LinkedIn·Wednesday, 5 August 2026·22d ago

You: Can we build this ourselves? AI: Done. First dashboard is ready. A month later: You: Can we connect it to live data? AI: Connected.…

Karel Callens

CEO & Founder at Luzmo

You: Can we build this ourselves? AI: Done. First dashboard is ready. A month later: You: Can we connect it to live data? AI: Connected. You: Can users edit it? AI: How do you want editing to work? Do you have a wireframe? Which parts can they change, and does it save per user or for everyone? You: Can we show what produced every number? AI: I can add that. Do you want me to log the query, the prompt, or both? And for how long? You: Can it use the same permissions in the product, Slack and ChatGPT? AI: Yes, I can start building the auth layer for all three. A few questions first: does Slack inherit the product role or the user's own? Do we sync roles or map them per surface? And what happens when someone's role changes in one place? You: Can we audit what every user asked and fix the context when the answer is wrong? AI: I hit something. Two users asked the same question and got different numbers. Both queries are valid. Which one should be correct? You: ... Many tokens and a lot of lines of code later: Two engineers are on it full time. Neither is on your roadmap anymore. The person who built the first version has left. The number in the dashboard doesn't match the invoice and nobody can explain why, so sales stopped using it. Your biggest customer's security team is asking about audit trails in a questionnaire, and now it has a deadline. The dashboard is no longer a dashboard. It's an analytics product your team now owns, with no PM, no docs and nobody who wanted the job. That is the new build-versus-buy trap. The first step is an afternoon. Self-service and auditable is a year, and you only find that out from the inside. And by then, switching to a vendor feels like admitting the last year was wasted. So you keep going. That is the expensive part.