Developers already know this, but small business owners are learning it the hard way: generating code or content is fast, and turning it into a reliable system is slow.
The puzzle
Think of any business system (an email program, a bookkeeping setup, a customer database) as a 10,000-piece puzzle. An AI assistant hands you pieces quickly. You still have to know where they go, spot the wrong ones, and rebuild sections when they collapse. For a non-technical owner, that loop can run for weeks.
What the research says
A Harvard Business School and BCG study of 758 consultants found AI boosted speed and quality on tasks inside its strengths, but people using AI were 19% less likely to get the right answer on tasks just outside them. The researchers called it the jagged frontier.
MIT's 2025 GenAI Divide report found buying specialized AI tools succeeded about 67% of the time, while internal builds succeeded about a third as often.
A simple rule
If it is a 10,000-piece system you will use every week for years, start with something already built. If it is a 50-piece one-off (a script, a formula, a quick report), build it with AI and move on.
The longer argument, including where AI fits as a second-in-command, is here: AI may run your whole business one day. Not yet.
Top comments (2)
The 50-piece versus 10,000-piece rule is a useful filter. I'd add maintenance to it. Even a small script turns into a big puzzle once three people depend on it.
Matches what I've seen. Generating the pieces is fast, deciding which pieces you actually need is still slow. I've started writing the boring spec first and only then letting the tooling fill in code.