MGI Tech's Zhiku Yunshu upgrade adds Genpilot agent: a live single-cell figure reproduction drops from months to hours
Synopsis
At its 2026 academic annual meeting, MGI Tech announced an upgrade to the Zhiku Yunshu biosample value-transformation platform, centered on the DCS Cloud biomedical analysis platform and the research agent Genpilot. Genpilot takes natural-language requests, decomposes tasks, generates code, runs analyses and produces figures and drafts; in a live demonstration it reproduced Figures 1–4 of a single-cell dataset end to end with results consistent with the original paper's core conclusions, compressing output from months to hours. On the storage side, AI vision-assisted calibration and natural-language retrieval cut sample-location time from minutes to seconds.
Interpretation
The software entry point to a biobank shifts from manual search and hand-run analysis to asking a question in natural language that locates samples and starts analysis. Previously the biobank focused on storing samples well, sample retrieval relied on manual lookup, and cross-institution integration and value conversion were limited. In the upgrade MGI Tech announced at the annual meeting, the ZSM Pro biosample management system added natural-language retrieval and a knowledge assistant, cutting location time from minutes to seconds.
Genpilot turns a researcher's stated goal into an executable analysis pipeline and delivers figures and drafts directly. Previously researchers wrote code, ran pipelines, then drew figures and wrote up results themselves. In a live demonstration, a single request to reproduce all main figures from a given dataset using a paper's method led the system to decompose the task, generate code, run analyses and retry fixes, delivering draft-quality figures and text.
AI in the storage step lets equipment see where samples are and check inbound records. Storage and operations previously depended on manual location and checking. The CLab-LT series introduced AI vision-assisted calibration for locating cryo-racks, boxes and tubes, plus inbound record checks, low-temperature position-offset correction and tube-picking fault detection.
Perspective
The demonstrated scope is the platform upgrade MGI Tech released at the annual meeting and its live demonstration: an end-to-end reproduction of Figures 1–4 for one single-cell dataset, plus the description that ZSM Pro natural-language retrieval cuts location time from minutes to seconds. The intended readers are biobank operators and bioinformatics analysts, in settings of an integrated hardware-software smart biobank and a cloud analysis platform.
The statement that reproduction results match the original paper's core conclusions comes from a live demonstration, without the dataset's size or a quantitative comparison of the reproduction. The claim that fewer than one in ten biobank samples are used and nearly nine in ten sit dormant is cited from international research, not measured by this upgrade itself.