#Ai-Coding
17 notas · última:
Why We've Tried to Replace Developers Every Decade
The article traces fifty years of attempts to make software creation simple enough to need fewer developers, and argues that the real limit is the complexity of the work, not the tool.
The Future Of Software Engineering with Anthropic
Roundtable on how AI is changing software engineering: closed-loop development, test-first workflows, evolving code review, long-horizon agents, and the limits of adoption in regulated settings.
How I use Claude Code
Boris Cherny describes a largely vanilla Claude Code setup built around parallel sessions, shared instructions, slash commands, subagents, hooks, and strong verification loops.
AI Coding Degrades: Silent Failures Emerge
The author says AI coding assistants have plateaued in 2025 and now fail more quietly, often producing code that runs but gives wrong results. He links this to training on user-accepted outputs.
Changed my mind: AI is actually good for coding
The author says AI coding tools help with repetitive work, are most useful in the hands of experienced developers, and are already creating real business and training concerns.
What the Team Behind Cursor Knows About the Future of Code
Cursor’s team argues that code work is shifting from typing in an IDE to directing and reviewing AI agents, with planning, model choice, and long-running workflows becoming central.
The 66% Problem
The text argues that AI assistants write code that is “almost right,” but hard to debug, creating new bugs and a deceptive sense of productivity.
Eight Software Markets That AI Will Transform Differently
The text argues that AI coding will reshape software unevenly because different markets have different constraints. Where the limit is developer time or skill, AI can unlock more production and create Jevons-style growth. Where the limit is procurement, regulation, safety, or market demand, cheaper code changes less.
Cursor debuts Composer and multi-agent suite in Cursor 2.0
Cursor 2.0 ships Composer, a coding model tuned for sub-30-second responses, alongside an agent-centric interface that runs multiple agents in parallel via git worktrees or remote machines.
Every Layer of Review Makes You 10x Slower
Avery Pennarun argues AI coding speed doesn't fix engineering velocity because review layers, not typing speed, are the bottleneck, and draws on Deming's quality theory to argue for eliminating review stages, not stacking more of them.
The Next Two Years of Software Engineering
Addy Osmani lays out five open questions on how AI reshapes software engineering through 2026: junior hiring, core skills, the developer role, specialization, and education.
AI Writes Code Faster. Your Job Is Still to Prove It Works.
Osmani's framework for code review after AI: solo devs lean on tests as a safety net, teams need a PR Contract stating intent, proof, risk, and where to focus review.
At Amazon, Some Coders Say Their Jobs Have Begun to Resemble Warehouse Work
NYT reports Amazon engineers describe AI-driven speed-ups echoing warehouse robotics: faster output quotas, less time to think, and a shift from writing code to reviewing it.
AI-assisted development needs automated tests
Simon Willison argues that comprehensive test coverage is what makes LLM-assisted coding trustworthy: tests let you prove convoluted AI-written code actually works.
AI Tooling for Software Engineers in 2026
Pragmatic Engineer's 2026 survey of 900+ engineers: Claude Code overtook GitHub Copilot and Cursor in eight months, and the more senior the respondent, the more they love it.
The New Software Lifecycle
Addy Osmani pulls the load-bearing ideas from a Google whitepaper he co-wrote on AI and the SDLC: an agent is 10% model and 90% harness, and verification is what separates vibe coding from engineering.
My LLM coding workflow going into 2026
Addy Osmani lays out the workflow he converged on after a year of AI-assisted coding: spec before code, small chunks, heavy context packing, and treating every AI commit as his own responsibility.