03 from Dragonfly and No-Code Vision AI: When to Buy, When to Build
Build or buy a no-code AI platform
A platform that sells computer vision as a subscription, hardware, pre-trained models and a drag-and-drop editor, is a good trade for some cases and a slow trap for others. Which one depends on the case, not the vendor, so this is the test I would run on any of them.
Buy when
- The case is standard: defects on common parts, counting, logistics, the work the vendor's models were trained on.
- Nobody on the team will own a model for years, and nobody plans to hire for it.
- Time to a working pilot is the constraint, for a proof of concept or a first answer for leadership.
Build when
- The case is yours alone: proprietary parts, unusual lighting, anomalies only your line produces. A drag-and-drop editor meets its ceiling there, and you spend the months the platform was meant to save.
- The platform is hardware. At five hundred cameras, leaving means ripping out physical infrastructure, not exporting a file.
- A miss is expensive: medical devices, aerospace, anything where a false negative can hurt someone. You will need to explain each decision and keep the audit trail, and a no-code platform often gives you neither.
Three questions before the pilot
- Is the case standard enough for a platform to handle?
- Will someone on the team maintain a custom model for as long as the line runs?
- What does a wrong answer cost? If a miss is tolerable, buy. If a miss can hurt someone, build, or buy only with an audit trail you control.
The answer is often both: the platform for the routine work, your own engineering for the part where an error is expensive, and one business outcome that both are measured against.
Before you sign
- Your images and labels export in an open format, on request, without a fee.
- The model can leave too: the weights, or an export another runtime can load, written into the contract.
- Pricing for the scale you plan, in writing. “Contact us for enterprise pricing” is where the cost moves once you are locked in.
- An uptime commitment for the devices at the edge, not only for the cloud console.
Buy the standard case. Own the one that can hurt you.