AI News
AI news that matters. Updated daily.
No stories match your filters.
Developers Find HTML Files Are Better Claude Code Context Docs Than Markdown
Most Claude Code users default to markdown files and terminal prompts for giving the agent context. A technique gaining traction among practitioners flips that assumption: HTML files, not markdown, work better as persistent scratchpads for complex, multi-step coding sessions.
Claude's Strength Is Working With Messy, Unfinished Thoughts
Most AI writing tools are built for the polished-input moment. Feed them a decent draft and they'll improve it. Feed them chaos and they'll either refuse or rewrite everything so thoroughly that your original point disappears.
Enterprise AI Agent Swarms From Lemonade, CrowdStrike, and Siemens Released as Browser Templates
What do an insurance company, a cybersecurity firm, and a manufacturing giant have in common? All three have built multi-agent AI systems - networks of AI models that coordinate to complete complex tasks by handing off work to each other - and someone has reverse-engineered how they work.
Claude's New Auto-Completion Keeps Finishing Your Sentences Too Early
Claude's auto-completion feature, rolled out recently to the web interface, is drawing complaints for firing too aggressively. The feature - which suggests text completions as you type in the chat input - is jumping in before users have finished forming a thought, leading to accidental submissions and broken sentences.
DeepSeek Walks Away From Alibaba Investment to Stay Independent
DeepSeek said no to Alibaba.
Claude Desktop App Adds Context Usage Indicator on MacOS
Anthropic quietly pushed a small but genuinely useful update to Claude for Desktop on MacOS: the app now displays how much of your context window you've consumed in a given conversation.
Anthropic Retiring Claude Sonnet 4.5 - Developers Need to Migrate
Claude Sonnet 4.5 is being retired. Anthropic is phasing out the mid-tier model from its Claude 4 lineup, meaning developers and applications pointing to that specific model version will need to migrate before the endpoint closes.
Claude Code's Surprising Strength: HTML That Actually Works
What happens when an AI coding assistant turns out to be genuinely better at one thing than everything else? Developer and researcher Simon Willison explored exactly that question in a blog post about the unreasonable effectiveness of HTML when using Claude Code.
ChatGPT vs Claude: How Practitioners Actually Split Their Work
Ask anyone who uses both ChatGPT and Claude daily and you get a similar answer: they stopped trying to pick a winner and started treating them as different tools for different jobs.