Why AI Pilots Fail, Why 2026 Matters, and How Entrepreneurs Win in the Age of Agents | DisrupTV Ep. 425
Why AI Pilots Fail, Why 2026 Matters, and How Entrepreneurs Win in the Age of Agents
On DisrupTV Episode 425, co-hosts Vala Afshar, Chief Evangelist at Salesforce, and R “Ray” Wang, CEO and Founder of Constellation Research, tackled one of the most urgent questions facing leaders today:
Why does so much AI promise fail to turn into real business results—and what changes next?
Joining them were two voices with very different but highly complementary perspectives on the AI transition:
Vernon Keenan, founder of Keenan Vision and longtime industry analyst, known for his work advising enterprises and hyperscalers on AI strategy.
Nicholas Thorne, co-author of Me, My Customer, and AI and founder of Autos, focused on how AI is reshaping entrepreneurship, venture creation, and small business economics.
Together, the conversation moved beyond surface-level AI hype to unpack why 95% of GenAI pilots are labeled “failures,” what actually blocks adoption, and how AI is quietly reshaping jobs, consulting, and company creation—often without dramatic headlines.
Why 95% of GenAI Pilots “Fail” (And Why That’s Misleading)
A widely cited MIT statistic claims that 95% of enterprise GenAI pilots fail. According to Vernon Keenan, this number obscures more than it reveals.
Keenan challenged the methodology behind the study, noting its overreliance on balance sheets and survey instruments that fail to capture what’s happening inside organizations. His own research at UC Berkeley Haas, based on interviews with roughly 45 ecosystem participants, surfaced a different root cause:
The Real Problem: Activation Energy
Most enterprises underestimate the work required after deploying an LLM. Simply embedding a model into a chat interface doesn’t create value.
Enterprise AI success requires:
Orchestration patterns – agents coordinating tasks across systems, not just responding to prompts
Context assembly – harmonizing data across CRM, ERP, finance, support, and operations
Agentic processes – AI acting inside real workflows, not alongside them
The models aren’t the bottleneck. System design, integration, and organizational will are.
2026: The Year AI Gets Real (But Diffusion Still Lags)
Keenan described 2026 as “the year AI gets real”—not because the technology suddenly appears, but because economic pressure finally forces adoption.
Key dynamics shaping this next phase:
AI agents are already real and generating serious ARR at startups and AI-native vendors
Diffusion remains uneven, especially across the mid-market and small businesses
Enterprises are still learning how to operationalize agents at scale
Over the next five years, Keenan expects a full transition into what he calls the “age of the agent,” where embedded and overlay solutions unlock a new model of growth.
Once optimized, agents can be replicated infinitely at near-zero marginal cost—introducing virtual employee economics and forcing executives to rethink growth, productivity, and headcount entirely.
Quiet Erosion: How AI Reshapes Jobs Without Headlines
Rather than mass layoffs, Keenan warned of “quiet erosion”—the gradual hollowing out of entry-level and cognitive work.
Examples include:
Cognitive commoditization, where much of what MBAs and junior consultants know now lives inside LLMs
AI-first boutiques replacing traditional consulting leverage models
Small teams using agents to do the work of dozens
The real risk isn’t losing your job to AI—it’s losing your company to competitors who adopt AI faster and more effectively.
Tiny Teams, Donkeycorns, and a Million AI-Powered Businesses
Where Keenan focused on enterprise friction, Nicholas Thorne focused on what AI makes newly possible.
AI dramatically lowers the cost of starting a company—but it also raises the bar for differentiation.
Through his company Autos, Thorne uses agent orchestration to help founders:
Generate landing pages, videos, and lightweight apps
Set up CRM, lifecycle emails, and ad tests
Handle payments and operational workflows
His ambition is bold but grounded: enable one million people to build $1M/year businesses, which he calls “donkeycorns”—small, focused, profitable companies that grind like mules and party like unicorns.
Relationship Capital: The Only Durable Advantage Left
In a world where everyone has access to the same models, Thorne argued that relationship capital becomes the real moat.
Winning companies will:
Define themselves by who they serve, not just what they build
Maintain deep, continuous feedback loops with early customers
Use customer insight to prompt, iterate, and evolve faster than competitors
Your rate of innovation isn’t constrained by the model—it’s constrained by how well you understand your customers.
Key Takeaways from DisrupTV Episode 425
AI pilots fail due to lack of activation energy, not bad models
Orchestration and context matter more than raw AI capability
2026 marks the start of a multi-year “age of the agent”
AI erodes jobs quietly through productivity, not mass layoffs
Tiny teams can now compete with legacy firms using agent leverage
Relationship capital is the most defensible asset in an AI-saturated market
Final Thoughts: AI as Electricity, Not Experimentation
Across enterprises, startups, and boardrooms alike, the message from DisrupTV 425 was clear:
AI is no longer a proof of concept
Leaders now demand outcomes, not demos
Agents, orchestration, and customer intimacy define winners
Like electricity before it, AI becomes invisible once it’s essential. Organizations that master activation energy, agent-driven workflows, and relationship-led innovation won’t just survive quiet erosion—they’ll define the next era of work and entrepreneurship.
Related Episodes
If you found Episode 425 valuable, here are a few others that align in theme or extend similar conversations:
- Disrupt Yourself: Personal Growth, Leadership, and Designing Work That Thrives | DisrupTV Ep. 424
- From Drift to Discipline: How AI-Powered Leaders Hone Strategy & Reinvent Enterprise Software | DisrupTV Ep. 421
- Leadership in the Age of AI-Driven Cyber Threats | DisrupTV Ep. 423
Published
Authors
