PlayAgit / Notes for builders
Build. Play. Explore.
A technology blog covering game development, AI, automation, and industry news — notes from what we actually build and run.
-
The Signed Receipt for a Write That Never Happened: How Agent Execution Records Became Evidence
The premise behind this whole category is that an AI execution can be watched but not re-run.
Latest from the workbench
-
One Repository, Two Complete Frameworks: What XuToWei’s GameDevelopmentKit Bundles for Unity Client and Server
GameDevelopmentKit is a GitHub repository published by the user XuToWei, at the repository path XuToWei/GameDevelopmentKit.
-
Three Books With Almost the Same Name: Which SFML Repository Belongs to Which, and Who Owns It
Search for the source code that goes with an SFML book and at least three titles come back tangled together:
-
The Sandbox Is Part of the Evaluation: Google’s ADK Live Evaluation, Model-Judged Scoring, and the Answer an Agent Can Find Without Knowing It
Google’s developer blog has published guidance on evaluating live voice agents in the Agent Development Kit (ADK), describing native live evaluation aimed specifically at live and voice agents (source).
-
The Decision Tree Is Not the Boundary: Solo-Built AI Operators, Their Revenue, and a 1988 Permissions Bug Underneath Them
A cluster of posts over the past weeks has converged on the same object from opposite ends.
-
Hugging Face Kernels Ships 200+ WebGPU Kernels — and Helion Support That Was Never About the Browser
Hugging Face ships 200+ WebGPU kernels for local AI, then adds Helion support — a sign the browser is one kernel target among several, not the goal.
-
NVIDIA CUDA Use Cases: Scientific Simulations and Large-Scale AI Training
NVIDIA’s developer blog article identifies scientific simulations and large-scale AI training as use cases for CUDA.
-
Xbox App Is Coming to TCL Smart TVs Through a New Microsoft–TCL Partnership
Xbox has announced a new partnership with TCL that will bring the Xbox app to select TCL Smart TVs, with availability expected in the coming months.
-
Cloudflare Now Ships Two Kinds of Automated Remediation — and Only One of Them Is Allowed to Touch Your Code
Cloudflare has published two separate automation stories, and the interesting thing is not that they arrived close together but that they draw the line on autonomy in opposite places.
-
Evaluations That Rank Without Distinguishing: BenchMIRT, a 50-Question Evaluation, and the Case for Behavioral Evaluations
The question BenchMIRT puts on the table is not which model wins but what a benchmark score is a measurement of.