Tag: generative ai

This is a welcome reset for an argument too often conducted with demos and vibes. The authors separate code generation from software delivery, point out that developers spend only a fraction of their time typing code, and reject lines produced as a proxy for value. AI assistance can help, but its effect varies with the task, the developer, the codebase, and the surrounding organization. Buying licenses does not redesign review queues, clarify requirements, or make legacy and regulated systems less consequential.

The practical correction is to measure the whole system. Faster implementation can simply move the bottleneck into verification, coordination, security, or operations; more generated code may increase the amount that experienced engineers must understand. None of that makes the tools useless. It makes "productivity" a claim that needs an outcome, a baseline, and a time horizon. Teams should ask whether dependable changes reach users sooner without increasing defects or hidden maintenance work, not whether the editor produced more text.

Nicolo argues that AI weakens one of SaaS's favorite properties: near-zero marginal cost. A conventional feature can be copied endlessly once built; an AI feature keeps consuming inference for each user and request. At the same time, cheaper software creation gives buyers and competitors more credible alternatives, putting pressure on prices while compute costs push from below. That is a real squeeze, especially for products whose value is little more than a model call behind a polished interface.

Inference is only one line in the ledger, though. Support, security, compliance, integrations, distribution, and the cost of earning trust often dominate the code itself, while model prices may eventually behave more like ordinary hosting. The deeper threat to undifferentiated SaaS is not merely a worse gross margin; it is that a customer can assemble a sufficiently tailored substitute. Durable value moves toward proprietary workflows and data, reliable operations, distribution, and outcomes that remain difficult after implementation gets cheap.

Roge draws a needed line between a generated artifact and a product. AI can produce a convincing interface or feature demo in an afternoon, but a product has users, a market, and a problem important enough to solve repeatedly. The implementation is only one component. Distribution, onboarding, support, reliability, security, and the accumulated knowledge of what people actually need are the parts a localhost reveal cannot show.

YouTube, Google, and Instagram are imperfect tests because their defenses include enormous networks and economies of scale, not just hard-to-write code. A more portable definition is dependability over time: people can discover the thing, trust it with real work, and expect someone to keep it useful. AI lowers the cost of testing an idea, which is genuinely valuable. It does not manufacture demand, differentiation, or commitment. The gap between demo and product is where most of the work moved, not where it disappeared.