$21 million. That’s what Runable just raised on the pitch that AI agents can not only build businesses but grow them. I bring that number up first because it explains the mood heading into TechCrunch Disrupt 2026 better than any speaker announcement does. Money is flowing toward agents that promise to run things, not just draft things. And the conference lineup is starting to reflect it.
The headline news is that Anthropic and OpenAI are both taking the AI stage at Disrupt 2026. Amjad Masad, CEO and co-founder of Replit, is on the Disrupt Stage. Applications are open to host Side Events, and there’s an Audience Choice vote deciding which sessions make it to the TechCrunch Founder Summit. That’s the full set of confirmed details, and I’m not going to pad it out with imagined announcements.
Why two frontier labs sharing a stage is less interesting than it sounds
I test AI tools for a living, which means I’ve watched a lot of conference talks that aged badly within a quarter. Frontier lab appearances follow a predictable shape. There’s a demo, a gesture at what’s coming, a careful sentence about safety, and a closing note about how much faster everything is getting. Useful for reading strategy. Close to useless for deciding what to actually put in your workflow next month.
That’s not cynicism about Anthropic or OpenAI specifically. It’s the structure of the format. A lab on stage is speaking to investors, enterprise buyers, policymakers, and developers at the same time, and the resulting message has to be vague enough to serve all four. You get direction, not detail. Direction is worth something. It’s just not worth what the ticket costs.
What would make these two sessions genuinely worth the time is specificity about failure. Where do their agents break down on long-horizon tasks? What percentage of multi-step runs need human intervention? What does the cost curve look like when an agent works for six hours instead of six seconds? Those are the questions that determine whether any of this survives contact with a real business. I’d love to be surprised.
The Replit slot is the one I’d circle
Masad runs a company that has been shipping agentic coding into the hands of non-engineers for a while now, at a scale where the rough edges are impossible to hide. That’s a different kind of stage presence. When your users are actively building and breaking things all day, you tend to have opinions grounded in support tickets rather than roadmaps.
There’s also an honest tension in that business worth hearing about. Agentic coding tools sell speed and accessibility. They also generate code that somebody eventually has to maintain, secure, and explain. Anyone selling agents to people who don’t write software has to have a position on that tradeoff, and hearing it stated plainly on a stage is more valuable to me than another capability demo.
What the funding signal actually tells us
Back to that $21 million. The framing around Runable is the tell. “Building businesses” was the previous pitch. “Growing them” is the new one. That’s an escalation in scope from a system that produces artifacts to a system that owns outcomes over time.
I’m skeptical, and I want to be clear about why. Building something is a bounded task with a checkable result. Growing something requires judgment across weeks, memory that survives context limits, and the ability to notice when a strategy stopped working. Those are exactly the areas where current agents are weakest in my testing. Capital moving toward that pitch tells you where the ambition is, not where the capability is.
The parts of the agenda nobody covers
Two items in this news cycle get treated as housekeeping and shouldn’t be. Side Event applications are open, and the Audience Choice vote is live for Founder Summit sessions.
- Side Events are where you get the unpolished version. Smaller rooms, fewer comms people, founders who will tell you which parts of their product don’t work yet.
- Audience Choice voting is a rare moment where attendees shape what gets stage time instead of accepting what sponsorship dollars picked. If you have a view on which conversations matter, that’s the lever.
My honest read on Disrupt 2026 so far: the frontier lab sessions will draw the crowd and generate the headlines, and the actual signal will come from operators talking about what broke. The gap between agent marketing and agent reliability is the defining problem in this field right now, and it doesn’t get closed by a keynote. It gets closed by people who’ve shipped, watching their systems fail in public, and being willing to describe it.
I’ll be watching for anyone brave enough to do that. If it happens, you’ll read about it here. If it doesn’t, you’ll read about that too.
🕒 Published: