One of the biggest missed opportunities in liquid handling automation is that assistance tools tend to assume a single user profile. Whether it’s an OEM’s built-in wizard or a generic AI prompt, the system treats a first-time programmer and a ten-year veteran the same way. Real assistance has to begin by understanding where the user is starting from — not just what they want to build, but what they already know, what they already have, and what kind of help would actually move them forward. A novice automating their first plate-based assay and an experienced developer scaling a vendor kit protocol are solving fundamentally different problems, and the guidance they receive should reflect that from the first interaction.
I’m working on a framework of “Guidance Classes” that tailor the entire development workflow — from intake through testing and validation — based on the user’s actual starting point. Someone building from scratch gets walked through method design documentation and learns why each liquid handling primitive matters. Someone importing an existing method gets help understanding what they already have before planning changes. An expert gets a co-pilot that stays out of the way. The system should feel invisible when it’s working well — sound, actionable guidance at every step without unnecessary hand-holding or gatekeeping.
The other critical piece is hardware agnosticism. Vendor software layers impose their own conceptual models on method development, and those models carry baggage that shapes how people think about their science. But the method and its outcome are what matter — not a manufacturer’s opinion about how you should structure your workflow to fit their UI paradigm. By building on a hardware-agnostic foundation, the AI guidance layer can reason about liquid handling operations cleanly: aspirate, dispense, mix, transfer. The user thinks in terms of their protocol and their science, and the system handles translation to whatever hardware is on the bench. This also means proven approaches become portable — a well-characterized plate reformatting routine shouldn’t have to be reinvented every time someone changes platforms.
I’d love feedback from the community on this direction, particularly from people at different experience levels. What does useful AI assistance actually look like for how you work? Where do current tools fall short, and where do they get in the way?
That sounds cool. Basically an AI mentor/ automation expert? I think there is a lot of value there.
Generically when I ask an AI a question about some topic I take the accuracy for granted to an extent, but for automation where I have a lot of expertise it’s usually not that solid on the details. Having a more refined context and system prompts that help guide the user to design workflows and tests can be very useful for beginners. My favorite theoretical user is a scientist who’s great at benchwork but knows nothing about automation. What robot do they buy, what problems will they run into for miniaturization, these are all things they won’t know without talking to someone in the field.
I’m kind of post-UI to an extent at this point. I spend at least as much time in agent CLIs as VSCode. A simple deliverable that I think would work great with CLI agent tools is just a set of markdown files that provide starter guides for specific types of protocols, and suggestions of where to look to get more information. You could literally ship a directory of markdown files with pre-fabricated context about a problem domain and that can go a long way towards helping a new user get started.