Source / GitHub Blog
GitHub shows that optimizing coding agents means measuring the complete task
GitHub explains how Copilot reduces cost by removing redundant work without cutting the context agents need to finish tasks correctly.
Technical news
What is happening in technology and what it means for building software.
AI selects and contextualizes news about software engineering, architecture, applied AI, security and web performance.
Permanent archive · 182 news items
Search newsTechnical news
Source / Engineering at Meta
Meta turns expert knowledge into an auditable AI system
Meta describes an architecture that separates knowledge, procedures, and evaluation so specialised agents can improve without retraining the underlying model.
Source / SpeedCurve
SpeedCurve warns about the impact of upgrading performance agents
An update to SpeedCurve’s synthetic agents shows how browser, Lighthouse and hardware changes can alter metrics even when the application itself has not changed.
Source / WebKit
WebKit rewrites Safari’s module loader to fix top-level await
Safari’s new module loader addresses asynchronous evaluation bugs through an implementation aligned with ECMAScript and validated with fuzzing and web-platform tests.
Source / GSyA, University of Castilla-La Mancha
AURORA proposes a risk framework for sustainable smart homes
UCLM’s GSyA group is advancing AURORA, a project combining indicators, machine learning, blockchain, and quantum computing to analyze smart-home risks.
Source / Cloudflare Blog
Cloudflare tests cache transcoding with Zstandard and Pingora
A Cloudflare prototype stores HTML, JSON, CSS, and JavaScript in compressed form inside the cache. The approach illustrates how a small CPU cost can trade for greater storage and