The problem. DBTL is easy to draw and hard to run as one integrated system. Carbonell et al. built the whole loop — design through machine-learning-guided redesign — inside a biofoundry, and reading it turns the cartoon into an engineering account.

The idea. One concrete pipeline for producing fine chemicals in microbes. Design: computational pathway enumeration (retrosynthesis) and enzyme selection to get from a host’s metabolism to a target molecule. Build: automated DNA assembly of the candidate constructs. Test: analytics measuring product titers across the library of variants. Learn: statistical/ML models over that test data to nominate the next, better round of designs. The point is that it’s automated and closed — the output of Test feeds a model that drives the next Design — realized on the SYNBIOCHEM foundry infrastructure.

Why it matters. Trace one iteration and the disciplinary boundary jumps out: Design and Learn are data engineering and modeling wearing a lab coat, while Build and Test are the physical, capital-heavy steps. That’s the same conclusion the DBTL review argues in the abstract, but here it’s demonstrated on a real product, which makes it far more convincing about where computational leverage actually sits.

Verdict. A valuable existence proof that the full loop can be automated, tempered by the honest reality that it’s still resource-intensive and the paper shows the machinery more than a long series of profitable iterations. Read it right after the DBTL survey: the survey is the map, this is one route driven end to end.