OpenAI President Greg Brockman: Our Plan To Merge Chat And Agents

Summary

Greg Brockman says OpenAI’s core goal is not just a better chatbot, but a persistent agent or “personal AGI” that can use tools, access context, and accomplish tasks on a user’s behalf. He argues early plugin efforts failed because models lacked memory and enough context, whereas current systems can connect to hundreds of tools, file systems, and enterprise workflows. Brockman frames trust, permissions, observability, and spend controls as central requirements for agentic AI, especially as OpenAI pushes further into enterprise use cases. On the competitive and infrastructure side, he says no layer of the stack disappears, compute remains scarce, and OpenAI is investing in its own chip strategy because demand will outstrip supply. He also dismisses the idea that near-term progress will come from a sudden discontinuity, instead expecting cheaper access to today’s intelligence followed by a better frontier model, and he is especially bullish on AI in health, where he says usage is already massive and becoming standard.

Notes

• OpenAI’s target product is AGI-like software that goes beyond chat and can act for users.
• The intended interface is nearly invisible: a persistent entity that accomplishes goals with little UI.
• Brockman describes the product shift as moving from conversational intelligence to agents that do work.
• OpenAI already uses agentic workflows internally for nontechnical staff.
• Tool use now spans Gmail, Slack, calendar, and event workflows.
• He says delegated access is already real: the model can draft or send email depending on permissions and trust.
• Early ChatGPT plugins failed because the models were not capable enough yet.
• The original bottleneck was tiny context windows and effectively no memory.
• Modern systems can expose hundreds of tools and even entire file systems to the model.
• Brockman claims current models are now solving unsolved math and physics problems.
• Trust is a first-class requirement in the agent era.
• OpenAI sees partner integrations as mutually beneficial rather than zero-sum.
• He rejects the operating-system framing and says this is a broader technology interface shift.
• Voice is strategically important, but current stacks are still clumsy.
• Today’s voice systems still rely on awkward turn-taking rather than natural conversation.
• He says voice is valuable for work because typing long feedback is cumbersome and interrupts the flow.
• Brockman says scaling laws still hold and most apparent walls turn out to be bugs or implementation errors.
• OpenAI says it has to inspect every layer of the stack because neural nets have few abstractions.
• Compute is described as the binding constraint on the next phase of AI.
• He says OpenAI has been building its own chip program for multiple years.
• OpenAI is pushing enterprise readiness with spend controls and observability.
• He says the company is shifting from model drops to end-to-end business outcomes.
• On pricing, Brockman says OpenAI has historically increased intelligence while lowering the cost of a fixed capability level.
• He expects a one-year cadence where current-quality intelligence gets much cheaper, then a much better frontier model appears.
• Brockman says no layer of the AI stack disappears from the value chain.
• He argues compute, models, and downstream orchestration all remain valuable.
• He says model competition is good for customers, but OpenAI still expects to lead on the hardest problems.
• OpenAI’s models are positioned as especially useful for accelerating science.
• He distinguishes scientific models from conversational agents that book travel or manage calendars.
• OpenAI’s enterprise value lies in domain expertise and workflow design, not just raw model APIs.
• He says regulated industries and education need careful multi-party orchestration.
• Brockman says the main goal is AI adoption in the economy to transform it and uplift people.
• On GPT-5.6, Brockman would only confirm that it is supposed to be coming.
• He refuses to validate rumor details and only says users should expect better, faster, smarter.
• He jokes that Twitter rumors are not reliable.
• Brockman says AI-health use will become standard rather than remain anecdotal.
• He says many people already use ChatGPT for health questions at large scale.
• Brockman says patients are often the accountable decision-makers in healthcare.
• He says AI is already helping him manage his wife’s health conditions.
• Brockman argues AI can improve personalized medicine, diagnosis, and drug discovery.
• He frames healthcare burnout as a major reason AI matters.
• Applying AI to medicine is a personal motivation for him and OpenAI.
• He closes by saying he is very confident in AI’s health upside.