One term. A lot of very different ideas about what it means.

The problem with the word

AGI has moved out of research papers and specialist conversations and into everyday discussions about technology, work, business and the future.

The term appears constantly.

But scratch beneath the surface and something becomes obvious quite quickly: people aren’t always talking about the same thing.

For some, AGI means AI that can perform most intellectual tasks a human can.

For others, the important threshold is economic: a system capable of doing a significant proportion of useful human work.

Some focus on learning and adaptability.

Some question whether AGI is a useful category at all.

And increasingly, there’s another question underneath all of them: how would we actually know if we were getting close?

That’s what Understand AGI is here to explore.

Why I built it

I didn’t have a date circled on a calendar

I’m Sarah Wood, founder of Creative Sauce AI and the person behind Understand AGI.

I didn’t build this site because I have a date circled on a calendar for when AGI will arrive. I don’t.

I built it because the more deeply I worked with AI, the more interested I became in what was happening beyond the obvious improvements.

Models were getting better at writing and coding, but that wasn’t the interesting bit on its own. They were becoming capable across more kinds of tasks. They could work with different forms of information. They could use tools. Systems could be given more agency. Tasks that had required constant human input could increasingly be carried further without it.

At the same time, some surprisingly basic weaknesses remained.

That combination fascinated me. How can a system appear extraordinarily capable in one moment and strangely incapable in the next? And what would actually have to change before we could reasonably describe an AI system as general?

The more I looked into those questions, the less useful the simple headlines became. So I started building somewhere to examine them properly.

The better question

It’s bigger than “when will AGI arrive?”

It’s an irresistible question. When will we get AGI? You’ll find plenty of answers.

The problem is that a date means very little until you know what the person giving it means by AGI.

Instead, Understand AGI looks at the capabilities underneath the prediction.

Can a system apply what it knows to genuinely unfamiliar problems? Can it learn something new and use that knowledge elsewhere? Can it pursue a complicated goal over a long period? Can it recognise when its approach isn’t working and change it? How much human supervision does it still need? How reliable is it outside the conditions in which it performs best?

Those questions are less dramatic than a countdown to AGI.

They’re also much more useful, and they are what the syllabus on this site is built around.

Thresholds

AGI may not be a single moment

One of the ideas we explore throughout Understand AGI is whether we’ll recognise AGI as a distinct event at all.

There may never be a morning when the world wakes up to an announcement that says: AGI achieved.

Instead, we may see systems become progressively more capable, adaptable, reliable and autonomous. Different people may decide that an important threshold has been crossed at different points. And those thresholds may matter for different reasons.

Scientific capability isn’t the same thing as economic impact. Autonomy isn’t the same thing as intelligence. Passing a benchmark isn’t the same thing as being reliable in the real world. Being impressive across many tasks isn’t necessarily the same thing as being generally intelligent.

Understanding those distinctions is a big part of understanding AGI.

Disagreement

We’re interested in the disagreement

Understand AGI isn’t here to convince you that AGI is imminent. It isn’t here to convince you that it isn’t.

Some of the people closest to the technology disagree significantly about what AGI means, what capabilities are still missing and how quickly those capabilities might develop.

That’s not an inconvenience we need to edit out. It’s part of the story.

When someone makes a major claim about AGI, we want to know what sits underneath it. What definition are they using? What has actually been demonstrated? Is the capability general or narrow? How reliable is it? Is this something a system can do today, something demonstrated experimentally, or something somebody expects to happen?

And perhaps most importantly: what evidence would change the argument?

Why me

I came to this through technology, not academic research

My route into this subject is probably part of the reason I look at it the way I do. I didn’t come to AGI through academic AI research. I came to it through technology.

I’ve spent around 15 years immersed in web development and digital, watching the internet evolve from relatively self-contained websites into an enormous interconnected layer of platforms, databases, APIs, cloud services, ecommerce, analytics and business systems.

Before that, I worked in sales and infrastructure, including being part of a core team building a new business within a major global company and helping put the processes and infrastructure behind it in place.

Then AI arrived as another technological shift. I started experimenting with it, then using it, then building with it. Today I work with AI every day through Creative Sauce AI, developing products, applications, integrations and systems while following how the underlying technology is changing.

And working with it has made me increasingly interested in the gap between what AI appears capable of doing and what it can reliably do in the real world.

That’s the gap Understand AGI spends a lot of time looking at.

Who this is for

You don’t need to be an AI researcher to understand where this is going

But I do think you need to be willing to go further than the headlines.

Technical research matters. So do model evaluations, system cards, demonstrations and the actual capabilities appearing in products.

But eventually those capabilities encounter the rest of the world. Businesses. Software. Jobs. Schools. Governments. Science. Families. Economies.

The interesting question isn’t only whether an AI system can do something. It’s what happens when millions of people can use a system that can do it.

Understanding the technology matters. Understanding what happens when it leaves the lab matters too.

Method

How we approach it

There’s an enormous amount of prediction in the AGI conversation. Understand AGI tries to keep prediction and evidence separate.

When an AI lab releases something important, we want to look at what was actually released. When research makes a significant claim, we want to find the research. When somebody says AGI is three years away, we want to know who said it, when they said it and what they mean by AGI.

And when credible people disagree, we’d rather explain the disagreement than manufacture a neat answer that doesn’t exist.

The information here will change too. A page describing the state of AI in 2026 shouldn’t quietly sit unchanged while the technology moves around it. Where timing matters, we date the information. Where the evidence changes, we update it. Where nobody really knows yet, “we don’t know” remains a perfectly acceptable answer.

Disclosure

Yes, AI helps build a site about AI

Of course it does. I use AI extensively in my work, and it’s part of the research, organisation and development process behind Understand AGI.

But there’s an important distinction. AI can help investigate a claim. It isn’t automatically evidence for the claim. Important factual claims should lead back to something you can inspect for yourself: research, documentation, data or the original words of the person making the claim.

Using AI has also given me another perspective on this subject. It’s very easy to become convinced by what these systems can do. It’s equally easy to focus entirely on what they still get wrong. The interesting part is holding both ideas in your head at once.

Today’s AI can be remarkably capable and remarkably limited. Understanding how that changes may tell us far more about the road towards AGI than any prediction about the year it arrives.

So

Are we heading towards AGI?

Possibly.

But that’s not a very satisfying answer. The more useful questions are: what’s changing? What capabilities are emerging? What’s still missing? Which limitations are proving stubborn? What do different people mean when they say AGI? And what would we need to see before that description became convincing?

Those are the questions behind Understand AGI. Not because we already know the answers. Because this may be one of the most consequential technological transitions of our lifetime, and it’s worth understanding while it’s happening.

Sarah Wood
Founder, Understand AGI
Founder, Creative Sauce AI

Last reviewed 19/09/26