Unpopular Opinion → AI Agents for Everyday Use is “Not There” Yet
I played around with AI Agents for everyday normal human use, and it was just meeeeeh. Here's what I found.
Every week, someone on Twitter (sorry, X) posts a video of an AI agent ordering their groceries, booking a flight, and doing their taxes before lunch. The comments are full of “This changes everything!” and “The future is here”.
Please, the future is not here yet. At least, not in that way.
The Hype vs The Reality
AI agents sound magical in theory: a digital assistant that can understand context, plan tasks, and execute them across different apps.
Imagine saying, “Plan my trip to Tokyo,” and your AI books flights, hotels, dinner reservations, and even arranges your pet sitter.
Sounds cool? Yeah, if it were in a movie.
In reality? Hmmm…. that’s a long shot.
I tried it, and realised that the gap between slick demos and everyday reliability is massive.
Issue #1 → Context Fragility:
Everyday decisions aren’t as simple as “book me a flight”.
Booking a flight actually requires a mountain of context: Which city am I flying from? Which dates? Do I prefer Emirates or Qatar? Window or aisle seat? Do I need extra baggage? Should it be refundable? What’s my budget? Am I travelling alone, with family, or with an infant? Do I need to land in the morning so I can make a meeting, or is evening fine?
By the time I feed an AI agent all that detail, I might as well have just opened the airline’s website and done it myself.
Agents today aren’t good at asking the right follow-up questions without overwhelming the layman, so the whole “let the AI handle it” promise just falls apart in real life.
Issue #2 → Tool Reliance:
Agents need access to third-party websites, apps and APIs to do anything useful. But most of the time, those connections are flimsy.
I asked an agent to “compare flight prices on Emirates and Qatar Airways”. It loaded one, got stuck on Qatar Airways’ website because Cloudflare blocked it, then declared, “Flight not available on Qatar Airways”.
The need to rely on third-party systems that probably weren’t part of the AI’s training data makes it quite a dynamic concern.
Issue #3 → Trust Issues:
I asked an AI agent to “check Carrefour and order a banana”. After a few minutes, it hit me with a login screen.
And then I had that moment of, “Do I really want to hand over my email and password to a random browser session controlled by an AI?”. All that for a banana?
It’s not like it wouldn’t work… I could go on, but the question is, at what cost? I could do it myself, or spend 10 minutes going back and forth with this agent after handing my email and password over to it.
But… It’s Day ONE, and This Is the Worst It’ll Be
It’s easy for me to dunk on AI agents right now, because they’re in the same stage as self-driving cars: they’re great at demos, they give a "I’m not sure about this” feeling on the road, and they’re not something you’d rely on without a human hand firmly on the wheel. At least, for everyday use.
But let’s be fair, this is version 1.0 of the idea; they’ll only get better.
Over time, they will learn to hold context longer. Integrations will get smoother (MCPs are already doing some good work). And trust frameworks will evolve.
So yes, right now they’re overeager interns. But interns don’t stay interns forever. This messy “day one” is the worst they’ll ever be, and that’s the exciting part.
I’m down for it.
Disclaimer: This piece is about AI agents for everyday personal use, such as shopping, booking, or running errands. AI agents already do a great job in other areas like research, automation, testing, and even creative brainstorming, but that’s not the conversation for today.