Custom software used to need a business case. Months of build time meant you only wrote software for problems big enough to justify it. Everything smaller got a spreadsheet and a workaround.
That maths is gone. With AI doing the assembly, a working tool — a tracker, a booking flow, a report generator — ships in hours. Not a demo. Deployed, on a real URL, doing real work. I’ve shipped a second-market version of a working product in a day.
Most people haven’t updated their instincts to match. Three that changed for me:
Build for one client. Software economics used to demand many customers to spread the cost. Now the cost is an afternoon, so software for exactly one client’s workflow is viable — and it fits them better than any off-the-shelf product could. Charge for the outcome, not the code.
Build to throw away. When rebuilding is cheap, being wrong is cheap. Skip the long requirements document — build the rough version, put it in front of the user, and let the prototype be the spec. Version two costs less than the meeting to plan it.
Say yes to small jobs. The internal tool nobody would quote for, the one-off report format, the little automation between two systems — these were never worth custom software. Now they are. The backlog of “too small to build” is where the easy wins sit.
One caution: speed moved the bottleneck, it didn’t remove it. Building is cheap; knowing what’s worth building, and verifying it actually works, is not. A wrong tool shipped in two hours is still a wrong tool. The judgment — what to build, whether it’s right, whether you’d stand behind the output — is still the job.
The scarce skill is no longer construction. It’s selection and verification. Spend your time there.
Thoughts on building intelligence services, automation, and business in Malaysia.
More notes