There are moments when artificial intelligence truly astounds, and I recently experienced one while troubleshooting a minor issue at work with the help of Claude. What followed was a surprising revelation about how quickly these powerful tools can consume resources.
Each morning, I rely on Claude to generate a daily AI news briefing. It dispatches virtual “agents” to scour Reddit, social media, and industry websites, delivering a curated report of trending topics. Crucially, it also filters out content I’ve already covered in my reporting, ensuring fresh insights.
This briefing is automated through Claude Cowork, a feature allowing Claude to perform tasks directly on my desktop. The finished report is saved as a text file on my home Mac mini, accessible remotely through iCloud. It’s a helpful supplement to my own research, occasionally uncovering story ideas I’d otherwise miss – like the surprisingly popular trend of prompting Claude to speak like a caveman.
I was working from PCWorld’s Times Square office when I noticed the morning briefing hadn’t appeared in iCloud. The local Claude Cowork automation on my home system had likely failed. Being away from home, I couldn’t manually restart it.
Then I remembered Claude Dispatch, a feature enabling remote control of local Claude Cowork sessions via the mobile app. Could I use it to trigger the automation from afar? It was worth a try.
I opened the Claude app on my iPhone, navigated to the Dispatch tab, and sent a simple prompt: “Can you access the daily news brief automation in the AI story ideas project?” Projects are essentially dedicated workspaces within Cowork, containing specific instructions, automations, and relevant files.
Claude responded quickly. “Found it,” it confirmed, detailing the automation. “Is that the right one? And what do you want to do with the news brief automation—check it, modify it, or something else?”
I instructed Claude to manually trigger the automation. “Done,” it replied. “I’ve triggered a manual run for the daily news briefing, firing in about 1 minute.” Moments later, the briefing appeared in iCloud, exactly as expected.
It was a satisfying solution, a testament to Claude’s capabilities. But the real surprise came when I checked my Claude usage meter. I’m subscribed to the $20-a-month Claude Pro plan, which provides a limited usage window.
The brief Dispatch exchange and the subsequent automation run had consumed a staggering 66 percent of my five-hour allowance – all within minutes. The speed at which Claude processed the request was remarkable, but the cost was equally significant.
I wasn’t entirely shocked. I’ve observed firsthand how Claude’s agentic features and coding abilities rapidly deplete AI tokens. I recently exhausted a week’s usage allowance for Claude’s new Design tool in just half an hour.
The exact reason for the high consumption remains unclear. Perhaps the initial failed automation attempt used some tokens, or maybe the second run, combined with the Dispatch dialogue, sent Claude’s agents into overdrive. Regardless, the incident highlighted a crucial point.
Claude’s Dispatch feature proved invaluable in resolving the automation issue. However, this experience underscored that even affordable, flat-rate plans may struggle to accommodate these “cool” moments of advanced AI functionality.
