A close family member runs a private psychology practice, and for years she juggled client records, session notes, invoices and payments across a mess of spreadsheets, a paper diary, and whatever app seemed easiest that particular month. Nothing talked to anything else. So one weekend I decided to actually fix it, using the same AI-assisted approach I now use for clients.
The brief, in her words
"I want to see a client, their upcoming session, and whether they've paid, without opening four different things." That was it. No spec document, no wireframes — just a real problem, described plainly.
What actually surprised me
I expected the AI to be fast at writing code. It was. What surprised me was how much of the weekend was spent on things that had nothing to do with typing: deciding what fields actually mattered, figuring out the right structure for the data, and testing it against real (anonymised) scenarios to make sure nothing broke when, say, a client cancelled and rebooked. That part didn't get faster. That part is still the job.
The other surprise was how naturally it fit around tools she already used, rather than forcing her to abandon everything and learn a brand new system. That mattered more to her than any feature list.
What I'd never leave to AI alone
Anything involving real client data needs a human checking it, every time. I reviewed every piece of it before she touched it — not because I don't trust the tool, but because the AI has no idea what's actually at stake if a psychology practice's client records aren't handled properly. That judgement call is still entirely mine to make, and always will be.
What it cost, roughly
A comparable custom-built system from a traditional software house would likely have started around R100,000–R150,000, with a multi-month timeline. This took a weekend to get a working first version, and a bit more time afterwards refining it based on how she actually used it day to day. That gap is the whole reason I do this work now.
The honest takeaway
AI didn't replace the thinking. It replaced the typing. She now has one dashboard instead of four disconnected tools, and I have a very clear answer the next time someone asks whether this "AI coding thing" is actually useful, or just hype.
If you're sitting with a similar mess of spreadsheets and apps that don't talk to each other, that's exactly the kind of problem this is good for.