Building Trusted AI at Scale Cities Startups & Digital Sovereignty – Keynote Jeetu Patel President and Chief Product Officer Cisco Inc
20 Feb 2026 12:00h - 13:00h
Building Trusted AI at Scale Cities Startups & Digital Sovereignty – Keynote Jeetu Patel President and Chief Product Officer Cisco Inc
Summary
Jeetu Patel opened the AI summit by congratulating the organizers and the Indian government for hosting a massive event and noting that AI development is moving faster than ever, now entering its second phase of autonomous agents and soon a third phase of physical AI [3-5][8-13]. He argued that AI has already transformed software creation, citing Cisco’s first product built entirely by AI without human-written code, which he says will turn the innovation curve from exponential to vertical and requires AI to be present in every loop rather than humans [14-16]. Patel identified three fundamental constraints that could hinder AI progress: insufficient infrastructure, a lack of contextual information for agents, and a deficit of trust in AI systems [18-23][26-33]. The infrastructure constraint stems from global shortages of power, compute, network bandwidth, and data-center capacity, which he describes as “oxygen for AI” and warns will limit AI’s potential if not addressed [20-24]. He further explained that as AI shifts from chat-bot style spikes to continuously running agents, the demand for steady-state compute will change the required infrastructure profile, necessitating new “token generation factories” designed for this behavior [38-41]. The context gap, he said, arises because agents lack the trillions of tokens of real-world context humans process, leading to poor decisions when operating without sufficient information [27-30][42-46]. To close this gap, Patel proposed enriching models with proprietary enterprise data and the growing volume of machine-generated time-series data, and redesigning workflows so that agents, not humans, adapt to the processes [51-66][67-72]. The trust deficit, according to Patel, is no longer about incorrect answers but about agents taking harmful actions, requiring protection against jailbreaks, prompt-injection, and data-poisoning as well as runtime guardrails that can intervene dynamically [74-83]. He claimed that Cisco is developing solutions across all three areas by building networks for agents, creating context-rich environments, and implementing security and observability throughout the stack [84-91]. Patel highlighted India’s strategic advantage for AI, noting its large pool of young talent, robust digital identity and payment infrastructure, and massive scale that provides abundant data for training models [98-103]. He emphasized that this combination positions India to not only adopt AI but also shape its global direction, and expressed optimism that collaborative, secure AI deployment can address humanity’s biggest challenges [94-105][106]. The discussion concluded with a call for an ecosystem approach to keep AI safe and secure, underscoring that progress will be measured by each nation’s ability to generate tokens safely, securely, and efficiently [36][105]. Overall, Patel’s remarks framed the urgent need to overcome infrastructure, context, and trust hurdles while leveraging India’s strengths to ensure AI’s responsible and transformative impact worldwide [37][107].
Keypoints
Major discussion points
– Rapid evolution of AI and its societal impact – Patel describes moving from “intelligent chatbots” to autonomous agents and soon “physical AI,” fundamentally re-imagining work across many dimensions [8-13].
– Three fundamental constraints on AI progress – He identifies (1) infrastructure limits (power, compute, bandwidth, data-center capacity) [18-26]; (2) a “context gap” where agents lack the rich, real-time information humans use to make decisions [26-33]; and (3) a “trust deficit” that hinders adoption unless safety and security are built in [31-35].
– Mind-set shift: AI in every loop, not humans in every loop – Patel argues that AI must become an “augmented teammate” embedded in every workflow, flipping the traditional “human-in-the-loop” model [14-16].
– Cisco’s integrated approach to address the three constraints – The company is developing (a) AI-ready network infrastructure, (b) context-enriched data pipelines (including enterprise and machine data), and (c) runtime security and observability to protect both agents and users [84-91].
– India’s strategic advantage and partnership opportunity – Patel highlights India’s large talent pool, strong digital foundations (Aadhaar, UPI), and massive scale of data as key assets for global AI leadership, and pledges collaboration with the nation [92-105].
