Everyone wants AI on the roadmap. Few people want to own the failure modes.
By 2024 I was adding AI to project work only when there is a measurable job: summarize this inbox, extract fields from these PDFs, suggest a reply, flag anomalies. Not when the goal is to sound modern on a proposal. My clients are in different countries. The hype sounds the same in every language.
Questions I ask first
- What happens when the model is wrong?
- Who reviews output before it touches a customer?
- What data leaves your systems, and where does it go?
- Can you still run the business if the feature is turned off?
If question four has no answer, you are not ready.
Who I build for
Most of my clients are not training foundation models. They run restaurants, logistics, clinics, trading companies, and small offices that need software they can actually run. They need tools that respect margin and attention.
That is where automation and careful AI help. Not a chatbot on the homepage that nobody maintains.
Further reading I actually recommend
Andrew Ng's work on prompting and practical AI education is worth your time if you want fundamentals, not hype.