Supporting scientific discovery: two weeks using EMET
Dr. Melissa Chengalroyen is a senior researcher at the University of Cape Town, working on drug discovery for tuberculosis. Like many researchers, her time is divided between the laboratory, student supervision, committee work, administration and writing, which often happens in short bursts between other responsibilities. She had two weeks with EMET, BenchSci's agentic workbench for preclinical R&D, and came to it with live work rather than example prompts: a review article she had started and stalled on, a grant application already drafted and an open question about a gene of interest.
Breaking through a writing block
A review article had reached a standstill. "I hit a mind block and just couldn't find the time to push through it." EMET produced a detailed comparison of the benefits and challenges within the scope of the topic. This enabled Dr. Chengalroyen to rapidly gain an overview of this component before exploring the most relevant papers in greater depth. For Dr. Chengalroyen, the time savings were substantial.
"The only time I get to read and write is maybe half an hour, during a busy day. So for me, it probably saved months."
Experimental planning, and a sounding board that is always available She worked on a research idea with EMET and got back the strengths and the drawbacks. What she had not expected was that it went further.
"What I really liked is that it also gave me alternatives as well."
The value here is availability. Within her community, scientists are often travelling or extremely busy with ongoing work - it’s not feasible to have regular scientific discovery discussions. Having EMET is like having a regular sounding board.
"Something popped into my head and I want to know right now, is this feasible? I've never been able to do that before."
A gene essentiality database that is normally too big to enter
Novel target identification is the standing job in her group, and it means interrogating a lot of genes. The question she brought to EMET was a conditional one: under which in vivo conditions is her gene of interest induced. The answer lives in a mycobacterial gene database that she can access perfectly well on her own. Access was never the constraint. Scale was.
"It's got thousands of genes and different induction conditions, with different experimental parameters. Even when you do go into it on your own, it's just an overwhelming amount of information. And this was able to give me exactly what I needed."
EMET went into the database, worked through the induction conditions, and returned the subset that answered her question. The information was always there, but extracting it manually would have taken hours of effort that competing priorities rarely allowed.
Strengthening a grant that was already written
The grant application was finished before EMET came anywhere near it. This is a useful distinction, because the common assumption about AI and grant writing is that the tool produces the prose. That is not what happened here. She had the argument, the science, and the draft. What she wanted was a second reader.
"I like the fact that I could actually ask EMET if there were ways to strengthen it."
It came back with suggestions for strengthening the application. Anyone who has waited on a colleague for a read of a grant will recognise what she got: a critical pass on her own writing, available in the window she actually had free, from something that had read the field.
Everything was already within her ability
What is striking about Dr. Chengalroyen’s two weeks is what she did not need. No new expertise, no new data, no new collaborators. Every one of these four pieces of work was already within her ability. The review, the experimental design, the database query, the grant critique. She had the training for all of it and the judgement to check all of it. That is the constraint an agentic workbench actually lifts. Not the science, which stays hers, but the distance between having a question and being able to afford to ask it