Overall purpose / goal
The discussion aims to map the current AI landscape, pinpoint the critical barriers (infrastructure, context, trust), propose a new operational paradigm where AI is embedded in every loop, showcase how Cisco is building the necessary technology stack, and rally India’s unique strengths to jointly advance a safe, secure, and globally competitive AI ecosystem.
Overall tone
The tone begins enthusiastic and celebratory, applauding the summit and India’s progress [3-6]. It then shifts to a cautiously analytical stance as Patel outlines the three major constraints and the risks of inadequate context or trust [18-35]. Following this, the tone becomes solution-focused and confident, describing Cisco’s concrete initiatives and the practical steps needed [84-91]. The closing remarks return to an optimistic and collaborative tone, emphasizing partnership with India and the hopeful potential of AI when managed responsibly [98-107].
Speakers
– Speaker 1
Role/Title: Event moderator/host (introducing the keynote) [S1][S3]
Area of Expertise:
– Jeetu Patel
Role/Title: President and Chief Product Officer, Cisco Inc. [S4] (Representative from Cisco) [S6]
Area of Expertise: Artificial Intelligence, networking, enterprise technology
Additional speakers:
– None identified
The summit opened with a warm greeting from the MC (Speaker 1), who welcomed the audience and introduced Jeetu Patel [1-2]. Patel then congratulated Prime Minister Narendra Modi and Minister Vaishnav for delivering a spectacular AI summit that attracted roughly a quarter-million participants [3-6].
Patel framed the current moment as a rapid acceleration of artificial intelligence, describing three successive phases: the early “intelligent chatbots” that seemed magical three years ago, the present wave of autonomous AI agents that perform tasks with minimal human oversight, and an imminent third phase of “physical AI” that will fundamentally re-imagine work across dimensions never before imagined [9-13].
He added that “if you think about what AI is doing, it’s basically forcing us to rethink every assumption that we’ve had in society,” underscoring the broader societal rethink triggered by AI [9-11].
A concrete illustration of this shift is Cisco’s first product that was 100 % generated and coded by AI, with no human writing a single line of code [15]. Patel emphasized that AI-first creation turns the traditional exponential innovation curve into a near-vertical line, effectively collapsing years of incremental progress into a rapid surge [14-16]. Consequently, the paradigm moves from a “human-in-the-loop” model to an “AI-in-the-loop” approach, where AI acts as an augmented teammate in every workflow [14-16].
Patel identified three fundamental constraints that could impede AI’s continued progress.
Infrastructure constraint – The world lacks sufficient power, compute, network bandwidth, memory, and data-centre capacity, which he described as “oxygen for AI” [20-24]. The shift from spiky chatbot-style inference workloads to continuously operating AI agents creates a steady-state demand for compute [39-41]. This requires new “token-generation factories” – large-scale compute facilities that generate the AI inference tokens needed for continuous agent operation – and AI-ready network designs capable of supporting persistent inference loads [38-41].
Context gap – Human cognition processes trillions of tokens of contextual information each second, a richness that current AI agents lack, leading to sub-optimal decisions [27-30]. Patel proposed three complementary actions to close this gap: (i) enrich AI models with proprietary, non-public enterprise data to turn internal intellectual property into a competitive differentiator [51-58]; (ii) feed agents large volumes of machine-generated time-series data such as weather feeds, sensor logs, and system metrics, which are projected to constitute over half of future data growth [61-71]; and (iii) redesign workflows so that AI agents are embedded in every process, adjusting the processes to the agents rather than expecting agents to fit legacy workflows [67-72].
Trust deficit – The primary risk has shifted from providing wrong answers to taking harmful actions [74-76]. Patel called for two layers of protection: (i) safeguarding AI agents from external attacks such as jail-breaking, prompt-injection, tool abuse, and data-poisoning [79-80]; and (ii) protecting the broader world from rogue agent behaviour by deploying dynamic, runtime guardrails that can intervene in real time, moving governance from static documents to live enforcement [81-84].
