I used the same prompt on both platforms: build me a command center for ADHD.
One app to rule them all. Because context switching is exhausting when you're managing too many apps, tabs, and tools. I dictated the prompt chaotically, let both platforms run, and watched what happened.
Here's what the full piece covers:
- Google AI Studio vs Claude Code head-to-head: Both built working apps. Both needed about two prompts to get there. The real difference is what they're built for, not which is "better". Google handles logins, Firebase, and AI features automatically. No infrastructure thinking required. Claude Code went deeper on the idea without being asked.
- Claude Code's new Dispatch and Channels features: Scan a QR code, send tasks from your phone, work is done when you return. Channels hooks into Telegram or Discord via MCP. If you've been building this kind of async workflow manually (I have), this is Anthropic shipping it out of the box.
- Qwen 397B running at 5.5 tokens/second on a MacBook Pro M3 Max: Dan Woods built a custom inference engine in pure C and hand-tuned Metal shaders. The whole model, not a small one, on a laptop. The "you need more hardware" assumption about local models just changed.
- Anthropic's vibe shift: They're shipping fast and engaging differently. Opus 4.6 with 1M context became the default in Claude Code. They doubled usage limits for two weeks. Small things too, like actual conversations on social. Something changed in how they operate.
I don't usually do news-style posts. This time there was enough happening that I wanted to put my honest take on it somewhere.
Full post: https://thoughts.jock.pl/p/ai-opinions-march-2026-google-claude-anthropic
Free newsletter on AI agents, automation, and practical experiments: https://thoughts.jock.pl
Top comments (0)