How Ansur's AI Workers Find Suppliers That Aren't in the Database using TinyFish
At a glance
- Company: Ansur, deploying AI workers in manufacturing environments to automate operations
- Use case: Supplier lookup and web research for AI workers running inside mid-market and enterprise manufacturers
- Previously: Exa
- Now: TinyFish Search and Browser
- Why TinyFish: Faster, more accurate web research that drops directly into Ansur's existing agent loop
- The split: Ansur's workers own the manufacturing judgment; TinyFish handles the part where the answer isn't in the customer's systems
Company overview
Ansur deploys AI workers into manufacturing environments to automate the repetitive, mundane parts of operations.
The goal is to move skilled people away from staring at spreadsheets all day and toward higher-value work: talking to customers, managing supplier relationships, and making decisions.
They're starting with mid-market and enterprise manufacturing, where those AI workers operate directly inside existing workflows and systems.
But there's a limit to what those systems know.
The problem
An AI worker is only as useful as what it can see. Most of the time, what an Ansur worker needs already exists somewhere inside the manufacturer's systems. But some of the most useful tasks require information that isn't there yet.
Supplier discovery is a good example.
A manufacturer's database knows the suppliers it has already worked with. It doesn't necessarily know all the suppliers it could work with.
When an Ansur worker needs to find one, it has to leave the customer's database and go out to the open web: searching, browsing, and retrieving enough information to continue the workflow.
That creates a web infrastructure problem underneath the manufacturing problem.
Ansur could build and maintain that layer themselves. But web search and browsing aren't what differentiate their product. The manufacturing context, workflows, and judgment of their AI workers are.
So they use TinyFish for the web layer.
From Exa to TinyFish
Before TinyFish, Ansur was already running the same supplier lookup and web research workflows on Exa.
And it worked.
This wasn't a case of replacing infrastructure that was failing. The question was whether the same workload could perform better on speed and accuracy, without forcing the team to rethink how its AI workers were built.
Testing that was easy because TinyFish fit cleanly into the existing agent loop.
"At the end of the day it's an API switch, so there wasn't much standing in the way of trying it."
- Foudner, Ansur
The integration was effectively drop-in and took an afternoon.
After running TinyFish on its workloads, the team found the results noticeably faster and more accurate based on its own usage and customer feedback. The difference was consistent enough for Ansur to move the workload over.
How It Works
TinyFish now acts as the web layer underneath these AI workers.
The flow is straightforward:
- A worker hits a gap. The task needs information the customer's database doesn't hold, most often a supplier the manufacturer hasn't used before
- TinyFish Search goes out to the open web and finds candidates.
- TinyFish Browser handles the sources that only give up their content inside a running browser.
- The worker keeps working. The information returns to the agent loop, where the worker applies the manufacturer's context and continues the workflow.
This division of labor is important.
TinyFish isn't deciding which supplier a manufacturer should use. The AI worker owns the manufacturing context and the judgment about what to do next.
TinyFish handles the part where the information it needs isn't in the database yet.
Most usage already sits inside live customer workloads rather than internal experiments, so the feedback loop comes directly from manufacturers using these workflows in production.
The Results
The team has seen a consistent signal across its own usage and customer feedback:
- Drop-in integration: TinyFish fit directly into the existing agent loop, with the switch completed in an afternoon.
- Better accuracy and quality: Customer feedback points to stronger results for sourcing and basic data enrichment compared with the previous setup.
- Noticeably faster results: Response times have been faster in production use.
- Running in customer workloads: TinyFish is used primarily inside production deployments rather than internal testing.
- Scales with deployments: As more AI workers are deployed, TinyFish usage expands alongside them.
"For basic data enrichment and sourcing, TinyFish has been way more accurate than our previous solutions, and it's been much faster to get results."
- Foudner, Ansur
What's Next
For Ansur, the next step is straightforward: keep deploying more AI workers.
As those deployments grow, so does the need for information beyond what already exists inside a manufacturer's systems. TinyFish scales alongside that growth, giving each worker the same web layer without requiring new infrastructure or per-customer integration work.
That leaves the team focused on what it actually wants to build: AI workers that understand manufacturing operations, while TinyFish handles the web underneath them.
Building something that needs the web as a reliable data layer?
