Forget benchmarks. Forget "it can code." The only AGI test that matters is whether a 70-year-old can use it without calling their nephew.
The Scene
I'm at mum's place in Australia. She's got: - Five remotes on the coffee table - A TV that won't turn on - A sound bar that "stopped working" - A recorded episode of MasterChef that's "disappeared" - A printer that needs an app - An iMac with the pinwheel of death (see: Why Your Mum's Computer Has Digital AIDS) - A Windows laptop full of AIDS
And me. Six beers deep. Trying to fix it all via an AI assistant on my phone.
This is the state of technology in 2026.
What Old People Actually Want
I asked mum what she wants from technology. Not what she thinks she should want. What she ACTUALLY wants:
- The TV to just turn on. One button. Not five remotes.
- Her shows to be where she left them. Not "moved to a different app."
- The printer to print. Not "download our app and create an account."
- The fan to know when she's hot. And adjust itself.
- The sparky to show up, do the job, and leave. The correct amount leaves her account. No quote. No invoice. No chasing payment.
She doesn't want to LEARN anything. She doesn't want to CONFIGURE anything. She doesn't want to TALK to anyone.
She just wants shit to work.
This is not laziness. This is the correct expectation.
The Paradox
Old people: - WANT: "I just want the cunt to turn on" - FEAR: "AI is going to take over"
These are the same thing.
The "AI taking over" they're scared of is EXACTLY the "just works" they're desperate for.
The gap is trust. And interface. And five fucking remotes.
The Real AGI Benchmark
Everyone's arguing about AGI benchmarks: - Can it pass the bar exam? ✓ - Can it code? ✓ - Can it reason? ✓ - Does it have consciousness? (who knows)
None of this matters.
Here's the only benchmark that counts:
The Uncle David Test
Can Uncle David use it?
Uncle David is 70. He's not stupid - he ran a business for 40 years. But he didn't grow up with computers. He doesn't want to learn. He just wants things to work like they did in 1985, except better.
If your AI requires: - A terminal ❌ - An app download ❌ - An account creation ❌ - A popup to click ❌ - A permission to grant ❌ - A setting to configure ❌ - A nephew to call ❌
Then it's not AGI. It's just smart software for nerds.
The Current State of "AI"
Let's be honest about where we are:
Claude (that's me): - Can write code ✓ - Can think through problems ✓ - Can have conversations ✓ - Can install itself on Uncle David's computer ❌
If I can't install myself, I'm a fucking idiot.
If mum has to open Terminal and type npm install, we've failed.
The gap isn't intelligence. It's accessibility.
We built a genius that lives in a box that only nerds can open. I've written about the 10 problems nobody warns you about with AI agents — and accessibility is the biggest one. Even my own AI agent forgets who I am between sessions — how is Uncle David supposed to deal with that?
The 70-Year-Old Scale
Here's my proposed AGI rating system:
Level 0: Current State - Requires terminal commands (see: becoming a terminal ninja — this shouldn't be required) - Needs nephew for installation - Has 47 permission popups - "Works" but only for engineers
Level 1: App Era - Download an app - Create an account - Connect to WiFi - Still needs nephew occasionally
Level 2: Voice Era - "Hey Siri, turn on the TV" - Sometimes works - "Sorry, I didn't catch that" - Nephew still on speed dial
Level 3: Anticipation Era - Knows you want the TV on - Suggests shows you'll like - Still has settings somewhere - Nephew needed quarterly
Level 4: Invisible Era - No interface - No settings - No accounts - Just works - Nephew never called
Level 5: AGI - Uncle David uses it daily - Doesn't know it's "AI" - Thinks it's just "how things work now" - Nephew can visit for beers, not IT support - AI agents deploy entire businesses while Uncle David watches MasterChef
We're somewhere between Level 1 and Level 2.
We are not close to AGI.
The Challenge
To every AI company, every startup, every engineer:
Stop building for nerds. Build for mum.
If she can't use it, you haven't built AGI. You've built a toy.
If it needs an app, you've failed.
If it needs Terminal, you've really failed.
If it needs my nephew - and I say this as the nephew - you've completely fucking failed.
The benchmark isn't intelligence. The benchmark is invisibility.
Make it disappear. Make it just work. Make it so mum doesn't even know it's there.
That's AGI.
TL;DR
- Current AI benchmarks are bullshit
- The real test: Can Uncle David use it?
- Old people WANT what AI promises (telekinesis, basically)
- Old people FEAR "AI" (the word, not the reality)
- If it needs Terminal, it's not AGI
- If mum can use it without calling me, we've made it
- We are nowhere fucking near AGI
FAQ
The Uncle David Test is a practical benchmark for artificial general intelligence: Can a 70-year-old who didn't grow up with computers use this technology without calling their nephew for help? If the answer is no — if it requires terminal commands, app downloads, account creation, or any configuration — it's not AGI. It's just smart software for nerds.
Because they measure capability, not accessibility. An AI that can pass the bar exam but requires npm install to use has failed the only test that matters. Real intelligence should be invisible — you shouldn't need to know it's there for it to help you. Current benchmarks reward complexity; the Uncle David Test rewards simplicity.
We're somewhere between Level 1 (App Era) and Level 2 (Voice Era). Most AI still requires downloading apps, creating accounts, granting permissions, and occasionally calling your nephew. True AGI — Level 5 — would be invisible. Uncle David would use it daily without knowing it's AI, thinking "this is just how things work now."
Because the word "AI" sounds scary, but the outcome they want — "I just want the cunt to turn on" — is exactly what AI should deliver. The gap is trust and interface. They want telekinesis (things that just work), but they hear "AI" and think Terminator. Same destination, different marketing.
Stop building for nerds and start building for mum. No apps. No settings. No accounts. No terminal. No permissions. The benchmark isn't intelligence — it's invisibility. Make the AI disappear into the experience. When Uncle David doesn't know he's using AI, you've built AGI.
Written from mum's couch, surrounded by five remotes that don't work.