Key Moments
AI’s third era: the rise of persistent AI coworkers | Tara Seshan (OpenAI’s product lead)
Lenny's PodcastKey Moments
AI is moving beyond chatbots to persistent co-worker agents, but product development must focus on the next 2-3 months, not the present or distant future. The key to success lies in rapid iteration and ambitious goals, rather than theoretical perfection.
Key Insights
OpenAI operates with a highly transparent, "open" strategy, where ideas quickly become public products or messaging, unlike traditional companies with secret roadmaps.
Product development must target AI model capabilities 2-3 months in the future, as building for current models or a year ahead both lead to failure.
The core of product management is shifting from documentation to rapid hypothesis testing and empirical validation, emphasizing speed and user feedback over theoretical rigor.
The future of work involves 'steering' rather than 'rowing,' with AI agents handling tactical tasks while humans provide direction and strategic vision.
Ambition is the new differentiator; AI tools enable individuals to expand their capabilities, allowing them to realize more complex visions, much like film directors or artists.
The distinction between coding and knowledge work is crucial: coding is output-verified (tests), while knowledge work requires inspecting the process and reasoning behind AI-generated outputs.
The 'open' nature of OpenAI's strategy
Tara Seshan, OpenAI's product lead for ChatGPT Work, reveals a surprising aspect of working at the frontier lab: its profound openness. Unlike many high-growth companies with founder-led, secretive strategies, OpenAI's approach is remarkably transparent. Ideas, product strategies, and operational philosophies quickly become public, either through product releases or public messaging. This "open" culture, while initially surprising to Seshan who came from more traditional, founder-led environments, fosters a sense of collective ownership and rapid iteration, akin to a founder's journey where the distance to the market is minimal.
Navigating the AI development timeline
A critical challenge in AI product development, as highlighted by Seshan, is accurately targeting the pace of model advancement. Building for current AI model capabilities is a recipe for failure, as they are quickly surpassed. Conversely, projecting too far into the future—say, a year ahead—is equally misguided. The optimal strategy involves focusing on the capabilities expected in the next two to three months. This requires a deep, constant connection with research teams to understand their roadmaps and focused efforts on specific model improvements. This tight integration ensures that product development remains aligned with, and ahead of, the evolving AI landscape, rather than being reactive or overly speculative.
The shift from theoretical to empirical product management
The dynamic and emergent nature of the AI market necessitates a significant shift in product management strategy. In more static markets, like payments, rigorous theoretical planning and long-form reasoning documents were key. However, in the AI era, success hinges on being prolific and empirical. The focus moves from writing extensive theoretical documents to rapidly prototyping and testing ideas with users. The most crucial aspect of this empirical approach is defining a sharp, "eigen" hypothesis—the single most important question to test. This allows for faster learning loops, where rapid iteration and data collection inform the next steps, rather than getting bogged down in exhaustive upfront planning. This empirical mindset is not limited to PMs but permeates engineers, data scientists, and designers across the company.
From 'rowing' to 'steering' with AI agents
The future of work is increasingly characterized by 'steering' rather than 'rowing.' AI agents will handle the bulk of the tactical, labor-intensive tasks (the 'rowing'), freeing humans to focus on strategic direction (the 'steering'). This steering is evolving to higher levels of abstraction, moving beyond simple task execution to goal-setting and continuous feedback loops. While data and intuition play roles, the ultimate direction-setting requires human opinion, ambition, and a vision for the future. This collaborative model extends to teams working together with their agents, creating a more integrated and intuitive workflow that resembles a multiplayer game where agents execute tasks and humans guide the overall strategy.
Ambition as the key differentiator
In an era where AI tools democratize the execution of many tasks, individual ambition becomes the primary differentiator. AI empowers individuals to expand their capabilities significantly, moving beyond automating routine tasks to tackling complex, ambitious projects. This is akin to the 'unicorn' individual of the past who combined product sense, engineering, and design skills. Now, with AI, individuals can ideate, design, prototype, and even model financial scenarios with unprecedented ease. This broadens the scope of what's possible, allowing for greater artistic expression and higher fidelity realization of visions. The challenge, therefore, is not in acquiring the capability but in expanding one's thinking and setting appropriately ambitious goals, leveraging the readily available tools to achieve what was once thought impossible.
The evolving role of writing in product development
Seshan distinguishes between 'writing as thinking' and 'writing as reporting.' While writing as reporting—status updates, launch plans—can be automated with AI, writing as thinking, such as developing product strategy or core ideas, remains a critical human activity. The act of outlining, drafting, and iterating on these ideas is essential for clarifying thought. However, the format of 'writing as thinking' is evolving. In contrast to the past, where detailed documents were prized, the current trend favors 'mocks, not docs' or prototypes. Demonstrable results, A/B test data, and interactive prototypes are more effective communication tools than long written documents. While Seshan continues to write extensively for herself, the emphasis for external communication has shifted towards tangible, interactive artifacts, reflecting the need for rapid, empirical validation.
