OpenAI's product lead explains the shift from chat to agents to persistent AI coworkers, and what it means for the future of work and human creativity.
AI's Third Era: How Persistent AI Coworkers Will Transform Work
Key Insights
- Three eras of AI products: Chat (first era), agents (second era), and persistent AI coworkers (third era emerging soon)
- Speed over perfection: Building two to three months ahead of current model capabilities—not for where models are now or where they'll be in a year
- Steering vs. rowing: Humans will increasingly focus on strategic direction while AI agents handle execution and tactical work
- Ambition as competitive advantage: When everyone has access to the same AI tools, human vision and opinionated choices become the differentiator
- The future is collaborative: AI agents working alongside human teams in shared workflows, not isolated one-on-one interactions
Understanding AI's Evolution: From Chat to Coworkers
The landscape of AI products is shifting fundamentally. The first era focused on chat interfaces—conversational AI that answers questions. The second era introduced agents, AI systems that can take autonomous action and get things done. The emerging third era centers on persistent AI coworkers: systems that work alongside you continuously, capable of collaborative problem-solving over time.
This progression reflects a deeper truth: as AI capabilities expand, the relationship between humans and AI must evolve from tool to teammate. A persistent coworker isn't something you interact with once; it's something you work with across days, weeks, and projects—syncing up at different cadences, reviewing in-progress work, and iterating together.
Building at the Right Pace: Two to Three Months Ahead
One of the most counterintuitive insights for product teams: you fail if you build for where models are now, and you fail equally if you build for where you think they'll be in a year. Both approaches are wrong.
The right strategy is building two to three months ahead. This requires constant communication with research teams about roadmaps and focused improvement efforts—knowing, for example, that coding capabilities or writing skills are getting better in specific ways. Close coordination between product and research ensures alignment, so product constructs don't limit the model's future potential.
This philosophy reflects the reality of exponential progress: the market moves too fast for long-term prediction, but completely reactive building leaves capabilities unused.
The Shift from Documents to Prototypes
Product management is undergoing a radical transformation. Traditional PM work—writing detailed reasoning documents, creating presentations, reasoning from first principles—worked in slower, more predictable markets. Today's hyperfast AI landscape demands a different approach.
Writing as thinking remains essential: briefs, strategies, and problem definitions still require human effort and iteration. But writing as reporting—status updates, summaries, structured communications—should be automated. The real shift is from "mocks not docs" and "prototypes not docs" to showing results from actual experiments.
Rather than a polished document proposing a direction, teams now lead with mocks people can interact with, or even better, A/B test results showing why one path works better than another. This isn't laziness; it's recognizing that in a world where AI can generate long documents easily, the document itself is no longer proof of thinking. Only concrete evidence—a prototype, an experiment, a result—signals genuine work.
Steering vs. Rowing: The Human Role in an Agentic Future
As AI agents take on more execution work, the human role transforms from "rowing" (doing tactical work) to "steering" (setting direction). But steering isn't passive. It requires:
- Intuition and opinionated choices: Deciding which direction the product should go, not just what the data says
- Vision and artistry: Software, like film, benefits from authorship and intentional vision
- Accountability: Ultimately, a person owns whether the outcome is what was needed
The most effective teams won't be those with the smartest agents—they'll be teams where humans provide clear direction, agents execute intelligently, and feedback loops tighten continuously. The multiplayer aspect matters too: as more work gets done with agents, the question becomes how teams collaborate through their agents, not just individuals working one-on-one with their own.
Ambition as the New Differentiator
When AI tools become commodities—available to everyone—the playing field flattens. The hard stuff becomes easy. The easy stuff becomes trivial. What separates people and companies now is ambition.
Previously, a "unicorn hire" was someone equally skilled at product thinking, engineering, and design—someone who could compress multiple functions. AI has democratized those skills. Now, the rare advantage is the human ability to want something ambitious and articulate it clearly.
With AI tools, you can now:
- Spin up designs, prototypes, and financial models yourself
- Learn to build something almost instantly
- Execute ideas that were previously beyond your reach
The limiting factor is no longer capability; it's imagination. Teams and individuals who expand their thinking about what's possible—who ask "What if we tried 10x bigger?" or "What if we did this three months faster?"—will pull ahead. Internal teams at OpenAI operate by three core memes: being maximally ambitious, being maximally accelerated, and mainlining the product (using it intensely to give feedback).
Knowledge Work vs. Execution: Why Transparency Matters
AI excels at structured, testable work. Code works or doesn't; you can verify it. Knowledge work is different. You can't blindly trust a presentation deck or financial model just by looking at the output. You need to see:
- The reasoning and chain of thought
- Citations and sources
- The inputs that led to the conclusion
- How the AI arrived at its answer
This distinction shapes how AI should collaborate on knowledge work. Rather than just returning a final answer, AI should show its work—letting users verify accuracy, understand assumptions, and catch errors. The interface should support this transparency, not hide it. As human collaborators join the workflow, the AI's role must also evolve to be a transparent partner, not a black box.
Creative Writing as Thinking: When to Keep Humans in the Loop
Not all writing should be automated. The distinction between writing as thinking and writing as reporting is crucial.
Writing as reporting (status updates, summaries, structured communications): Automate freely. AI excels here, saving time.
Writing as thinking (briefs, strategies, problem definitions, creative vision): Keep this human. The act of outlining, drafting, cutting, and iterating—wrestling with an idea on the page—is how thinking actually happens. Using AI to generate the first draft or polish the prose short-circuits that process and atrophies thinking ability.
Briefs written through this human thinking process, shopped around for feedback, poked at by colleagues—these become stronger, more defensible, and more truly yours. The cognitive work of writing is the work.
Persistent AI Coworkers: The Near-Term Reality
The third era isn't theoretical. It's starting now. Early signs at OpenAI include:
- Agents that persist across sessions: Work with them one day, come back the next, continue where you left off
- Shared agent workflows: Multiple people working with their own agents toward a common goal, agents communicating with each other
- Local and cloud infrastructure: Agents running locally (with access to your files and data) and in the cloud (with access to external systems, databases, and integrations)
The technical foundation matters as much as the intelligence. An isolated cloud agent is useless if it can't access your Slack, Google Docs, company database, or third-party systems—just like a colleague locked in a room with no access to tools would be ineffective.
What Remains Fundamentally Human
As roles shift and tools evolve, certain things remain irreducibly human:
- Accountability: Someone must own the outcome—whether it's good, whether it's what was needed
- Expression and artistry: The opinionated choices about what to build, how it should feel, what vision it serves
- Care and relationship: How teams elevate each other, learn together, and work toward shared goals
- Steering direction: At higher and higher levels of abstraction, humans must still decide where we're pointing
These aren't limitations on AI's rise—they're the foundation of what makes building meaningful.
Conclusion
AI's third era is arriving not as a single breakthrough but as a gradual shift in how we work. The move from agents you consult to coworkers you collaborate with day-to-day will reshape not just how work gets done, but who succeeds.
The teams and individuals who thrive will be those who master the balance: using AI to expand what's possible, staying ambitious enough to attempt more, and keeping their intuition sharp. Automation isn't the goal. Amplifying human creativity and vision is.
Try the tools—build a site, visualize your data, take a risk on something bigger. But remember: the AI is rowing. You're steering. That's where the real work happens.
Original source: AI’s third era: the rise of persistent AI coworkers | Tara Seshan (OpenAI’s product lead)
powered by osmu.app