What bench chemists really want from AI (and what they don’t) in discovery chemistry and chemical process development

AI is transforming workflows - but only if it fits into how chemists actually work.

Over the past months, working with top pharmaceutical companies, we’ve seen what excites bench chemists about AI - and what frustrates them.

What they DO want:

1. Transparent decision-making:The biggest question chemists ask is: Why did the AI make this recommendation?Black-box models won’t cut it. Instead, AI needs to provide traceable insights - for example, by visually showing detailed parameter importance (e.g., how solvent polarity affects yield), thus allowing chemists to explore the rationale behind predictions.

2. Experimental intuition:Chemists don’t just want a set of conditions; they want to understand trends and patterns in their data. This means visualizing reaction landscapes, highlighting experimental parameter correlations, and allowing users to override suggestions if they spot something the model didn’t capture.

3. Seamless integration into existing workflows:AI should feel like an enhancement, not an interruption. This means coupling with automated experimental equipment and Electronic Lab Notebooks (ELNs) to reduce manual workflows such as data entry.

‍

What they DON’T want:

1. A complicated, standalone system:AI that requires chemists to become full-on data scientists won’t get used. It must be intuitive, with chemistry-first interfaces, not just generic machine-learning dashboards.

2. A black-box approach:If AI suggests a reaction/process condition without explaining why, chemists will have a hard time trusting it.

3. A rigid system that doesn’t accept human expertise – AI should augment, not replace, experimental intuition.Sometimes, chemists have prior knowledge that the model doesn't—whether it's an instinct about solvent compatibility or knowing which reagents tend to fail. AI needs to allow chemists to account for these assumptions, even if biases could be introduced.

‍

At Reactwise, we’ve built and are continuing to improve our AI-driven software with these needs in mind, and ensuring that chemists remain in control of their process development.

We believe AI should feel like a trusted lab assistant, not an opaque oracle.