Coding versus knowledge work: distinct AI challenges
A key distinction in AI product development lies between coding and knowledge work. Coding tasks are highly output-oriented, allowing for validation through tests and direct output assessment. AI can generate code that is then rigorously tested for correctness. Knowledge work, however, is different. An AI-generated deck or financial model cannot be validated solely by its final output; the process, inputs, and reasoning are equally critical. This necessitates a product design that emphasizes collaboration, allowing users to see in-progress work, citations, and the AI's reasoning. Features like transparency in citations, chain-of-thought explanations, and collaborative interfaces are crucial for building trust and ensuring the quality and utility of AI-generated knowledge work outputs.
The 'co-worker' model and collaborative AI
The next frontier in AI interaction is the 'persistent co-worker' model, moving beyond one-on-one agent interactions. This involves collaborative workflows where multiple team members and their agents work together. Instead of sharing screenshots of agent interactions, the vision is for a seamless, multiplayer experience where teams collectively steer their agents. This extends to OpenAI's product strategy, aiming to integrate AI capabilities into tools like ChatGPT, making them natural extensions of existing workflows. The ultimate goal is a unified interface where users don't need to choose between different AI modes (chat, coding, work), but rather interact with a versatile AI that understands their needs and facilitates collaboration, both with the AI and with human colleagues.
Mentioned in This Episode
●Software & Apps
●Companies
●Organizations
●Books
●People Referenced
Product Management in the Age of AI
Practical takeaways from this episode
Do This
Avoid This
Common Questions
The third era of AI products, following chat and agent-based systems, is expected to be characterized by persistent AI coworkers that can actively get things done alongside humans.
Topics
Mentioned in this video
A company mentioned in relation to WorkOS's enterprise client base.
A sponsor of the podcast, offering business banking and expense management services, including AI agent card capabilities.
The AI research and deployment company where Tara Sash leads product for Codex and ChatGPT Work. Known for its open approach to strategy and rapid product iteration.
A company mentioned in relation to WorkOS's enterprise client base.
An environmental data company where Tara Sash previously led product.
A company founded by Ari Weinstein that was acquired by OpenAI.
Where Ari Weinstein worked before founding Sky.
A financial services and technology company where Tara Sash previously worked as one of the first product managers and was recognized as a top performer.
A sponsor of the podcast, providing APIs for enterprise features like single sign-on and SCIM for B2B SaaS companies.
A lengthy book recommended for slow reading, with a breakdown available on 99press.
A book recommended by Tara Sash for its message about pursuing passion and striving for excellence, even without achieving perfection.
A book by Hilary Mantel being slow-read on Simon Hazel's Substack.
A classic novel recommended for its layered narrative and how it can be understood differently at various stages of life, reflecting personal growth.
The conversational AI model from OpenAI, which is evolving to incorporate agent capabilities and work modes, aiming to become a persistent coworker.
An AI model developed by OpenAI, integrated into ChatGPT's Work mode, focused on assisting with coding and knowledge work tasks.
A company mentioned in relation to WorkOS's enterprise client base.
An AI model mentioned in comparison to Codex regarding its perceived popularity on Twitter.
A company mentioned in relation to WorkOS's enterprise client base.
A private social network developed by Tara Sash's friend Sebastian, described as a 'private Twitter' for a small group of friends.
Mentioned for his idea that product ideas rarely end up as initially conceived, requiring a process of discovery.
Founder of the Teal Fellowship program.
Author whose four tenets on work serve as Tara Sash's life motto.
A notable and successful alumnus of the Teal Fellowship, founder of a company.
Former Chief Product Officer who stated, 'This is the worst the models will ever be,' emphasizing the rapid improvement of AI.
Author of 'Barbarian Days', a book recommended for its exploration of passion and striving for excellence.
Mentioned for a phrase related to defining the 'igen question' or the most important thing to test.
Director of the film 'Rashomon', pioneering the technique of telling a story through multiple perspectives.
A Teal fellow and founder of Sky (acquired by OpenAI), an expert on Mac usage and creativity.
Runs a Substack that does a slow read of important books, such as 'War and Peace' and 'Wolf Hall'.
A colleague at Stripe who, like Tara Sash, valued writing and sharing briefs.
A former colleague of Tara Sash at OpenAI who had the meme 'Is this maximally accelerated?'
More from Lenny's Podcast
View all 47 summaries
85 min84 minutes of enterprise sales alpha | Jen Abel
91 minThe playbook for building high talent density teams | Adam Ward, Head of Talent at Cursor
85 minThis CPO regrets that product management exists | Tom Verrilli (CPO of Whatnot)
94 minHow Anthropic builds products like Claude Code before the AI models are ready | Dianne Penn
Ask anything from this episode.
Save it, chat with it, and connect it to Claude or ChatGPT. Get cited answers from the actual content — and build your own knowledge base of every podcast and video you care about.
Get Started Free