What Startups Can Teach Engineers About Thriving Amid AI Coding Agents

As AI coding agents proliferate—used not just by software engineers but increasingly by product managers, designers, and marketers—the role of professional engineers faces an existential question: what differentiates a specialized engineer when nearly anyone can now write code? Ben Greene, building the startup Tessi, argues that survival in this shift mirrors the lessons startups have always had to master through sheer necessity.

Start with the Simplest Version That Works#

The pressure to build fast and scale immediately has never been stronger, yet the fundamental principle remains unchanged. When code becomes cheaper to produce, demand only rises—a phenomenon known as Jevons paradox. As more people gain the ability to build software, requests for more features, faster delivery, and broader complexity multiply.

Yet systems—whether technical infrastructure, product experiences, or business models—follow a timeless pattern. Fred Brooks observed that working systems invariably evolve from smaller systems that already worked. Starting with a flawed belief that a complex system can be engineered wholesale from scratch invites failure. Coding agents amplify this temptation by making it trivially easy to say "yes" to every request and layer on more features before the foundation truly works.

Build What You Actually Understand#

Perfection is impossible; releasing before you feel entirely ready is not weakness but wisdom. The polish added before real-world use often becomes wasted effort, discarded when underlying assumptions prove wrong. But understanding what you've built matters enormously—it's the linchpin of making changes when they become necessary.

AI-powered code generation can construct entire codebases that remain mysteriously opaque to the people maintaining them. When an engineer doesn't comprehend what exists, the only paths forward are wholesale replacement or workarounds that accumulate as technical debt. Crucially, some technical debt is rational: you shouldn't engineer for a ten-year vision when the idea's viability remains unproven. The key is knowing what tradeoffs are being made.

Push Into Unexplored Territory#

If a startup isn't dramatically better than existing alternatives, it vanishes. That requires doing things not yet tried. Yet expertise itself can be a trap—the Einstellung effect (coined by psychologist Abraham Luchins) showed that prior experience narrows perception, locking experts into familiar patterns even when better solutions exist.