Key Takeaways

  • Betaworks founder John Borthwick reports that software construction costs are collapsing toward zero, shifting startup founders from engineering builders to agent steerers.
  • Borthwick built and shipped functional applications himself this year, demonstrating that non-engineers can now direct models to produce working software.
  • The classic lean startup playbook is reversing: instead of manual hypothesis testing, cheap compute lets models search possibility spaces first, mirroring Move 37 in AlphaGo.
  • Betaworks adjusted its twenty-year diligence stack, elevating founder adaptability over initial product specifications.

The Collapse of the Engineering Moat

Borthwick spent decades backing software companies, including an early pre-seed check into Hugging Face. For twenty years, his diligence stack followed a clear order: thesis, product, founder. The core gate was whether a team had the engineering talent to build what they designed.

That filter has broken down. “The cost of building technologies as a technology is trending towards zero,” Borthwick told David Weisburd. Ten years ago, early-stage capital funded expensive technical sprints. Founders lived inside codebases. Today, creation belongs to operators who manage models. “Now they are more often than not sort of steering a set of technologies, a set of agents, a set of AI tools towards sort of a goal.”

Borthwick tested this shift directly. Despite running an investment firm rather than an engineering team, he shipped code and built working applications himself this year. When a venture investor can assemble functional software through agentic tools, writing code is no longer a defensible barrier to entry.

AlphaGo's Move 37 and Data-First Iteration

This collapse in software creation costs upends the traditional lean startup method. For over a decade, venture-backed companies operated like labs: formulate a hypothesis, build an MVP, test it on users, and repeat. That loop was slow because testing was expensive and engineering hours were scarce.

Borthwick argues that the dynamic is now inverted, pointing to DeepMind's AlphaGo and its infamous Move 37 against Lee Sedol. The algorithm did not start with human chess or go theory. It traversed millions of board states to discover a move human masters had never conceived.

“The machine is basically traversing the possibility space and finding new solutions that we as humans have never imagined before,” Borthwick explained. “The old world was, define a hypothesis and then go test it and keep prodding the possibility space. Now you can actually, because crawling the possibility space is so cheap, you can actually just go crawl it and then come back with like hey, Move 37 actually works.”

Instead of starting with human guesses, founders can feed structured data to agent swarms and let them chart functional solutions. The startup becomes data-first rather than hypothesis-first.

Reweighting the Pre-Seed Diligence Stack

Because code is cheap, product roadmaps are fragile. What takes a six-person dev team three months to build today can be replicated by a single operator over a weekend tomorrow.

That reality forced Betaworks to rethink its evaluation criteria. “If you were to say that thesis, product, founder was the stack for me 20 years ago, now it's probably founder is relative higher,” Borthwick said.

When product technical barriers disappear, the founder's taste and steering agility become the primary variable. The winner is not the founder who codes fastest, but the operator who directs agents effectively across changing tool sets.

Why It Matters

Software creation is being demonetized. When the cost of building software approaches zero, valuations based on proprietary code or technical difficulty will fail to hold up. Capital will shift away from technical builders and toward two assets: proprietary workflow data that models can traverse, and operators with rapid distribution channels.