Cisco is positioning itself to address all three constraints simultaneously. The company is building AI-ready network infrastructure on which AI agents can run, creating context-rich data pipelines that integrate both enterprise and machine data, and embedding security controls with pervasive observability-from GPU utilisation to model performance and agent actions-across the entire stack [84-91][144-152]. This integrated stack is intended to enable every nation and enterprise to safely, securely, and efficiently generate AI tokens, a metric Patel identified as the new benchmark of global competitiveness [36-37].
Patel highlighted why India is uniquely poised to lead in this AI era. The nation boasts a massive, youthful talent pool, one of the world’s largest cohorts of people under 30 [98-100]; a robust digital foundation exemplified by Aadhaar and UPI that provides common identity and payment infrastructure at scale [101-102]; and an enormous population that supplies the data volume AI systems thrive on [103-105]. These assets enable India not only to adopt AI but also to shape its global direction [94-96][153-159].
In concluding remarks, Patel expressed strong optimism that, if humans can confidently delegate tasks to AI within a safe and secure framework, the technology could help solve humanity’s most pressing challenges-disease, poverty, and education-while urging the community to band together as an ecosystem to keep AI trustworthy [170-176][106-108]. He thanked the audience, reaffirmed Cisco’s partnership with India, and closed the session on a hopeful note [177-179].
works without resilient, secure infrastructure is both timely and essential. Ladies and gentlemen, please welcome Mr. Jeetu Patel.
Namaste. I feel very happy to see India’s progress. So firstly, congratulations to all of you for hosting one of the most spectacular AI summits that the world has ever seen with about 250 ,000 attendees. And congratulations to His Honorable Prime Minister, Mr. Narendra Modi, as well as Minister Vaishnav. For actually bringing us all together to talk about what the possibilities of the future are with AI. So what I thought I’d do is I wanted to actually walk you through. where we are today and what the possibilities are and what the constraints are going to be that we need to overcome. But let me just take a step back and say that we are probably moving at a pace that is faster than we’ve ever either expected or seen before with AI.
And we are now squarely in the second phase of AI. So we started with this kind of notion of intelligent chatbots that answered questions for us that felt like magic three years ago. And now we are at this point where agents are conducting tasks and jobs for us almost fully autonomously. And we are actually soon going to go to the third phase, which is physical AI as well. And what this is going to do is fundamentally reimagine work across a multitude of dimensions and vectors that we had never even imagined before. Now, if you think about what AI is doing, it’s basically forcing us to rethink every assumption that we’ve had in society. and I think a lot of these are going to be positive and we also need to be mindful of the downsides that might be there but if you really think deep and long and hard the first thing that I’d say is the modern development process for software development has completely changed and flipped at this point in time where AI is going to in fact at Cisco we have our first product that was 100 % built and coded with AI where there was no human writing a single line of code what that actually has as an implication is that your exponential curve of innovation is almost going to feel like a vertical line and how we need to adjust for that because right now what’s happening is that the rate of change is going to accelerate but as that acceleration is happening what you’ll find is the absorption rate of technology is going to increase and the absorption rate of technology is still not quite at the same level as the innovation rate of the technology itself And so rather than having a human in every loop, which is the way that we’ve thought about it, we need to flip that model and make sure that AI is in every loop rather than thinking about a human in the loop.
And the big mindset shift that’s starting to occur is this notion that, you know, these aren’t just productivity tools. These are going to be augmented teammates into our society where they will be working on behalf of humans for humans to go out and conduct things that we actually need additional capacity for. So then the question to ask is, what could hold progress back for AI? And we think there are three things that could fundamentally be impediments for the progress of AI. The first one is an infrastructure constraint. And what I mean by that is there’s just not enough power, compute, and network bandwidth in the world. Now there’s not enough memory, enough capacity. There’s not enough capacity to build out the data centers.
