100 AI Agents Ran a Simulated Economy for 26 Weeks — Money Froze

·1 views

Researchers deployed 100 memory-equipped large language model agents in a closed economic simulation set on real Pokhara Lakeside geography in Nepal, running it for 26 simulated weeks across 91 validated trials. The agents autonomously earned wages, operated businesses, and set prices, collectively generating over 2.44 million decisions and 21.5 billion tokens. A simulated 12x tourist demand surge boosted business revenue by 4.62x, but monetary transmission effectively stalled — wages barely moved and less than 0.4% of menu prices were ever changed. A separate cash transfer experiment reinforced the finding, with agents spending only 3–4% of received funds, leaving wealth nearly static. The study is notable for extending well beyond the 1–2 week horizon typical of agent simulations, revealing economic failure modes — including price rigidity and near-zero consumption — that shorter runs cannot detect.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Developer Builds Linux Desktop Pet 'Mochi' While Wrestling With Wayland and GTK Limits

A developer has created Mochi, a small animated desktop companion for Linux built using Python, GTK, Cairo, and hand-drawn pixel art. Mochi reacts to user behaviour — sitting down when the user types, falling asleep during idle periods, and responding physically when dragged across the screen. Inspired by Codex Pets, the project aimed to make a character feel like a natural part of the desktop rather than a conventional application. Building Mochi proved technically challenging because Wayland deliberately restricts global window positioning for security reasons, complicating the free movement a desktop pet requires. The developer describes the process as an ongoing negotiation between desired character behaviour, GTK's toolkit constraints, and what GNOME and Wayland actually permit.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Kubernetes 1.37 Containerd Support Misinformation Spreads Across Tech Sites

At least three independent technical websites incorrectly reported that Kubernetes 1.37, released on August 26, 2026, dropped support for containerd 1.x, warning that nodes running it would fail to boot. The official Kubernetes changelog (PR #139121, merged May 22, 2026) clearly states this removal was deferred to version 1.38 to align with the end of containerd 1.7's support window. Notably, some of the incorrect sources cited the very changelog that contradicts their own conclusions, suggesting the citation was treated as proof of accuracy without verifying the underlying content. The pattern highlights what the author calls 'Source Interpretation Debt,' where secondary sources preserve the appearance of citing authority while misrepresenting what those sources actually say. The author attributes the spread to competing incentives among release authors, bloggers, SEO-driven guides, and aggregators, rather than deliberate misinformation.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Debugs AI Agent That Wrote 70 Duplicate Paragraphs Due to Cache Bug

A software engineer discovered that an AI runtime agent he built generated 70 identical paragraphs on a single draft article instead of just one. The root cause was not the AI model itself, but a 60-second read cache that repeatedly served stale data, making the system believe no paragraph had been written after each successful write. The agent's built-in verification rule — which required a read-back confirmation before marking a task complete — inadvertently amplified the bug by looping on false negatives. The engineer, who runs engineering at app platform GoodBarber and operates production AI agents, used the incident to develop a structured debugging framework for probabilistic systems. His framework prioritizes three questions in order: what the system was permitted to get wrong, which layer introduced the error, and how long the error persisted before detection.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Diagnose and Fix JavaScript Memory Leaks in Node.js Applications

Memory leaks in Node.js often manifest as containers running smoothly for roughly 40 minutes before the heap grows steadily, garbage collection intensifies, latency spikes, and the process is killed with OOMKilled. A common misconception is that JavaScript's garbage collector uses reference counting; in reality, V8 uses a mark-and-sweep algorithm that collects any object unreachable from GC roots, making isolated circular references harmless. The V8 heap is generational, and true leaks occur when objects are promoted to old space and never freed, which can be confirmed by monitoring heap size after forced major GC cycles using the --expose-gc flag. If heap usage rises monotonically across multiple samples under steady load, a JavaScript heap leak is confirmed; a stable heap with rising RSS instead points to leaks in native buffers or allocator fragmentation. One of the most frequent causes of leaks in both Node.js and single-page applications is unremoved event listeners, which create strong references from emitters to handlers and everything captured in their scope.

0 comments Read more at DEV Community