The Setup: Two AIs, One Desperate Person
January rolled around and I did what millions of other people did. I opened Claude on one tab, ChatGPT on another, and decided I was going to let AI handle my entire monthly planning. Not just the big stuff. Everything. Calendar blocks, project breakdowns, priority matrices, decision trees for what to eat when I couldn’t decide.
The logic seemed sound. Both Claude 3.5 Sonnet and GPT-4o are genuinely impressive tools. OpenAI alone was reporting over 400 million weekly active users across its products by early 2025, and everyone kept telling me that the real productivity gains came from outsourcing the thinking work. So I outsourced it.
What I found was messier than the productivity gurus suggested. Not in a “this is a learning opportunity” way. In a “I think I broke something in how I make decisions” way.
The Paradox of Saving Time While Losing Confidence
Here’s what the data says, and here’s what I experienced. According to the Stanford HAI 2025 AI Index Report, knowledge workers using AI for scheduling and planning saved about 2.3 hours per week on average. I probably saved more than that. I was outsourcing entire thinking processes.
But there was this weird cost hiding underneath the time savings. The same Stanford research showed a 19% increase in decision fatigue for heavy AI planning users. Decision fatigue sounds abstract until you’re living inside it. By week three, I couldn’t prioritize my own tasks without running them through an AI first. I’d sit at my desk and think, “Should I work on email or the project?” and instead of trusting my gut, I’d open Claude and ask. Every single time.
This wasn’t laziness. It was something stranger. I had gotten used to outsourcing the decision, and now my brain had stopped doing that work altogether. A MIT Media Lab working paper published in January 2026 called this “reduced temporal self-efficacy” – basically, your confidence in managing your own time tanks after about 90 days of heavy AI planning tool use. I hit that wall in month one.
Where Claude and ChatGPT Actually Diverged
The two AIs failed in different ways, which was almost more frustrating than if they’d failed the same way. I could have picked a winner. Instead I got two distinct flavors of disappointment.
Claude wanted to have longer conversations. Anthropic’s internal research showed that Claude users need about 7.2 back-and-forth exchanges before getting an actionable plan. But when I asked Claude to plan my month, I expected maybe two or three iterations. That mismatch meant I’d get a plan that was close but not quite right, so I’d refine it, and Claude would refine it, and we’d go back and forth until I had something usable. By then I’d spent 40 minutes on what should have taken 10.
ChatGPT was faster at getting to a plan, but shallower. It would give me a structure that looked good on the surface – categories, time blocks, priorities – but when I tried to actually live by it, the plan fell apart because it didn’t account for the weird rhythms of how I actually work. I don’t work in neat blocks. I get momentum on something and ride it. ChatGPT’s recommendations assumed I was a different person than I am.
Neither one understood the actual shape of my life. They understood the inputs I gave them. That’s different.
The Dependency Trap No One Talks About
Here’s what genuinely concerns me, and what the data backs up. According to Microsoft’s 2025 Work Trend Index, 46% of AI-assisted workers now feel they can’t confidently prioritize tasks without checking with an AI first. That number was 19% just two years ago. It doubled. And I became one of those people.
I didn’t notice it happening. You don’t wake up one day suddenly dependent on something. You just start using it because it’s useful, then you use it more because it’s faster than thinking, and then one day you realize you genuinely don’t trust your own judgment anymore. By mid-month, I was running every small decision through one of the AIs. Not because I didn’t have the information to decide. Because I’d gotten comfortable letting something else decide for me.
The worst part? Both AIs were reliable enough that this felt reasonable. They weren’t giving me bad advice most of the time. They were giving me competent, generic advice. And competent, generic advice is a dangerous thing to get used to, because it trains you to stop trusting your specific, weird, non-generic instincts.
What Actually Happened When I Stopped
By February, I went cold turkey. Deleted both apps from my phone. Went back to a paper planner and my own brain.
The first week was uncomfortable in a way I didn’t expect. Not because I was less productive. Because I had to sit with uncertainty again. I’d make a plan and have to live with it without checking if an AI thought it was optimal. I’d prioritize something and couldn’t immediately validate that choice through another tool. The doubt came back.
But something else came back too. I started noticing patterns in how I actually work. Things Claude and ChatGPT never could have picked up, because they don’t know me over time. I know that Monday mornings I have about three hours of real focus before my brain gets scattered. I know that I’m not a “batch similar tasks” person no matter how many productivity systems tell me I should be. I know that sometimes I need to work on something small just to feel like I’m making progress on something big.
The AI plans ignored all of this. Not because the AIs were stupid. Because they couldn’t see it. They saw data points. They didn’t see the person.
If you’ve tried this experiment yourself, I’d genuinely like to hear where it went for you. Not the polished version – the actual version. Where did the AI planning help you? Where did it make things weirder? I’m curious whether this experience of outsourcing confidence is something everyone’s hitting or if I’m an outlier who was maybe too willing to hand it all over. The technology’s not going anywhere, and neither are we. Figuring out where the line actually is seems worth doing.