A conference stage devoted to practical scaling strategies is the most useful thing TechCrunch Disrupt 2026 could offer AI founders, and also the most likely to leave the hardest part unsaid.
The Builders Stage returns to Disrupt 2026, running October 13-15 in San Francisco, and TechCrunch is framing it around founders, startup operators, and investors talking through what it actually takes to build and scale. The stated focus is growth challenges: raising capital, acquiring talent. Around 10,000 founders, VCs, and operators are expected. That is a serious concentration of people who make funding and hiring decisions, and the agenda reflects what those people ask about most.
My angle on this is narrow and unapologetic. I study how agent systems are put together, and the scaling problems I see in agent-based startups are not primarily capital problems or hiring problems. They are architecture problems that present themselves as capital and hiring problems. When a founder tells me they need another round and three more engineers, I usually want to see the call graph first.
Why growth advice lands differently for agent companies
Classic SaaS scaling has a comforting property: unit costs fall as you grow. You amortize infrastructure, your margins improve, and the pitch to investors writes itself. Agent products often invert that. Every user interaction can trigger a variable number of model calls, tool invocations, retries, and verification passes. The cost curve is shaped by how the agent thinks, not by how many customers you have. Growth can make your economics worse before it makes them better.
That changes what “practical scaling strategy” means. For an agent company, the highest-return move in a given quarter might not be a hire or a raise. It might be:
- Cutting the number of reasoning steps per task without degrading task completion
- Replacing an expensive general model with a smaller one on the subset of calls that do not need frontier capability
- Building an evaluation use so you can tell whether either of those changes broke anything
- Caching and reusing intermediate state instead of re-deriving it on every turn
None of these are fundraising activities. All of them determine whether your fundraising story holds up under a diligence spreadsheet.
The talent question is really a topology question
Talent acquisition is on the Builders Stage agenda, and it should be. But agent startups tend to hire against a mental model borrowed from web companies: frontend, backend, some ML. The roles that actually break agent products are less legible. Someone has to own evaluation. Someone has to own the tool layer and its failure modes. Someone has to own observability across non-deterministic execution paths, which is a genuinely different discipline from monitoring a request-response service.
If you do not know which of those functions is your bottleneck, hiring faster just distributes confusion across more salaries. The useful version of a talent panel for this audience would be less about sourcing and closing and more about which competencies an agent company needs at ten people versus fifty.
What to actually ask on the floor
Ten thousand attendees means the hallway conversations carry more signal than most sessions. If you build agent systems and you are going to San Francisco in October, the questions worth carrying with you are uncomfortable ones:
- What is your cost per completed task, and how has it moved over the last two quarters?
- How do you know a prompt or model change did not quietly reduce reliability?
- Where does your agent hand off to a human, and is that boundary designed or accidental?
- What breaks first when concurrent sessions increase tenfold?
Founders who can answer those crisply have done the architectural work. Founders who cannot are usually the ones who most need a stage telling them to focus on growth, and least able to act on it.
Credit where it is due
I do not want this to read as dismissive of the format. Practical conversations among founders, operators, and investors are exactly the right shape for this moment, and the Builders Stage framing suggests TechCrunch understands that the audience is past inspiration and wants mechanics. Capital and talent are real constraints, and pretending otherwise would be its own kind of naivety.
My argument is about sequencing. For agent companies specifically, the money and the headcount amplify whatever system you already have. A well-structured agent with clean tool boundaries and real evaluation gets dramatically better with more resources. A tangle of nested prompts calling each other gets more expensive. Same round, same hires, opposite outcomes.
So go to the Builders Stage. Take notes on the capital strategy and the hiring playbooks. Then go home and read your own traces, because that is where your scaling story is actually written.
🕒 Published: