GPT-5, Opus 4.1, and Shrinking CLAUDE.md

View in browser | Past Issue | Subscribe / Unsubscribe

SitePoint Source

Welcome, Developers! 👋

This week had a lot of releases and more follow as Claude improved their frontier model Opus and OpenAI released their next generation of models.

We also cover AI coding practices including shrinking Claude configuration files, building services entirely with AI agents, and debates over whether developers must adopt AI tools or risk obsolescence.

From our sponsor: Codemetrics

Understand your development team like never before

Codemetrics transforms your team's commits and PRs into actionable insights. Our AI-powered platform reveals who's excelling, who needs support, and how your performance compares to industry standards like DORA metrics.

With developer leaderboards, code reviews, and daily reports, engineering leaders get complete visibility while developers receive well-deserved recognition. Try our free beta with seamless GitHub integration today.

Start for free

🔖 The Reading Room

Articles we have hand-picked for you:

GPT-5 is here

GPT-5 is the generational successor to GPT-4 and comes with built-in thinking which was previously reserved for O-series models. With this release, the naming conception has simplified a lot with GPT-5, GPT-5-mini and GPT-5-nano. For ChatGPT users, the previous models have been retired from the UI and same will likely follow for API users too.

Some people are disappointed that GPT-5 didn't serve the hype but it does seem to top the intelligence charts, leaving Grok 4 behind. Coding benchmarks look good too so there's progress and improvement but if you were expecting a generational improvement like GPT 3.5 -> 4 was then you would be disappointed.

GPT-5 comes with a healthy 400k context size (more than Claude, less than Gemini) and costs the same as Gemini 2.5 Pro ($1.25 input / $10 output) with a knowledge cutoff of Oct 2024.

By OpenAI →

How I shrunk my CLAUDE.md file to 13 lines
Key Insight: use universal README files for context and environment tools for constraints. This approach works across all coding agents (Cursor, Claude Code, Codex, and Jules). Environment tools include type checkers, linters, and tests. Since modern AI agents can now iterate until all quality gates pass, manual rules are no longer reliable.

By Holden Matt →

The Best Companies Are Dictatorships

"Everyone wants democracy until they see what it builds"

The most successful tech companies operate as benevolent dictatorships, not democracies. Founders who obsess over every detail and maintain unwavering vision create more impact than collaborative decision-making committees. While uncomfortable to admit, direction beats democracy when building exceptional products.

By Nikunj Kothari →

Built with Borrowed Hands

David spent two months building a production service using only AI agents to write code. The results were mixed - while agents can generate functional code, they struggle with architecture, create unmaintainable duplicates, and often require more time than traditional coding. The current AI tools work best as coding assistants, not replacements.

By David Cramer →

Developers, Not Operators

Victor challenges the notion that all developers must embrace AI or become obsolete. He argues that the "adapt or perish" mentality ignores the fundamental right of developers to choose their own tools and approaches. Great software comes from thoughtful decision-making, whether that includes AI or not.

By Victor Wynne →

⏳ Back in Time

Most clicks from last newsletter:

🔗 The Link Lounge

Unordered finds from around the web:

Find something cool? You can send us links to feature here via email.

🧰 The Toolbox

Tools and products we're excited about today:

ForesightJS

ForesightJS is a lightweight JavaScript library that predicts user intent through mouse movements and keyboard navigation to prefetch content before it's needed, improving perceived performance across desktop and mobile devices with zero configuration required.

Learn more →

React Axe Reporter

React Axe Reporter is a React component that integrates with axe-core to display web accessibility testing results in an interactive, visual interface, helping developers identify and fix WCAG compliance issues with detailed reports and scoring.

Learn more →

StaticSearch

StaticSearch is a client-side search engine for static websites that uses JavaScript and JSON data files instead of server-side technologies, creating searchable indexes and providing web components for easy integration into any static site generator.

Learn more →

Opal

Opal is an experimental Google Labs tool that lets users build and share AI mini-apps by chaining prompts, models, and tools using natural language descriptions and visual editing, without requiring any coding knowledge.

Learn more →

🎤 Your Voice

Your feedback shapes what comes next! We read every email, so simply hit reply and tell us what's on your mind.