AGI Isn't Here. Let's Build Something Useful Instead.
The debate over AGI is a distraction from what matters. We break down what current AI models can and can't actually do, and argue for focusing on building practical systems that solve real problems today.
July 24, 2026 · 1 min read · SuperThinking team
Every few weeks, someone on Twitter declares AGI is here, or just a few months away. Then someone else writes a ten-page paper on why it's decades away. The debate is exhausting, and frankly, a bit boring.
It's a conversation about definitions, not capabilities. Is it about consciousness? Passing the Turing Test? The ability to reason abstractly? Who cares.
Let’s reframe the question to something more useful: Can an AI agent do the median human office job, autonomously, for a week? Can it manage a project, deal with ambiguous feedback, and figure out what to do next without a human holding its hand?
By that standard, we are not even close. And that’s a much more interesting place to start, because it focuses on what we need to build next.
What Models Actually Do
GPT-4 and its peers are incredible. They are autocomplete on steroids, phenomenal pattern-matchers trained on a vast portion of the internet. They are great at tasks that fit within a single, clean context window.
Give it a well-defined task, and it will often nail it:
- Code Generation: