EXPLORATORY SINGLE-CASE STUDY · DIGITAL LAB · KLAUS ARENT

A long-term observation of intensive use of generative AI in knowledge work

Since mid-2025 I have been documenting how working methods, productivity, decision load and project scope change when an experienced knowledge worker uses AI agents extremely intensively. Investigator and subject are the same person: me.

n = 1, self-reported, no control group. Study design and limitations: study design (German).

THE KNOWLEDGE GRAPH

The study's memory, relinked every night.

Every night at 3 a.m. an agent breaks the day's work into notes, links them and updates this graph. Only the anonymized topology is shown: dots and connections, no titles, no content. Curated main nodes are clickable.

DRAG = ROTATE · SCROLL = ZOOM

8,966 NOTES · 17,347 LINKS · AS OF 2026-11-05 · UPDATED NIGHTLY AT 3 A.M.

Guided tour · same data, fixed camera path along the six largest topic nodes

01 COLLECTED DATA

33
DOCUMENTED CASES
6,277
OWN GIT COMMITS
2.000+
DEPLOYMENT EVENTS 2026
~20
RUNNING SYSTEMS

These numbers measure activity, not success. Commits are filtered by author identity and deduplicated by hash; they include experimental and discarded changes. Deployment events include fixes and automated releases.

As of July 2026 · from my own git analysis across 101 repositories · total recorded across all repositories, including cloned third-party projects: 14,263

04 WORKING PAPER

One Operator, Twenty Systems

Working Paper v0.2, 17 September 2026, 19 pages. Not yet for citation. Figures in the paper come from a snapshot of 16 September 2026 and differ from the figures on this page (as of July 2026).

One Operator, Twenty Systems (PDF, 0.4 MB)

Investigator, subject and operator of the documented systems are the same person. Prior experience and resulting biases are disclosed. Author and conflicts of interest →

The full study — findings, discontinued cases, research journal, study design — lives on the German site. Full documentation (German) →

The observation is running.