The Real AI Crisis at Disrupt 2026: Enterprise Trust, Not Technology

Anthropic and OpenAI are about to expose what everyone's been avoiding—why AI deployments stall, how security frameworks are fundamentally broken, and why the SaaS playbook may be obsolete.

The AI conversation has been stuck in theory for two years. Everyone talks about what AI could do, what might disrupt, what could happen next. But almost nobody is talking about what actually happens after companies deploy it—where it works, where it completely stalls, and why some enterprises are still running pilots eighteen months later. That's about to change. Watch the full video or listen to the podcast episode for deeper context.

At TechCrunch Disrupt 2026 (October 13–15 at Moscone Center in San Francisco), the AI Stage is returning, presented by Google for Startups, and this year it's not chasing hype. It's interrogating the problems founders are actually losing sleep over.

The Business Model Collapse

AI didn't just change how startups build products—it broke the business models underneath them. How do you price a product when the underlying model is becoming a commodity overnight? How do you defend a moat when your competitor can spin up the same capability with an API call? These aren't edge cases. They're the central tension every AI-first company is facing right now.

The Security Time Bomb

Then there's agentic AI, now making autonomous decisions inside the most sensitive enterprise systems on the planet at speeds traditional security frameworks were never built to handle. The kicker: agent security wasn't designed from the ground up. It's being retrofitted in real time while the agents are already live inside enterprise infrastructure.

Application-level permission models are fundamentally flawed. Not slightly outdated—fundamentally flawed. That means enterprises are essentially rebuilding the basic elements of cybersecurity from scratch, under pressure, while the tools they're securing keep getting more autonomous by the month.

Who Actually Knows What's Happening?

Cat de Jong, Head of Applied AI at Anthropic, works directly with enterprises deploying Claude into critical workflows. Not the sanitized case study version—the real one. Where deployments succeed immediately, where they quietly stall, and what actually separates companies extracting real value from those stuck in pilot purgatory.

On the OpenAI side, Tara Seshan, Head of Productivity, is tackling something almost stranger: a job category that didn't exist two years ago. GTM engineering. Today it's one of the fastest-growing roles in the industry, with independent practitioners building million-dollar businesses around a discipline that had no name in 2024.

The Real Problem Isn't Technology

The technology got sorted out already. What's unresolved in 2026 is trust—at the pricing layer, the security layer, and the go-to-market layer, all at once.

Every session at Disrupt keeps circling back to the same underlying issue: trust. Which pricing model can customers actually trust to be sustainable? Which security architecture can you trust versus the one you literally cannot afford to touch? Who do enterprises trust enough to let AI touch their most sensitive systems?

Ric Smith from Okta is going deep on why infrastructure-level rebuilds matter more than app-level patches. Chet Kapoor from AWS, Katie Moussouris from Luta Security, and Wendy Nather from 1Password are unpacking why the cloud just got complicated now that AI is operating autonomously inside systems never designed for that level of trust. Arsalan Tavakoli of Databricks is arguing that the enterprise isn't broken—your assumptions about it are.

If you're a founder, this isn't a "watch and get inspired" event. It's a diagnostic. Whether your problem is pricing before commoditization, closing security gaps an agent could exploit, or building a go-to-market motion that had no name two years ago—this is where the people actually inside these companies get specific about it, on stage, in front of over 10,000 startup, tech, and VC leaders.

The conversation about what enterprise AI actually requires is happening whether you're there or not. The only question is whether you're in the room when it does. Current pricing windows are closing, and early-bird discounts of up to $200 off tickets end soon.