Team mode just dropped, and I'm genuinely curious if it'll hit the same sweet spot we had back in summer. Remember when Opus 4.6 was burning through tokens like there was no tomorrow? Yeah, that's back in focus.
I've been watching the token consumption metrics, and it's wild—you're looking at 3-5x the token burn with Opus 4.6 compared to what I expected. If you're paying per token, this is not a detail you can gloss over. You need to bake this into your planning from day one. Seriously. I've pulled a few screenshots showing what Claude's actually doing under the hood, and it paints a picture: the model is thinking out loud, working through problems step by step, being thorough. It's not cheap, but you can see why.
Here's where it gets interesting though. The new tasks tool—replacing the older todo approach—seems designed specifically for this. Instead of juggling random jobs, you've got real structure. Dependencies. A way to see what feeds into what. And that changes how you think about bigger projects and features. Suddenly you're not patching together micro-tasks in parallel; you're building something coherent.
Personally, I'm still gravitating toward smaller, focused tasks that I can run in parallel. That's where my brain works best. But I'm getting this sense that team mode could breathe new life into the whole swarm idea—that energy where multiple agents are moving in the same direction, picking up slack for each other, actually collaborating. It's the kind of thing that makes you wonder what's possible if you commit to it properly.
The token cost is the real question mark though. For hobby projects and experimentation, it's fine. For production work? You're going to feel it in your budget. But if the throughput gains are real, maybe it's a trade worth making.
Part of the #100DaysToOffload documenting agentic development in 2026
