Stalled AI prototype turned into a working system
Found what was blocking delivery and shipped the workflow in one month.
- 1 month: From a stalled project to a live system
- Desktop app: Packaged and ready for engineers to trial
Before
An energy company had a strong idea: automate a documentation-heavy engineering task that normally took days by hand, with the potential to do it more than ten times faster. But the project was stuck. The team was bolting ad-hoc fixes onto a flaky system with no clear path to launch, and the whole idea was at risk of dying in the build.
The hard part
The team couldn’t tell whether they had a bad idea or had just built it badly. They’d tried to do everything with a single prompt, with no sense of which parts needed a fixed, reliable answer and which parts the AI could write freely, and no real examples from actual engineering work to check the results against.
What got decided
Once the real issue was clear, the path opened up:
- The idea is sound. It’s the way it was built that needs reworking, not the concept
- Stop patching and start structuring: separate the parts that need a reliable, fixed answer from the parts the AI can write freely
- Ship it, which the team did the following month
Result
The blockers were cleared. The working system read and wrote the right engineering files correctly and was packaged into a desktop app ready for engineers to trial. The team also came away knowing how to build and manage this kind of AI themselves. Live within one month.
A stalled AI project doesn’t mean the idea is wrong. Often the most valuable thing you can do is separate a bad idea from a badly built one. Once that’s clear, everything else moves.
Download the PDF