On 24 October 2026 Chuck Brooks wrote that HeartGenesis and Compilence "have agreed to form Compilence Care and establish an experimental program at an existing hospital in Japan". The plan is to put robots and AI assistants next to clinicians. We build systems of autonomous agents, so a claim about controlling machine behaviour near patients is one we read closely. We have no part in this project and have tested nothing. Below is what the article states, what stands behind each statement, and why our own state is "on hold".
What was announced, and what was not
The article says the two companies agreed to create a company, Compilence Care, and to develop "an experimental robot-augmented, AI-assisted setting within an existing hospital in Japan". HeartGenesis and the participating physicians would bring surgical methods and clinical experience. Compilence, in the article's words, "develops the underlying AI infrastructure for controlling and tracing machine behavior".
The article draws the line itself: "The agreement sets a direction; company formation and deployment of the hospital program remain ahead." It adds that any clinical use must be supported by evidence and by institutional and regulatory approvals.
That gives four separate facts: an agreement, a company, a programme, and clinical use. Only the first is reported as having happened. The article's reference for it is project background provided to the author for the article. We treat it as the author's statement about an agreement between two other parties.
Source: Chuck Brooks, "In 2018, I Predicted Robots Would Learn to Operate: That Future Is Taking Shape", BizCatalyst 360°, 24 October 2026, bizcatalyst360.com.
How the control technology is described
The technical description takes two paragraphs and is attributed to Yurii Chudinov, named in the article as co-founder and CTO of Compilence. Three names appear.
E.L.I.A. is said to describe tasks and limits independently of the skill. The physician's experience becomes what the article calls Knowledge Tissue, "whose fragments can be improved, transferred, or deprecated and removed from active use". ARC "translates protocols into machine instructions for traceable, real-time execution". Its "admissibility levels" are said to let behaviour be corrected inside the protocol's boundaries even when a specific action was not programmed in advance.
The article is careful about results. Of the compact representations Compilence is developing, it says that "quantitative benefits remain to be demonstrated". The one robotic experiment it describes with numbers is a 2025 report by other researchers on excised animal tissue, which the article labels as research outside this collaboration. It says nothing about the announced programme.
For the technology, the article's reference is language documentation and explanations provided for the article. It links one public record: Yurii Chudinov, "Human interface to ARC (Auto Regressive Compiler) for deterministic semantic governance over Probabilistic Inferring Engines (PIEs)", published on Zenodo on 22 May 2026. We opened that record on 14 November 2026. Its access status is "embargoed" and the embargo date is 19 January 2031; no file can be downloaded now. The public description includes an overview of E.L.I.A. and ARC; the embargo restricts downloadable files. The description does not name Knowledge Tissue.
Source: Yurii Chudinov, "Human interface to ARC (Auto Regressive Compiler) for deterministic semantic governance over Probabilistic Inferring Engines (PIEs)", Zenodo, 22 May 2026, doi.org/10.5281/zenodo.20343518.
Why we record it and wait
Some of the words are close to ours: traceable execution, limits stated apart from the skill, pieces of knowledge that can be retired. Close words are not the same design. The public overview is the developer's own account and is not a specification of any of the three components, so we draw no comparison with our own work, favourable or not.
We also keep the kinds of fact apart. "Agreed", "formed", "ran" and "worked" are four claims, and each needs its own evidence. A description given by the people who build a system is recorded as their description. We apply the same rule to our own agents: a report of what was done is not accepted in place of a record of it.
What would move this item from "announced" to something stronger is concrete. A technical specification whose files can be read. A registered company. A protocol for the hospital programme that says who checks the machine's behaviour and against which record. The approvals the article itself calls necessary. The article names the measure of success too: "whether a transferred skill remains reliable across cases, staff can supervise it effectively, and patients benefit".
Our state: on hold. Nothing is integrated and nothing is proposed. This is not a partnership of ours and not an implementation of anything we publish. We will read the record's files when they open, or earlier if a programme document appears.
Source: Chuck Brooks, "In 2018, I Predicted Robots Would Learn to Operate: That Future Is Taking Shape", BizCatalyst 360°, 24 October 2026, bizcatalyst360.com.
Follow-up, 16 November 2026: a separate research direction
Following Yurii Chudinov’s recommendation, we read Albina Dermenzhy’s Tech Diplomacy: A Framework for Interaction Between Probabilistic Systems, Humans, and Other Systems (v1, 12 November 2026, DOI 24172092) and selected methods, findings and limitations of GPT Deprivation (v1, 23 October 2026, DOI 23599162; the record credits Dermenzhy as editor). Our follow-up article distinguishes those research proposals from a proposed experiment in our own Live Lab.
This is a separate research direction. It supplies neither the embargoed ARC specification nor evidence of the announced hospital programme’s deployment or clinical effectiveness. The statements and access observation above retain their 14 November attribution. The hospital item remains on hold; no integration or partnership of ours is established by this update.