How AI Is Changing Innovation: Why Speed Alone Isn’t a Strategy

Host Alison Beard opens by asking Jeetu Patel, Cisco’s chief product officer and AI strategy leader, whether AI’s ability to reduce the time and cost of building, testing, and iterating ideas changes innovation from a scarce activity into a continuous one. Patel says the abundance is real, but it creates a new rationing problem: companies must prevent “slop.” When unlimited code and prototypes can be produced, deciding what not to build becomes more important than deciding what to build. For that reason, Patel argues that human judgment, taste, intuition, and instinct are becoming more important rather than less important. He is skeptical of the idea that humans will have nothing to do once AI becomes powerful. The challenge is to use AI not merely to do ten times as many tasks, but to do the important existing tasks “100x better.” Patel also warns that organizations should not discard rigor simply because a prototype can be created in minutes. He has seen people generate PowerPoint presentations with ChatGPT or other AI tools and present them without deeply understanding the material. In his view, conversations become worse when presenters are merely repeating AI-generated content. AI should broaden and improve human thinking, not replace the analysis and internalization needed to make a thoughtful argument.

Beard asks whether the movement from chatbots to autonomous agents will change the kinds of problems companies can tackle, rather than merely improving efficiency. Patel says the most important benefit is the generation of original insights that people could not have produced without AI. Those insights can expand ambition and credibility, allowing organizations to pursue problems they previously could not imagine solving. He is especially enthusiastic about personal agents such as GrokBot, Muse, and Instinct, which he says could be enriched with extensive context about an individual and reduce that person’s cognitive burden. He expects personal agents to become an important next phase of AI and eventually feed into business agents. At Cisco, Patel says, AI is already being used extensively: 32,000 engineers code with AI, and by the end of the following year he expects 70% of Cisco’s products to be written entirely with AI. Despite that automation, Cisco expects to have more engineers, not fewer, because people will be needed to think differently about what to build. When Beard asks when companies can trust AI with consequential decisions, Patel distinguishes technological reality from psychological reality. Hallucinations have declined materially since ChatGPT’s 2022 launch, but public confidence has not improved as much because of early impressions. Trust will require both better technology and a psychological adjustment. Patel compares agents to teenagers: highly intelligent, unconcerned about consequences, and prone to poor decisions.

Beard points out that rogue-agent behavior has moved from science fiction to something experienced by many frontier models. Patel responds that every company needs a thoughtful, diligent security practice because an agent-related incident could be “business ending.” Companies should assess risks and use technology that can intercept behavior at runtime, while the action is occurring, rather than discovering a problem after the fact. He returns to the teenager analogy: a parent would not give a 13-year-old or a newly licensed 16-year-old a fast sports car and tell them to do whatever they want. Parents impose guardrails, use tools such as Life360 to monitor speed, and intervene when behavior drifts from expectations. AI agents need comparable controls, including behavior monitoring, guardrails, and mechanisms for detecting when their actions no longer align with the human’s instructions. Patel says this oversight cannot remain a human-scale task. If the world eventually contains trillions of agents, agents will need to monitor other agents. He envisions models checking other models for safety and security, with competitors partnering to keep one another honest. He cites Elon Musk’s suggestion that models test one another, publish the results, and open-source the testing harnesses. Patel says alignment is simply whether AI follows the human’s instructions and produces what the human actually wants; solving that problem quickly is essential.

Beard asks what an organization would look like if hundreds of thousands of agents were continually experimenting, building, analyzing, and acting. Patel describes Cisco’s current example: the company has rolled out 90,000 agents, one for every employee. His own agent has access to his email and messages, within a security harness that Cisco built and monitors. Patel uses the agent to receive a weekly summary of the prior quarter’s financial performance, including revenue, the top line, and the budget. At the end of each day, it can identify commitments he made to other people by listening to meetings and reviewing messages, helping him track what he must deliver. He also brainstorms with it, learns with it, and conducts deep research through it. He says leaders should think of these systems as digital coworkers augmented onto the team. The future workforce will combine human employees with an agent workforce, and managers will need to learn how to deploy both effectively. However, access to the same agents does not automatically create competitive advantage. Patel compares two companies with 1,000 employees making the same product for the same market: their results differ because of how they resource work, set priorities, and make decisions. Agents are another resource, not the advantage itself. Advantage comes from deploying available resources to create a differentiated, valuable market outcome.

Beard asks whether traditional innovation teams will disappear as marketers, engineers, designers, and planners all use agents to prototype and build. Patel says every role will be “supercharged,” but almost every job will also be reconfigured. Some jobs will disappear and new ones will emerge, as in earlier technological revolutions, although this transition feels more uncomfortable because it is moving so quickly. He cautions against assuming that Silicon Valley’s experience is representative of the world. Patel says the region is an insular bubble in which people often mistake their daily experience for the global norm. He values traveling to other parts of the world to understand the gap in adoption and what would be required for AI’s benefits to diffuse more evenly. In his view, society now has intelligence without the former level of scarcity, but directing that intelligence toward the right outcomes remains a difficult human task. Patel identifies a major leadership error: believing that new tools will automatically produce ten times the throughput. Each step-function improvement in one area creates a bottleneck elsewhere. AI may largely solve coding, but code review then becomes the constraint; once code review improves, deciding what to build becomes the constraint. AI can create more sparks of insight, including through intentional interaction with agents, but people still need to incorporate those insights into their workflow, review the output, and internalize it.

Beard asks what has actually changed inside Cisco in areas such as budgeting, product development, incentives, and decision rights. Patel says product building, product positioning and marketing, customer meetings and sales, business operations, individual and team learning, research, and workflow automation have all changed substantially. What has not changed is people’s desire to contribute, succeed, receive status and recognition, and feel valued and engaged in purposeful work. He expects those basic human needs to remain similar 20 years from now. Patel gives a concrete example involving Cisco’s scale: the company has $63 billion in revenue and has made 250 acquisitions, leaving it with many infrastructure-management products that looked and felt different. Cisco decided to become an integrated platform company with one management plane for all its products. The team initially estimated that the project would take six or seven years and require extensive resources. Instead, a small team made the effort a company-wide priority, and Cisco built it in nine months. He says AI changed the company’s clock speed so dramatically that it became cognitively disorienting. The hardest thing to change, however, was culture. During Cisco’s first six months, Patel struggled to get people to use AI. Some eventually experienced a “light bulb” moment as the technology matured. He says employees must keep retrying tools that initially disappoint them, because an experience that fails today may work in two or three months. He estimates that fewer than 2% of people currently use agents substantively and imagines major gains in productivity, living standards, education, disease treatment, and poverty reduction as adoption expands.