Is AI Killing Software? — With Bret Taylor

Episode Summary

Executive Summary: Brett Taylor argues AI is changing software more fundamentally than vibe coding suggests: the future is agentic software, not just custom CRUD apps. He sees dashboards, web forms, and even internet discovery being replaced by AI agents, with new business models like outcomes-based pricing. He also discusses Sierra’s rapid enterprise adoption, OpenAI’s monetization, the possibility of a tech bubble, and lessons from leaders like Benioff, Zuckerberg, Altman, and Musk.

Main Topics: AI is reshaping software form factors (Priority: 5/5): Taylor says the real shift is not merely faster app-building, but a transition from browser-based forms and dashboards to autonomous agents that act on databases and systems. He thinks incumbents face disruption because the software people use tomorrow will look different from today’s software. Vibe coding vs. agentic software (Priority: 5/5): He acknowledges the excitement around individuals building custom tools quickly, but argues that the deeper change is that software creation and maintenance will become easier while the interface itself evolves toward agents rather than traditional CRUD apps. Enterprise adoption, trust, and reliability of AI agents (Priority: 5/5): Taylor explains how Sierra gets large companies to trust AI agents with customer actions by using simulations, AI monitors, and human review. He argues AI agents can already be more reliable than human customer-service operations for many tasks. Business model transition in software (Priority: 4/5): He believes AI will change software economics from licenses and subscriptions toward outcomes-based pricing, and that the hardest part for incumbents may be shifting business models rather than technology alone. AI’s effect on the internet and consumer behavior (Priority: 4/5): Taylor predicts personal AI agents will become the front door to the internet, changing search, SEO, discovery, and demand generation/fulfillment. He expects consumer experiences to become more agent-mediated and less centered on websites and apps. OpenAI, scaling costs, and model progress (Priority: 4/5): As OpenAI board chair, Taylor defends monetization through subscriptions, ads, API use, and agents as necessary to fund expensive model training and inference. He says progress is still real, especially for reasoning and long-running tasks, even if some gains come from scaffolding. Tech history, leaders, and the innovation cycle (Priority: 3/5): In a lightning round, Taylor reflects on lessons from Benioff, Zuckerberg, Altman, Marissa Mayer, Sandberg, Larry Page, Sergey Brin, and Elon Musk, emphasizing long-term thinking, community, hiring, candid feedback, and bold vision. He frames the current AI wave as part of a repeated pattern of disruptive cycles and eventual consolidation.

Key Arguments: The main disruption in software is not just that individuals can build faster; it is that the interface and underlying product category are shifting from dashboards and forms to AI agents. Most of the cost of software is maintenance, so businesses will prefer agents and high-leverage platforms that absorb that burden rather than maintaining custom systems themselves. AI-native companies may displace incumbents because each technology wave flips prior advantages into disadvantages if incumbents cannot transform their products and business models fast enough. Outcomes-based pricing is likely to replace many seat-based software models, because agents create measurable results rather than mere access to a tool. The internet will increasingly be mediated by personal AI agents, changing SEO, advertising, and how consumers discover and buy products. AI agents are already useful in mission-critical enterprise workflows such as customer service, benefits support, mortgage workflows, and troubleshooting, because they remove analog bottlenecks like phone queues. Trust in AI can be engineered through simulations, monitoring, and staged deployment, making agents more reliable over time than many human-operated processes. OpenAI needs multiple monetization paths because model training and inference are extremely expensive, and monetization does not negate the public-benefit mission. Model improvements matter most in reasoning-heavy use cases like coding and science, even if casual consumer use does not always expose the gains. The AI boom is likely both transformative and bubble-like: some companies will become generational winners, while others will fail or be consolidated.

Data Points: Sierra annual recurring revenue: $100 million - Taylor cites Sierra’s scale to show enterprise demand for AI agents. Share of customers with over $1B revenue: 50% - Shows Sierra sells to very large enterprises. Share of customers with over $10B revenue: 20% - Indicates penetration among the largest global companies. Sierra founding year: 2023 - Used to emphasize the company’s rapid rise. Customer rollout speed at Cigna: Less than 2 months - Example of fast enterprise deployment of an AI agent. OpenAI pricing/monetization pillars: 3 - Taylor describes ChatGPT subscriptions, API use, and agents as the three main business lines. OpenAI customer-case support example: Per resolved case - Taylor says Sierra charges per resolved case for customer-service agents. AI agent rollout horizon for consumers: 3 to 4 years - Taylor predicts personal consumer agents will arrive within this timeframe. Time to build a custom CRM in anecdote: A night and a morning - Referenced as an example of vibe coding/custom software creation speed. Market move in software bundles: About 10% down this year - Used in discussion of market skepticism toward software incumbents. Historical platform shift example: 1997 bank login forms - Taylor cites early web banking as an example of how hard now-trivial software once was.

Pivotal Quotes: "I think the future of software is agents." — Brett Taylor: Core thesis on how AI changes software interfaces and workflows. "Business model transitions are harder than technology transitions." — Brett Taylor: Explains why incumbents may struggle even if they can adapt technically. "Your personal AI agent will be your front door to the internet." — Brett Taylor: Describes how consumer internet discovery and interaction may evolve.

Implications: Software vendors should prepare for agent-based products, outcomes pricing, and new trust/safety controls. Enterprises can automate more customer and operations workflows now, while consumers should expect AI to mediate search, shopping, and service.