AI automation in e-commerce: when it makes sense and how to avoid the pitfalls

Michał Zabielski pushes back on the hype around AI automation in e-commerce: when it actually pays off, and where the pitfalls sit that aren’t obvious at first glance.

An LLM as a mid-level developer

The core approach: treat a language model like a good but mid-level developer, it writes solid code, but needs precise instructions and business context, not just technical prompts. That’s why the role of a consultant working at the intersection of business and technology is growing, alongside a methodology called Spec-Driven Development: instead of writing code by hand, the developer writes a precise, structured specification of what the LLM should produce.

Real numbers, not marketing

Zabielski is direct about the scale of the speed-up: a 30-50% reduction in cost and time, not a tenfold acceleration, and even less the more complex the product gets. His example is building Endora Commerce: 5 months of work, 60 modules, 25 integrations.

Garbage in, garbage out, only worse

AI doesn’t fix imprecise instructions or poor-quality input data (say, sloppy product descriptions in a PIM), it amplifies the errors, by Zabielski’s estimate as much as tenfold compared to a developer’s own work. Input data quality ends up mattering more than the choice of model.

Three product types, three approaches to specification

Internal tools, one-off products, and market products that need to generalise each require a different level of specification detail, the wider the audience, the more has to be anticipated upfront.

What AI won’t solve

Model hallucinations and the risk of misinterpretation, dependence on API subscriptions and token consumption, a higher barrier for a developer taking over AI-generated code, and the growing cost of distributing and marketing AI-built products, all of it needs to go into the ROI calculation before the decision, not after. LLMs push back the point where automation stops paying off, but they don’t remove that point.

---

Considering AI-driven automation for your B2B platform and want to work out whether it actually pays off? Get in touch with Endora, and we’ll run the ROI numbers before writing a single spec.