I had four Chrome windows open last week. One for the extension I'm building, one for a client thing, one full of research tabs I hadn't touched in days, one I couldn't explain. Bookmarks were no help. I'd saved stuff with good intentions and stopped trusting that folder months ago. Tried using a new tab extension called Tabisto. It turns your new tab into a visual dashboard where you build named sections and drop links in. I set up one section per active project. Work and personal stay in separate workspaces so they don't mix. There's a command palette (Cmd+K) to jump to any bookmark without clicking around. Quick notes on the tab itself so I stop opening a random doc every time I need to write something down. Saved sessions that snapshot whatever tabs are open and bring them back as a group later. Free, no account, works offline. Honestly, I expected it to feel like every other new tab extension I've tried and then ignored. It didn't. Having the browser open to something actually organized every single time I hit a new tab changes how a work session starts more than it probably should. Anyone here have a proper system for managing browser stuff across multiple projects, or is everyone just dealing with the mess?

Anthropic x Amazon just got bigger Amazon is investing another $5B (total now $13B), and in return Anthropic is committing to spend $100B+ on AWS over the next 10 years. They’re getting access to Amazon’s custom chips like Trainium (their Nvidia competitor), including future versions that aren’t even out yet. Plus priority access to new capacity as it comes online. Feels very similar to Amazon’s recent OpenAI deal where cloud usage was a big part of the investment. Anthropic is basically locking in massive compute now to support future models. https://www.anthropic.com/news/anthropic-amazon-compute

Google is rolling out “Skills” inside Chrome, turning prompts into reusable workflows that can run across different websites. Instead of rewriting the same instructions over and over, users can now save a prompt and trigger it instantly on any page. This builds on Gemini’s integration in Chrome and pushes the browser toward becoming more of an execution layer for AI, not just a place to ask questions. With features like running across multiple tabs and a built-in library of workflows, this starts to look like lightweight automation directly inside the browser. Source: https://techcrunch.com/2026/04/14/google-adds-ai-skills-to-chrome-to-help-you-save-favorite-workflows/ Apple is taking a much stricter stance on vibe-coding apps like Replit, Vibecode, and Anything. Updates are being blocked, and in some cases apps are being removed entirely. The main issue is Apple’s rule against apps that download or execute code, combined with concerns that users could generate and run unreviewed or even malicious apps. At the same time, AI tools are driving an 84% increase in app submissions, which is putting pressure on Apple’s review process. This is essentially a clash between open-ended AI development tools and Apple’s tightly controlled ecosystem. Source: https://techcrunch.com/2026/04/14/how-vibe-coding-app-anything-is-rebuilding-after-getting-booted-from-the-app-store-twice/

ChatGPT now has a $100/month Pro plan OpenAI just introduced a new $100/month ChatGPT Pro tier, filling the gap between the $20 Plus plan and the $200 Pro plan. This is a solid middle ground if you needed more Codex usage but didn’t want to jump straight to $200. The plan offers about 5x more Codex usage than Plus and is clearly built for longer, more demanding coding sessions. There’s also a launch promo running until May 31 where usage is temporarily boosted to around 10x compared to Plus, so early users get quite a bit more headroom. This is also a direct move against Claude Code at the same price point, with OpenAI positioning Codex as a stronger option for active coding workflows. Source: https://techcrunch.com/2026/04/09/chatgpt-pro-plan-100-month-codex/

Z.ai dropped GLM-5.1 Z.ai just released GLM-5.1, and it’s pushing into top-tier territory for open models. It’s currently ranked #1 among open-source models and #3 globally across benchmarks like SWE-Bench Pro, Terminal-Bench, and NL2Repo, putting it right up there with frontier systems. It also suggests the gap between open and closed models might be closing faster than expected. What’s interesting is the focus on long-running tasks. It can run autonomously for up to 8 hours, iterating and refining its approach over thousands of steps. This feels like a clear move toward more persistent, agent-style workflows. Blog: http://z.ai/blog/glm-5.1