What Automation Looks Like, and Why It Matters More Than the Model
Meghav Verma
Talk to most labs about AI and the conversation goes straight to the model. What architecture, what's it trained on, how accurate is it. That's the layer everyone can picture, so it's the layer everyone talks about.
Almost nobody talks about what's underneath it. And that's a problem, because the model is not where the real advantage lives.
Automation Isn't One Step. It's the Whole Chain.
When people hear "automated sample prep," they often picture a single machine doing a single job faster: an automated pipette, a faster centrifuge. That's not really what end-to-end automation means, and it's not where the value is.
Real end-to-end automation means the entire sequence, extraction, evaporation, reaction execution, and everything between, runs as one coordinated process instead of a chain of separate manual handoffs. No sample sits waiting for a tech to notice it's ready for the next step. No step depends on which technician happens to be at the bench, how many samples came before it, or what time of day it is.
That distinction matters enormously. A lab can automate individual steps and still have inconsistency, because the handoffs between those steps are still manual, still variable, still dependent on human timing and judgment. End-to-end means the whole run behaves the same way every time, start to finish, regardless of who's on shift or how the day has gone so far.
Why This Matters More Than the AI Layer
Here's the part that gets underestimated: the model sitting on top of all this data is, in a real sense, replaceable. Model architectures improve every year. What one lab builds today, a competitor can approximate with an off-the-shelf tool next year. The model is not the moat.
The data feeding the model is the moat, and the data is only as good as the process that generated it. A brilliant model trained on inconsistent, drift-riddled sample data will always underperform a simple model trained on clean, consistently generated data. No amount of clever architecture fixes a foundation problem. It just gets better at confidently learning the wrong thing.
That's the real argument for automation at the sample prep level: it's not really about speed, or even about reproducibility on its own. It's about building the one asset that compounds over time, a body of data generated the same way, run after run, that gets more valuable and more trustworthy the longer it accumulates. That's the layer nobody can shortcut their way to, and it's the layer that ends up mattering most.
Naming the Right Layer
It's easy to get excited about the model because it's the most visible part of the stack. But visibility and importance aren't the same thing. The unglamorous, upstream work of making sample prep consistent, end-to-end, is what determines whether everything built on top of it is trustworthy at all.
At Axle Research and Technologies, this is the layer we think about first, not because the model doesn't matter, but because it's the easiest part to get right once the foundation underneath it is solid. Getting that foundation right is the harder, less visible problem, and it's the one worth solving before anything else.
We'll be talking through what that foundation looks like at Future Labs Live 2026. If it's the layer you're rethinking too, let's compare notes there.
