Cisco president and chief product officer Jeetu Patel says AI agents, which act autonomously on a person's behalf, pose the central trust and security problem for AI right now.
Patel says organisations must continuously monitor agents and stop drift from human expectations, since he likens agents to supremely intelligent teenagers with poor judgment.
Patel argues that AI revenue already exists, citing Anthropic's end-user revenue and Azure and Google cloud revenue, and says agents will reshape business models by aggregating information across retailers.
Cisco's Patel argues security will decide how fast agents spread
Cisco's Jeetu Patel argues that AI's next phase is agents that act on people's behalf, and that trust is what gates it. Agents only become useful once they can touch email, calendars and credit cards, which sharply raises the stakes of their poor judgment. Adoption therefore depends on security technology that watches agents and stops drift as fast as AI itself advances.
Takeaways
Trust is now AI's central problem. Patel says a chatbot answers questions while an agent carries out tasks autonomously on your behalf. Agents are only useful with access to tools, email, calendars and credit cards, and granting that access raises risk dramatically.
Agents act like brilliant, reckless teenagers. Patel describes agents as supremely intelligent, with no fear of consequence and often poor judgment. Organisations must monitor their behaviour continuously and intercept drift from human expectations right away. Cisco is spending billions of dollars on monitoring and safety technology.
Security must keep pace for adoption. Enterprise staff now use chatbots much like a search engine, but agent use is in its very early days, partly from fear. Patel himself uses Muse regularly yet has not moved his checking account onto it.
Usage surges once safety catches up. Pressed on hallucinations and alert fatigue, Patel says safety and security technology must innovate as fast as AI itself. Once it does, user comfort rises and usage will surge. He ties token efficiency directly to an organisation's security posture.
Guardrails are now a prerequisite. A consultant made clients' C-suites and boards sign a commitment to implement AI on a morally just basis before engaging. Patel says embedding an organisation's value system is uncontroversial and essential, especially for regulated firms, because unpredictable models need constant monitoring.
AI revenue is already flowing. Answering who pays for AI, Patel points to Anthropic's end-user revenue for applications and to cloud revenue reported by Azure and Google for infrastructure. Both layers of the stack, in his view, are already being paid for.
AI risk is real, not apocalyptic. Asked whether Anthropic's S-1 risk language is silly, Patel says the risk of things going wrong is considerable and should not be dismissed as hyperbole. He holds that real risk and humanity's survival can both be true.
How agents reshape competition
Asked by Tom Keene whether roll-ups and combinations could create monopoly or monopsony, Patel says the AI market has a fair number of players, not one or two.
Tap a name to go there · tap a line to see why they are linked
On this episode 3
Show 1
Agents will shift business models. A retailer that once sold direct through advertising and upselling may instead face an agent that aggregates information across many retailers and presents it to the consumer.