I Used Claude Science for a Formulation Study. Here's What Happened.

I did not want to write about Claude Science based on the announcement alone.

So I tried it. With a formulation problem I know well - the kind of study I have designed dozens of times across nearly a decade in pharmaceutical R&D.

Here is exactly what happened.

The setup

The research problem - A stability study for a small molecule formulation - excipient selection, concentration ranges, pH conditions, stability endpoints, analytical methods, timepoints. The kind of design that typically takes a formulation scientist two to three days of literature review, rationale building, and protocol structuring to get right.

I wanted to see what it would do with that.

What it produced

It moved fast. Within minutes, a design appeared.

On the surface it looked comprehensive. Conditions were listed. Excipients were suggested. Timepoints were included. It had the structure of a formulation study.

But when I looked closer, something was missing.

What was missing - Why this excipient over another? What literature supports this concentration range for this compound class? Why these timepoints specifically - what stability data or regulatory guidance drives that selection? What is the basis for the pH range?

I could not find clear answers. The design was confident. The reasoning behind it was not visible.

What I felt as a scientist

The design happened in front of me, not with me. I was not walked through the decision points. I was not asked about my lab's constraints, my available equipment, my regulatory context. I was not invited to challenge an assumption or adjust a variable and see how the logic shifted.

I was presented with a result.

In a computational workflow, that is often acceptable. You can validate the output against the data. But in formulation science, the reasoning behind the design is not separable from the design itself. A concentration range without a cited rationale is not a defensible concentration range. A timepoint without a regulatory or mechanistic justification is not a timepoint you can defend to your team or your submission.

The other thing I noticed: it was not iterative.

Formulation design is never linear. You propose a condition, then realize your stability chambers are booked for six weeks. You select a pH range, then find a paper that changes your thinking. You build a factorial design, then discover the analytical cost makes it unfeasible. The design evolves through that back and forth.

Claude Science produced a plan. It did not create space for that negotiation.

The best way I can describe the experience is this: I felt like an audience member.

What this tells us

Claude Science is not a flawed tool. It is a precisely built tool for a specific kind of scientist - one whose workflow is computational, data-heavy, and post-experiment.

For that scientist, what it does is genuinely powerful.

But for a formulation scientist - or anyone in wet lab research who needs a design they can build, justify, and defend - the pre-experiment layer is still missing.

The hours spent before a single vial goes into a stability chamber. The rationale that determines whether the study generates useful data or has to be repeated. That work is not what Claude Science was built to accelerate.