Home/ Resources/ Essay

Five things we don't build

A short, honest list of AI systems we've been asked to build and declined. For each, why, and what we recommend instead.

CategoryEssay
PublishedFebruary 2026
Read time6 min read
ByMarion Lowell

01. Website chatbots

We turn down roughly two of these a month. They're popular because they're visible, and because a CEO can point at one in a board meeting. They almost never produce measurable business outcomes. If you want a chatbot, buy Intercom. Don't pay us $80K.

02. "AI strategy" engagements

Three-month paid explorations of "where AI could fit in your business." No build. We decline these. Either we can see the two line items in the first call, or you don't need us. There is no strategy engagement worth your money that doesn't end in a system being shipped.

03. Proofs-of-concept that don't ship

A POC built with no intention to ship is a demo. Demos are cheap and we don't charge for them, they're the guarantee.

04. AI that replaces an entire department

We'll rewire a department. We won't replace one. Businesses don't absorb change at that rate, and the systems that tried to do this in 2023 are cautionary tales. Start with the worst-fit role, systematize it, redeploy headcount into higher-value work. Six months later, do the next.

05. Generic LLM wrappers

"Our version of ChatGPT, for our company." These rarely produce real leverage. What does: retrieval-grounded assistants that know your specific documents. That's a RAG build, not a wrapper. We do those.

Ready when you are

Book the 45-minute diagnosis.

No pitch. No fee. A written map of the two line items bleeding your business.