Steijn Pelle & Frédéric Renken on doing the job before you automate it
The founders of Lassie explain how they built AI agents that autonomously run dental back-offices — by first doing the paperwork by hand, engineering for full autonomy over years, and reaching the Main Street businesses whose only 'incumbent' was a human who quit.
Do the job before you automate it
Lassie's founders learned each back-office job by doing it by hand first, then automated their own work away.
Initially we were actually the humans in the loop. We kind of automated away our own problems.
Software finally does the labor, not just store it
For decades software only moved the filing cabinet into a database while people still did the work; now it can do the work itself.
So it just turns out that the work is orders of magnitude bigger than the storage of information that the work is done on.
It's a labor shortage, not a job takeover
In these small businesses the work goes undone because owners can't hire anyone — so agents fill a gap rather than displacing a worker.
It's not like, oh, AI is going to take the jobs. In many cases, you can't find somebody.
The real bottleneck was paper, not the model
Even capable models couldn't run these offices because most payments still arrive on paper — a federal mandate to go digital is what opened the door.
even if you have the models or had the models 5 years ago like yeah the payments are still on paper and we are digitizing that in the meantime
No human in the loop is the whole product
Because Lassie runs the practice instead of handing over a tool, it can't keep a human reviewer — so it had to be built to err on the side of correctness.
you all of a sudden need to build autonomous systems that run on its own and don't have a human in the loop. It runs the business for Dr. Sloop, which makes it technically super interesting.
The models still don't know how to do the work
Frontier models carry language and reasoning but not the office workflows, which live in staffers' heads rather than on the internet.
The models are trained on so much data and they're so large and yet they actually don't really know how to do any of this work.
The incumbent was Betty, and she quit
These categories never had a software incumbent to out-race — the only competitor was the human labor that walked out the door.
Well, the incumbent was named Betty and she quit two weeks ago. That's the incumbent.
Overhyped in the Valley, underhyped in Iowa
The hard part isn't skepticism — the customers are hundreds of thousands of busy, non-technical owners scattered across the country, not a handful of enterprises.
we literally need to go find like thousands, tens of thousands, hundreds of thousands of small businesses.