07/01/2026
Just completed one of our most challenging AI agent builds so far.
Created a procurement agent that searches and checks whether requested orders are in stock or out of stock by browsing supplier websites using catalogue numbers and CAS numbers.
The Problem:
The team had to manually check if suppliers had items in stock. If not, they’d check another supplier, then another, until they either found it or confirmed it wasn’t available anywhere. Time-consuming, repetitive work.
What We Built:
An AI agent that:
∙ Runs automatically at scheduled times
∙ Searches supplier sites using catalogue and CAS numbers
∙ Uses AI (including Perplexity) to scrape pages and check stock availability
∙ Delivers results with direct purchase URLs
∙ Allows the team to communicate with the agent to request specific order searches if anything gets missed
What Made This Hard:
Perplexity couldn’t reliably search every page or return accurate data, so we had to build manual search fallbacks. That’s where the project got significantly larger than expected.
Different supplier websites had completely different URL structures when using CAS and catalogue numbers. We had to account for so many different scenarios, different parameters, different search methods.
Every solution uncovered new problems. Classic learning experience for early-stage builders.
Scoping got harder as we went deeper. Complexity revealed itself layer by layer. But that’s exactly how you learn what’s possible and what’s not.
The Result:
Agent is built, tested, and ready. Scheduled automation handles what used to be daily manual work.
Projects like this are the most valuable learning experience - they teach you how to scope better, anticipate complexity, and deliver real systems. That knowledge carries into every future project.
Excited to move onto the shipping side of the company next and see how we can automate those processes.✅
