Can messy technical product data become usable for AI agents?
Partly · Agentic commerce · Hardware data
Hardware procurement is broken because technical product data across B2B manufacturers is highly fragmented, unstructured, and messy. Without clean, structured component data, AI agents cannot autonomously source, compare, or procure hardware — and the entire physical supply chain remains a manual bottleneck.
Context
Every manufacturer uses different schemas, terminologies, and PDF spec sheets. An actuator from Maxon is described completely differently than one from Faulhaber, even when they're functionally interchangeable. This makes comparison, automated sourcing, and AI-assisted procurement nearly impossible without significant human mediation. It's an incredibly hard data alignment problem.
What I did
At Partly, I explored how to parse, structure, and standardize complex B2B hardware catalogues so that agentic commerce workflows could reliably understand them. I treated data as infrastructure: instead of building a smart wrapper on top of messy inputs, I dug into the foundational schemas necessary to make technical data machine-readable and interoperable.
What happened
I mapped out the architecture required to turn unstructured manufacturer data into standardized, actionable inputs for procurement agents. The core insight was that the problem wasn't sophistication of AI models — it was the absence of structured ground truth. You can't build reliable agentic workflows on top of fundamentally broken data.
What I learned
The limiting factor for AI in the physical world isn't model intelligence — it's the lack of structured, high-quality data. Ground truth is messy. This sounds obvious in theory but becomes viscerally clear when you try to make an AI agent source a specific type of linear actuator from three different suppliers using three completely different naming conventions and measurement units.
Why it matters now
Any physical AI or robotics company will inevitably face massive data-structuring challenges. Procurement and supply chains run the physical world, and fixing them is a prerequisite for automation. This is also a huge unlock: whoever structures the B2B hardware data layer correctly could build the infrastructure that enables entire categories of autonomous procurement and manufacturing.