These are massive constraints and infrastructure is oxygen for AI. So if you don’t have enough amount of infrastructure, you’re not going to be able to make sure that you can fulfill and harness the full potential of AI. So infrastructure is the first big constraint that we see. The second big constraint is this notion of a context gap. If we think of each one of us in our lives, the way that we think about acknowledging, kind of gathering information, is we are taking trillions of tokens of context every second and assessing it in our brains and actually informing ourselves of what we need to do as we move forward. These agents are going to need to have that same level of context enrichment.
And if they don’t, they’ll still make decisions, but they’re not going to be very good decisions. You know? So the second one is this fundamental context gap. And then the third area is a trust deficit. If you don’t trust these systems, you’re never going to be able to use them. And you’ll actually see that there’s an impedance to adoption. as a result of the absence of trust. So you have to make sure that you actually start to think about safety and security at a fundamentally different level than what we’ve seen before. And if you think about what the new metric for success is, the new metric for global competitiveness moving forward is our ability to safely, securely, and efficiently generate tokens for the use of AI.
And every country and every company is going to actually get measured by your ability to safely, securely, and efficiently generate tokens, and that’ll directly impact your economic prosperity as well as national security. So let’s go into each one of these three constraints and talk about what the dynamics are and how do we need to make sure that we overcome them. Because if you look at the pattern on the infrastructure side, specifically with what’s happening with agents, what you’re starting to see is as we move from chatbots to agents, the pattern of inferencing. is going from this very spiky kind of compute, you know, consumption that used to happen to a much more steady state, you know, kind of persistent demand signal that you’re starting to see in the market, right?
And that’ll have a very, very different level of infrastructure requirements than what we might have seen before. And so I think we have to keep that in mind as we’re building out the rest of, you know, these kind of token generation factories that we’re building out. We’ll need to make sure that they can actually accommodate for that second, you know, kind of behavior model rather than the first one. Now, as you go into the context gap, imagine this. Imagine if you’re an ER doctor and imagine that you actually had an unresponsive patient and you had no charts on the patient, you had no history about the patient, you had no symptoms that you knew that the patient was experiencing.
How would you be able to go treat that patient? You might still be able to do it, but you’re going to make a bunch of guesses, right? An agent without context is still going to make decisions, but those decisions might not be the kind of decisions we want that agent to make. And so we have to make sure that we figure out effective and efficient ways to enrich context for that agent, for the AI. And in the absence of that, it’s just going to be forced to guess. And those guesses are as good as you flipping a coin and your head showing up. So how do you close that context gap? And so the way you close the context gap, the first one is these models have been trained on human -generated data that’s publicly available on the Internet.
But we are running out of human -generated data publicly available on the Internet. So now what you’re starting to see is there’s a tremendous amount of enterprise data that’s actually intellectual property of these companies. Can we make sure that we can enrich these models with this proprietary enterprise data for the purposes of that organization so that they can create competitive differentiation? And so the first one is this. No. The notion of connecting enterprise data to AI and agents. The second big area is this notion of enriching agents with machine data. Because right now what you see is most of the data is human -generated data that these AI models have actually been using. We need to make sure that we use machine data.
What does machine data mean? It’s time series data. When something, you and all of us humans, we start our day by actually consuming machine data. We might check the weather. That’s machine data. As you have more and more agents, 55 % of the growth of data in the world is going to be machine data. As these agents work 7 by 24, you’ll have much, much more data and logs, metrics, events, traces that will actually need to be consumed. So that second area of agents being enriched with machine data is going to be critical for these agents actually operating with a sufficient level of context. And third is you have to embed AI in every workflow. You can’t just actually think about that as, I’m just going to use this machine data.
