{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,8,25]],"date-time":"2026-08-25T09:23:24Z","timestamp":1787649804476,"version":"build-2736575974"},"posted":{"date-parts":[[2026]]},"group-title":"SSRN","reference-count":0,"publisher":"Elsevier BV","license":[{"start":{"date-parts":[[2026,1,1]],"date-time":"2026-01-01T00:00:00Z","timestamp":1767225600000},"content-version":"unspecified","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"abstract":"<jats:p>&lt;p&gt;Agentic artificial intelligence can produce substantial gains in speed, output, and problem-solving capacity. Yet the interaction process also creates a distinctive pattern of intermittent outcomes: complete successes, partial progress, plausible near misses, regressions, and failures arrive in an uncertain sequence. This paper introduces the Claw Machine Effect as a theoretical mechanism through which those outcomes may reinforce continued human engagement independently of verified value. The metaphor is intentionally narrower than a claim of addiction. A claw machine combines uncertainty with perceived skill, visible proximity to success, repairable partial progress, and the recurring intuition that one more attempt may complete the task. These features map closely onto human supervision of contemporary AI agents.&lt;span&gt;&lt;\/span&gt;&lt;\/p&gt;\n&lt;p&gt;The paper integrates research on partial reinforcement, reward prediction error, near misses, social-media reward learning, automation bias, human-AI productivity, software-engineering work, and work intensification. It then develops a formal model distinguishing objective progress, perceived progress, verification debt, and the decision to continue. Seven propositions specify when salient near misses, variable outcomes, rising verification complexity, and concurrent agents should extend sessions or widen the gap between perceived and verified value. An illustrative Monte Carlo analysis runs 100,000 simulated sessions in each of four scenarios. Relative to a calibrated variable-outcome baseline, adding a salient near-miss premium raises simulated completion from 13.0% to 19.0% and sessions exceeding 20 attempts from 3.4% to 6.4%, while mean objective value falls from 0.120 to 0.117 model units and a large verification gap emerges. Periodic verified checkpoints reduce sessions exceeding 20 attempts to 2.4% and restore objective value to 0.120, though simulated completion returns to 13.4%. These values are demonstrations of the model's implications, not estimates of human behavior.&lt;\/p&gt;\n&lt;p&gt;The analysis predicts a productivity-burden paradox: AI can genuinely accelerate work while also increasing verification, concurrency, after-hours extension, and organizational expectations. Because many business models benefit from repeated attempts, token consumption, retention, or apparent throughput, the resulting incentive wedge does not require intentional addictive design. The paper concludes with testable hypotheses, experimental and field-study designs, and implications for product metrics, organizational governance, and terminology. In particular, it recommends agentic over-engagement and verification fatigue for workplace phenomena, reserving clinical terms such as psychosis and addiction for evidence that meets diagnostic and causal standards.&lt;\/p&gt;<\/jats:p>","DOI":"10.2139\/ssrn.7328978","type":"posted-content","created":{"date-parts":[[2026,8,25]],"date-time":"2026-08-25T09:04:19Z","timestamp":1787648659000},"source":"Crossref","is-referenced-by-count":0,"title":["The Claw Machine Effect: Variable Reinforcement, Near Misses, and Work Intensification in Agentic AI"],"prefix":"10.2139","author":[{"ORCID":"https:\/\/orcid.org\/0009-0001-1325-5614","authenticated-orcid":true,"given":"Derek","family":"Snider","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"78","container-title":[],"original-title":[],"deposited":{"date-parts":[[2026,8,25]],"date-time":"2026-08-25T09:04:19Z","timestamp":1787648659000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.ssrn.com\/abstract=7328978"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026]]},"references-count":0,"URL":"https:\/\/doi.org\/10.2139\/ssrn.7328978","relation":{},"subject":[],"published":{"date-parts":[[2026]]},"subtype":"preprint"}}