancing competing interpretations of "benefit to humanity."
• The safety and security committee provides board-level oversight of model safety decisions; it is expected to evolve as models become more capable.
• Responsible iterative deployment — shipping incrementally while monitoring risks — is OpenAI's core methodology.

Lessons from Google, Facebook, and Salesforce

• Google: first-principles infrastructure thinking and vertical integration — essential for margin in AI businesses.
• Facebook: permission to iterate fast, experiment freely, and move without fear of imperfection.
• Salesforce: world-class go-to-market execution and the ability to generate excitement around enterprise technology.
• Sierra is an attempt to composite the best of all three — infrastructure discipline, product velocity, and sales rigour.

What makes Silicon Valley's startup ecosystem work

• Capital, talent, and culture are the three ingredients — all three must be present simultaneously.
• Natural talent flow between big tech and startups creates a self-reinforcing cycle of knowledge and risk tolerance.
• Silicon Valley is uniquely tolerant of ambitious failure — investors treat a failed founder as a better founder, not a damaged one.
• The best constraint on big tech is a healthy startup ecosystem, not regulatory limits on incumbents.

Building societal trust in AI

• Safety: AI companies must recognise that their individual actions affect the entire industry's credibility.
• Job displacement: treating reskilling as outside a tech company's responsibility is both insensitive and irresponsible.
• Sierra has allocated equity specifically toward job displacement and reskilling initiatives.
• Disinformation may actually improve as AI trains people to verify rather than blindly trust digital content.