ChatGPT Scheduled Tasks: How to Lose a User in Three Days

I love new technology. In fact, I probably adopt it earlier than I should. Over the last forty years I’ve gone from Sinclair Spectrums and Psion organisers to Azure, AI, mobile apps and SaaS platforms. If there’s a shiny new tool promising to make life easier, chances are I’ll give it a go.

So when ChatGPT introduced Scheduled Tasks, I was interested.

The idea is genuinely good. As someone juggling clients, software projects, gym sessions, grandchildren, chickens, invoices and approximately forty-seven active ideas at any given moment, an AI-powered reminder system sounds exactly like the sort of thing I need.

Unfortunately, reality had other plans.

The Great Chicken Incident

It started with a simple reminder to put the chickens away in the evening. One reminder. One task. Nice and simple.

A few days later I started receiving multiple chicken notifications. Then more. Then even more. Eventually ChatGPT informed me I had reached the maximum limit of fifteen active scheduled tasks.

The only problem? I didn’t have fifteen scheduled tasks.

Apparently my chicken reminder had discovered how to reproduce.

Things Started Getting Weird

Like most software bugs, this one wasn’t obvious at first. The more I investigated, the stranger things became.

I found duplicate reminders, duplicate notifications, incorrect active task counts, reminders linking back to old conversations instead of task management screens, and occasional failures when trying to list the scheduled tasks at all.

At one point I was receiving notifications reminding me that a reminder existed to remind me about another reminder.

As a software developer, I know bugs happen. Every piece of software I’ve ever written has had bugs. That’s not the issue.

The issue is confidence.

Trust Is Everything

The moment a system tells me something I know isn’t true, I start questioning everything else it tells me.

That’s exactly what happened here.

When ChatGPT insisted I had fifteen active tasks despite me knowing I didn’t, trust started to disappear. When I couldn’t easily see what was active, trust dropped further. When task management behaved differently between iOS and macOS, confidence took another hit.

Eventually I discovered that deleting tasks required a long press gesture on iOS.

Not on macOS.

Not through the scheduled task screen.

Not through the conversations the tasks linked back to.

A long press.

I eventually managed to remove everything, but by that point the damage was already done.

The Architect’s Curse

One downside of spending thirty-five years building software is that you can’t stop analysing other people’s software.

When something behaves unexpectedly, I don’t simply get frustrated. I start investigating. I look for patterns. I reproduce the issue. I look for root causes and workarounds.

Most users would have given up long before I did.

The irony is that if one of my clients presented me with a system that duplicated records, produced incorrect counts, linked users to the wrong screens and behaved differently depending on the platform being used, I’d be logging defects and scheduling a fix. I certainly wouldn’t be calling the feature complete.

The Final Straw

The thing that finally pushed me over the edge wasn’t the duplicate reminders.

It wasn’t the chickens.

It wasn’t even the incorrect task count.

It was discovering that clicking on a scheduled task often took me back to the conversation where the task had been created rather than to the task itself.

Then, when I asked ChatGPT to list all scheduled tasks, I was greeted with a message stream error.

At that point I’d stopped using the feature and started debugging it.

That’s never a good sign.

The Bigger Lesson

The feature itself isn’t the problem. In fact, I still think Scheduled Tasks is a great idea.

The problem is that productivity tools are built on trust. Users will forgive missing features. They’ll forgive limitations. They’ll even forgive the occasional bug.

What they won’t forgive is uncertainty.

Once users stop trusting the information being presented to them, they stop relying on the system. Once they stop relying on it, the feature is effectively dead.

Final Thoughts

Will I revisit ChatGPT Scheduled Tasks in the future?

Maybe.

The concept is solid and I can absolutely see the value. For the right person, in the right situation, it could be incredibly useful.

But right now, after duplicate reminders, phantom task counts, confusing navigation, message stream errors and chicken-related notification spam, I’ve decided to step away from it.

For now I’ll stick to tools I know I can trust.

The chickens, thankfully, are safe.

And for the first time in several days, nobody is reminding me to put them away.