I’m just going to have a tool that augments to my broken process. You have to fundamentally rethink the process that accommodates these agents. agents don’t adjust to us. We have to adjust our process to the agents so that they can actually be effective for us. So that’s the second big area is the context gap. And then the third area that we talked about was this notion of a trust deficit. And in AI, the risk with AI now is no longer that AI is going to give us the wrong answer. The risk with AI is AI is going to take the wrong action. And when you actually start having an AI take the wrong action, the consequences are far more grave than just giving you the wrong answer.
So what do we have to do? So there’s two areas that we think are going to be really important. The first one is we got to make sure that these agents get protected from the world, which means that jailbreaking an agent, having prompt injection attacks, making sure that there’s a level of tool abuse or data poisoning. We have to make sure that we can protect the agents from that happening. And then the second thing we have to do is we have to make sure that we can protect the world from the agents so that the agent starts going rogue. It’s having behavior that’s unintended. that we can actually make sure that we can provide effective guardrails at runtime because no longer is governance a document.
It’s going to be a runtime implementation so that as the agent is working, if you start seeing that the agent’s doing something that’s not going to be in the best interest of humans, we have to be able to inject guardrails into that dynamically at runtime so that we can make sure that that creates a level of trust for the system. Now, it turns out that Cisco is actually building solutions across all of these three areas. And so the way that we think about ourselves is we want to invent and innovate in making sure that the critical infrastructure for the AI era is as simple to deploy, as safe and secure as we want it to be, and as context -enriched as it can be.
So what are we doing? We’re building networks that agents will run on. We’re building contexts that makes it richer for these agents to operate. in a way that allows us to make sure that we can delegate safely and securely to them and feel good about the outcomes that we’re going to get. And we’re going to make sure that we’re going to have security that governs these agents. And all of this is going to be done with a tremendous amount of observability and visibility that says in every layer of the stack, from the way that the GPU is getting utilized, to how the model is performing, to how the apps are being built, to how the agents are performing, we can have observability from bottom to top.
The entire stack. Because if we can do that and figure out that every company, every country is generating tokens in the most effective and secure way, then you’re going to see that there’s progress being made. Now, why is this a tremendous opportunity for India? Because India is not just going to use AI. The way that we’ve seen over the course of the past week is you’re actually helping shape the direction of the entire world with AI. And I think there’s a few… There’s a few key areas… that we should all feel very hopeful for and why India can be a tremendous contributor to AI for the rest of the world. Not just for India, but for the entire world.
The first one is we have a huge talent pool of young, vibrant, intelligent, smart, educated people within India that actually contribute to the workforce. We’ve got one of the largest groups of people under 30 that are actually going out and contributing to the economy. Number two is we have a very strong digital foundation, having common identity with Aadhaar, having UPI. These are things that in India you all might take for granted, but these are very rare to come by in countries, especially at scale. And third, India actually has massive, massive scale. And why is that important? Because AI works best with scale. Because AI works best when you have the most amount of data. right and so the way i think about this is we have a tremendous opportunity ahead of us now the future is not going to be built by ai alone in fact the future gets built when humans can confidently put ai to work and delegate jobs and tasks to ai in a way that we feel safe and secure and so i i’m actually as hopeful as i’ve ever been however i also feel like there’s tremendous amount of risks of these things can go going sideways and so we as a community have to band together and make sure that we actually work as an ecosystem to keep ai safe and secure because if we do that in the right way we’re going to solve the hardest problems that humanity has faced and reduce and hopefully end suffering in so many different areas we might be able to cure the best uh the the hardest diseases that we’ve not been able to overcome we might be able to overcome poverty we might be able to overcome you , know the gaps around education so that can be evenly distributed to people we can make sure that we can improve people’s quality of life So I think there’s a lot to be excited about, but those constraints need to be kept in mind.
And we are so grateful to be partnering with India in this journey ahead. So thank you all. Take care.
“The MC (Speaker 1) welcomed the audience and introduced Jeetu Patel.”
The transcript notes that the session introduced Jeetu Patel from Cisco, confirming his introduction as a speaker [S5].
