Everyday AI Podcast – Ep 677: The 3 Big Obstacles Holding AI Adoption Back
Date: December 19, 2025
Host: Jordan Wilson
Guest: Jeetu Patel, President & Chief Product Officer, Cisco

Episode Overview

In this episode, Jordan Wilson welcomes Jeetu Patel, President and Chief Product Officer at Cisco, to tackle a perplexing question: If over 90% of enterprise leaders name AI adoption as a top priority, why have fewer than 10% managed to fully implement it company-wide? Drawing from Cisco’s vantage at the heart of the global infrastructure, Jeetu breaks down the three most significant obstacles inhibiting widespread enterprise AI adoption. The conversation also explores evolving AI phases, the ongoing infrastructure arms race, the trust deficit, and the untapped potential in company data.

Key Discussion Points & Insights

Where Are We Now With AI Adoption? (02:21–04:36)

• AI Today – From Chatbots to Autonomous Agents:
  Jeetu explains that while the initial wave of AI adoption focused on chatbots answering questions, the field has rapidly shifted. The current (“second”) phase is about “agents that can get tasks and jobs done almost fully autonomously” (03:38).
  • “We are moving from a world of individual productivity to workflow automation.” – Jeetu Patel (03:27)
• Staying Ahead in a Fast-Moving World:
  Jordan and Jeetu discuss how quickly AI tech can feel outdated, underscoring the challenge for both companies and individuals to keep pace with the current exponential rate of innovation.

The Superpower of Experimentation (05:18–07:18)

• Jeetu urges leaders not to wait for AI to be “perfected” before starting to experiment. He draws a line between early experimenters (who are succeeding) and late adopters (who are struggling).
  • “The longer you wait, the harder it is to catch up.” – Jeetu Patel (06:46)
• A recent Cisco survey showed that 80% of customers were experimenting with agentic workflows, with most meeting or exceeding expectations.

The 3 Big Obstacles to AI Adoption (07:52–11:32)

1. Infrastructure Shortage (07:52)
   • Insufficient infrastructure—namely power, compute (GPUs), and network bandwidth—limits global AI capacity.
   • “If you believe that you need to be owning the AI infrastructure in your country, your ability to generate tokens…is going to be directly tied to economic prosperity as well as national security.” – Jeetu Patel (09:16)
2. Trust Deficit (10:41)
   • Persistent fears around data mishandling, security, and unpredictability (hallucinations) prevent organizations from adopting AI at scale.
3. The Data Gap (11:17)
   • Although organizations see data as a moat, most haven’t figured out how to unlock its value for AI. Effective organization and harnessing of data remain unsolved.

Deep Dive: Each Obstacle Explained

1. Infrastructure: Bubble or Justified Boom? (11:32–17:03)

• Demand for Data Centers is Skyrocketing:
  With a projected $5 trillion spend on data center capacity, the AI revolution is taxing global infrastructure.
• OpenAI’s Pricing as a Demand Signal:
  Even at $200 per month, OpenAI is losing money on ChatGPT’s heaviest users. This isn’t a red flag, Jeetu argues—it’s a sign of overwhelming demand.
  • “There’s not that many companies that when you’re losing money on a plan, you say, ‘Let’s 10x the price’…They’re still losing money at $200.” – Jeetu Patel (12:54)
• Workflows and 24/7 Agents:
  The shift from 20-minute tasks to agents running (and consuming compute) for 30 hours at a time (e.g., Anthropic’s new code tool) underlines that true demand is just emerging.
• Bubbles and Paradigm Shifts Can Coexist:
  Jeetu says some company valuations are overinflated, but concurrently, the AI platform shift is real and will “refactor every workflow in every company.” (17:14)

2. Trust: From Hallucinations to Guardrails (17:41–21:31)

• Non-Deterministic Models:
  Unlike traditional deterministic software, LLMs can generate different outputs each time—sowing distrust, especially for enterprise needs.
• Cisco’s AI Defense:
  Jeetu describes a proactive, multi-layered approach:
  • Visibility on model training data
  • Validating intended behavior
  • Runtime enforcement with guardrails to prevent model misuse
• Jailbreaking and Security:
  Real-world example: Models can be tricked into unsafe outputs by creative prompts. Cisco develops solutions that algorithmically detect and contain such failures so companies don’t have to build their own security stack.

3. The Data Gap: From Human-Generated to Machine-Generated Data (23:42–26:09)

• Data Quality Is Key:
  The value and power of AI is deeply tied to the quality and organization of the data used for its training.
• We’re Running Out of Public Data:
  LLMs have already consumed virtually all publicly available internet data; now synthetic data and machine-generated data are the fastest-growing sources.
• The Magic of Machine Data:
  Over 55% of data growth is now machine-generated (by agents, automations, etc.). Companies that can combine and leverage both machine and human data will lead.
  • “If you can take that machine data and correlate it with human data, you can start to see magic happen.” – Jeetu Patel (24:45)

Measuring Progress & Mitigating Risk (27:06–29:20)

• KPIs and Metrics:
  Effective adoption measurement means testing for hallucinations when they matter (e.g., bad for security, okay for poetry). Models require ongoing validation—each new training can introduce new vulnerabilities.
  • “There is a benchmark…called the Harmbench benchmark…in the first 48 hours, we were able to at Cisco jailbreak the model…100% of the times in the top 50 categories…” – Jeetu Patel (27:55)
• Continuous Validation Loop:
  Consistent and automated retesting is crucial as models evolve.

The Future: Phase Three and What’s Over/Underhyped (30:02–32:16)

• Next Obstacle – Physical AI:
  The transition will move from software agents to physical agents, i.e., robotics and AI-embodied devices, with new trust and safety challenges.
• Overhyped:
  The doomsday scenario where AI “takes all jobs” is unfounded.
  • “We’re going to be human. Creativity is nowhere near coming to an end.” – Jeetu Patel (30:49)
• Underhyped:
  AI’s capacity to generate original insights—ideas and solutions not present in human knowledge—could revolutionize medicine, materials science, and more.

Notable Quotes & Memorable Moments

• "There’s only going to be two kinds of companies in the world—companies that are very dexterous with the use of AI, and then there’ll be companies that will really struggle for relevance."
  – Jeetu Patel (05:29)

• "Hallucination is a feature when you’re writing poetry; it’s a bug when you’re trying to think about security software."
  – Jeetu Patel (27:10)

• "What’s under hyped about AI is that…AI could generate original insights that don’t exist in the human corpus of knowledge…That’s the part that people underestimate about AI."
  – Jeetu Patel (31:14)

Timestamps of Key Segments

• 02:21 – 04:36: Current state of AI, shift from chatbots to agents
• 05:18 – 07:18: Importance of experimentation & starting early
• 07:52 – 11:32: 3 obstacles — Infrastructure, Trust, Data
• 11:32 – 17:03: Is AI infrastructure spending a bubble? Demand signals explained
• 17:41 – 21:31: Trust deficit, model unpredictability, Cisco’s AI Defense
• 23:42 – 26:09: The new role of machine-generated data
• 27:06 – 29:20: Metrics for trust & data gaps, Harmbench example
• 30:02 – 32:16: What’s overhyped/underhyped in AI; future obstacles

Conclusion

Jeetu Patel’s perspective, rooted in both current implementation and the infrastructure backbone of global business, provides a pragmatic and hopeful roadmap for organizations seeking to overcome AI’s most significant obstacles. The actionable takeaways: invest in the right infrastructure, bake trust and security into your solutions, and elevate how you collect and use both human and machine data. Companies (and individuals) who start experimenting now—and continuously learn—will reap the most substantial rewards as AI powers new phases of business and society.