Tag
A blog post arguing that strict small-PR 'annealing' policies hinder rapid change, and that AI-driven coding produces big, interconnected changes which sometimes can't be cleanly decomposed into small reviewable steps.
Sonar CEO Tariq Shakir’s talk points out that AI agents have caused code output to surge (+2,450% in lines of code, +240% in commits), yet releases have grown only 30% — the “scaling from production to delivery” problem remains unsolved. He proposes a “code confidence formula” built by multiplying intent, guardrails, verification, and evidence, helping software factories deliver confidence rather than blind trust.
Microsoft distinguished engineer Stephen Toub recounts how he single-handedly ported the 400,000-line GitHub Copilot runtime from TypeScript to Rust in about 14.5 weeks using AI coding agents, at a cost of roughly 136 billion tokens and $120,000, discussing language choice, rollout strategy, and regression prevention.
A quote tweet pushing back on trendy 'performance slop' in timelines, arguing that real performance engineering means profiling CPU, heap, disk, and network and shrinking the biggest bars on those profiles.
MIT Technology Review argues teams should measure AI development by cost per merged pull request instead of per-token spend, recommending model routing and owned open-weight capacity to curb the rising economics of agentic coding.
An empirical study finds that continual harness self-evolution of coding agents across multiple software engineering tasks causes severe catastrophic forgetting, with SWE-Bench performance dropping from 94.0% to 32.0% after evolving on FeatureBench, and offers insights for future continual agent-harness evolution.
Nader Dabit notes that at Cognition, as the cost and speed of merged pull requests drop due to AI, people are building and fixing software that wouldn't have been attempted a year ago — illustrating Jevons Paradox in practice.
Troy Hunt's Weekly Update 528 discusses the Proton CEO's controversial AI engineering manifesto about developers' roles shifting to guiding LLMs, alongside reflections on reading AI-generated code and the economics of solar.
CrossSWE introduces a multi-repository benchmark of 171 real-world instances showing that current LLM coding agents struggle with cross-repository consistency, with the strongest model (GPT-6.1 Sol) reaching only 58.48% Pass@1 and cross-repository inconsistencies accounting for 44.2% of failures.
Kent Beck argues that trust, not code generation speed, is the binding constraint on software delivery in the AI era, and that XP practices like testing and pairing manufacture the confidence needed to release; he also promotes his upcoming paid class on trust-optimized AI-assisted development.
A practical, evidence-based guide to GPT-6 Astra's real-world use cases, mapping its performance across 12 Scale AI SEAL benchmark categories to project recommendations, strengths, limitations, and cases where alternative models like Claude Fable 5.1 are preferable.
介绍了「Kluever Maneuver」这一 bug 修复实践：先提交一个能通过的失败行为测试来复现 bug，再在第二个提交中修复代码并同时修正该测试，从而证明修复真正有效且每个提交都能通过 CI。
SVPino argues that deterministic tests should replace written instructions when working with AI agents: anything checkable automatically—formatting, types, API contracts, security—should be enforced by tests, linters, and static analysis rather than by prompting the agent with rules.
A Code Simplicity blog post arguing that any time a developer stops to think while coding, it signals an underlying problem—usually incomplete understanding—and explains how deeper comprehension of tools, symbols, and the problem itself is the real secret to fast, simple programming.
The author explores what to do when AI coding agents produce code faster than a team can comprehend it, and how five layers of governance — compiler-enforced constraints, architectural skeletons, read-only contracts, refactoring loops, and disposable implementations — can counteract system entropy and human cognitive bankruptcy to preserve software maintainability.
Jack Ganssle's classic 2002 article lists the top ten reasons embedded systems projects get into trouble, covering issues like insufficient resources, premature coding, and the pitfalls of using C in safety-critical firmware.
Amazon发布新工程指南，告诫工程师停止为了显得高效而盲目堆砌AI token消耗，强调AI将处于软件工程实践的核心，但使用量本身并非目标。
Cisco president Jeetu Patel gave a 27-year-old engineer unlimited AI tokens and a mandate to refactor 98% of the company's legacy code within 24 months, tasking him with leading a broad AI-driven codebase rewrite.
Meta 分享其 Catching JiTTests 方案在生产环境大规模部署的成果：利用 LLM 自动生成会失败的即时测试，在代码合入前捕获传统测试和 LLM 代码审查难以发现的 bug，并配套发布了详细论文。
Podcast interview with Lean creator Leo de Moura on how AI is making formal verification practical, the need to verify AI-generated code and autonomous agents, and the vision of a verified software stack.