Sam Altman rarely describes the future of artificial intelligence (AI) in narrow terms. When asked where the technology ultimately leads, he moved quickly beyond software, portraying AI as a force capable of reshaping physical infrastructure, accelerating scientific discovery and driving broad economic growth.
“I can imagine billions of humanoid robots building more data centers and mining for material and building more power plants,” the OpenAI chief executive said on Tuesday (Feb. 3) during a conversation with Cisco’s CPO Jeetu Patel at Cisco’s AI Summit. “I can imagine just the economy growing at an unprecedented rate if there’s all sorts of incredible new services and scientific discoveries happening.”
But the image anchored a broader theme that surfaced repeatedly throughout his remarks: AI’s trajectory is no longer confined to narrow use cases or productivity gains. It is moving toward systemic change, even as most enterprises remain structurally unprepared to absorb it.
The Upper Limit: Fully AI Companies
When the conversation turned to where today’s AI systems ultimately lead, Altman focused less on near-term deployment and more on capability. “
The upper limit, I think, is full AI companies,” he said, describing organizations where AI systems are not tools layered onto workflows, but active participants in how work gets done.
A key inflection point, he added, is the shift from models that generate outputs to agents that can operate computers directly.
“Code is really powerful,” Altman said, “but code plus generalized computer use is even much more powerful.”
Agents that can navigate browsers, applications and authenticated environments allow AI to complete tasks end to end, rather than stopping at recommendations or drafts. Once that interaction model is experienced, Altman suggested, it becomes difficult to think of AI as a passive system waiting for human prompts.
He extended that logic beyond individual workflows to coordination and collaboration. Altman described the possibility of entirely new interaction models in which agents communicate with one another on behalf of humans. He positioned it as a natural outcome of increasing capability: interaction systems designed primarily for machines to exchange information and coordinate tasks, rather than for humans to manually manage those exchanges themselves.
Security and Data Access Remain the Hardest Problems
Despite rapid progress in AI capability, Altman said that the most binding constraints are no longer technical.
“How are we going to balance the sort of security and data access versus the utility of all of these models?” he asked. Existing security and permission systems were designed for human users making discrete, intentional requests. They are poorly suited to always-on agents that observe continuously and act across systems.
“It feels to me like there is a new kind of security or data access paradigm that needs to be invented for this,” Altman said. Until that happens, he said, organizations will continue to limit AI deployment even as capabilities advance.
Altman repeatedly returned to what he described as a widening gap between what AI systems can do and what enterprises are prepared to adopt.
That gap, in his view, is driven less by the technology itself than by unresolved questions around governance, security and data access. The result is slower adoption, even for tools that already exist.
“Figuring out how to set up enterprises such that they can quickly absorb these new tools,” Altman said, without years lost to internal friction and access debates, “feels very important.”
Delays, he warned, could carry competitive consequences. Companies that fail to adapt their structures fast enough may find themselves falling behind, not because the technology is unavailable, but because they are not ready to work alongside it.
“I don’t want to make it too dramatic of a prediction,” Altman said, “but I think the companies that are not set up to be able to adopt, let’s call them AI co-workers, very quickly, will be at a huge disadvantage.”