What AI can and cannot do for your business.
AI is genuinely useful for drafting, summarizing, pulling data out of documents, assisting with code, and running agents that take defined actions in your systems. It is not a mind that reasons for you, it can be confidently wrong, and it delivers nothing until someone decides what is actually worth building. Knowing the line between those is what saves the budget.
Where it is genuinely useful
- Draft, summarize, and rewrite text against your material
- Pull structured data out of messy documents and calls
- Assist with code, first drafts, and repetitive analysis
- Answer questions from your own documents and records
- Run agents that take defined actions inside your systems
Where it falls down
- Guarantee it is right; it can state a wrong answer confidently
- Truly reason or plan like a person on novel, high-stakes calls
- Work safely on sensitive data without real access controls
- Replace human judgment where being wrong is expensive
- Deliver value without someone deciding what is worth building
The expensive myth
“We just need to buy some AI.”
An MIT Media Lab report on enterprise AI in 2025 found only about 5% of pilots drove real revenue; most stalled with little measurable profit. The report's own read was a learning gap, not bad models: generic tools pointed at the wrong work, layered on top of old processes, with no one owning the outcome. AI pays off when it is aimed at a real, repetitive, expensive problem, with a human accountable for the result. That aiming is the work we do first.