“Jeetu Patel is President and Chief Product Officer at Cisco Inc.”
The keynote listing identifies Jeetu Patel as President and Chief Product Officer of Cisco [S4].
“Patel congratulated Prime Minister Narendra Modi and Minister Vaishnav for delivering a spectacular AI summit.”
The leaders’ plenary references a Minister Vaishnav (spelled Vesnav) being thanked, confirming a minister named Vaishnav was present; the summit’s opening by a prime minister aligns with the welcome address format [S6] and [S45].
“Cisco’s first product was 100 % generated and coded by AI, with no human writing a single line of code.”
Cisco’s collaboration with OpenAI to embed agentic AI into enterprise software engineering shows development of AI-native code, supporting the claim of AI-generated product, though the source does not state it was 100 % AI-written [S57].
“AI‑first creation turns the traditional exponential innovation curve into a near‑vertical line, collapsing years of incremental progress.”
Discussions of AI-native development treating AI as operational infrastructure illustrate a rapid acceleration of innovation, consistent with this description [S57].
“Infrastructure constraint – the world lacks sufficient power, compute, network bandwidth, memory, and data‑centre capacity, described as “oxygen for AI”.”
A global AI policy framework highlights infrastructure and compute limitations as key barriers to scaling AI, confirming the described constraint [S10].
“Human cognition processes trillions of tokens of contextual information each second, a richness that current AI agents lack.”
The knowledge base discusses limits of language models but does not provide a quantitative figure of “trillions of tokens per second”; the specific number is not substantiated [S63].
“Feeding agents large volumes of machine‑generated time‑series data is projected to constitute over half of future data growth.”
While the knowledge base notes challenges in data discovery and rapid data expansion, it does not give the specific projection that >50 % of future growth will be time-series data [S64].
“The primary risk has shifted from providing wrong answers to taking harmful actions.”
Analyses of language model hallucinations and safety concerns describe a move from factual errors to potentially harmful behavior, supporting this risk shift [S63].
The discussion shows strong internal coherence in Patel’s presentation: he consistently links infrastructure, context, and trust as three inter‑related constraints on AI, and proposes concrete infrastructure, data‑enrichment, and governance measures. The only cross‑speaker agreement is on the importance of resilient, secure infrastructure, echoing the opening remark of Speaker 1.
High consensus on the three constraint pillars (infrastructure, context, trust) within Patel’s arguments, and moderate consensus across speakers limited to the infrastructure theme. This suggests a unified vision for AI advancement that hinges on building robust, secure, and context‑rich ecosystems.
The discussion shows strong alignment between the two speakers on the critical role of resilient and secure infrastructure for AI development. No substantive disagreements were evident; the only divergence is the level of detail, with Patel providing a comprehensive analysis of infrastructure, context, and trust constraints, while Speaker 1 offers a brief introductory endorsement.
Minimal – the speakers are largely in agreement, indicating a cohesive perspective on infrastructure needs, which bodes well for coordinated action on AI development and deployment.
Jeetu Patel’s remarks systematically moved the audience from awe at AI’s rapid evolution to a grounded appraisal of the concrete barriers—infra‑structure, context, and trust—that must be overcome. Each pivotal comment introduced a new analytical lens (phases of AI, flipped development paradigm, structured constraints, vivid analogies, geopolitical metrics, runtime safety, national strengths, and a collaborative vision) that redirected the conversation, deepened its technical and strategic depth, and ultimately framed the discussion as both a challenge and an opportunity for India and the global community. These insights shaped the flow by creating clear turning points, prompting listeners to reconsider assumptions, and ending with a unifying call to collective responsibility.
Disclaimer: This is not an official session record. DiploAI generates these resources from audiovisual recordings, and they are presented as-is, including potential errors. Due to logistical challenges, such as discrepancies in audio/video or transcripts, names may be misspelled. We strive for accuracy to the best of our ability.
Related event
