What Is AGI, Really, and How Close Are We? An Honest Map of the Expert Shouting Match

Two of the most credentialed people in artificial intelligence looked at the same technology this year and told you two different stories. One says we have crossed the event horizon of the singularity. The other says we have made perhaps 2% of the progress and has the grace to admit even that number is a vibe. Both cannot be right, both are smarter than me on this subject, and one of them is asking you for trillions of dollars. In my graduate AI studies I read the papers; in my two decades in the technical workforce, since 2006, I learned to read the incentives too. So let us do what nobody in this shouting match seems willing to do: define the damn terms, lay out who says what with receipts, give the warnings their strongest form instead of a strawman, and then look at the wreckage and the paychecks that are already real while the philosophers argue.

Key Takeaways

• Short answer: AGI has no agreed definition, so "are we there yet" has no answer. Capability is compounding on a measurable curve. Human-level generality still looks like a decades-scale project. Your business decisions should be built for the jagged, supervised agents we actually have.
• Definitions disagree. Andrew Ng’s classic bar is AI that can do any intellectual task a human can, and by that bar we are "many decades" away (Fast Company via PYMNTS, February 27, 2026). Microsoft and OpenAI, in a now-defunct contract clause, effectively defined it as about $100 billion in profits (Simon Willison, April 27, 2026).
• Optimists are specific and incentivized. Sam Altman wrote that humanity is "past the event horizon" of the singularity (The Gentle Singularity, June 2025). Dario Amodei has pointed to something like "a country of geniuses in a data center" around 2026 (Business Insider via AOL).
• Skeptics are specific and mostly not selling AGI. Ng offered a deliberately loose estimate of "a solid 2% of progress, with another 49 years to go," and 76% of surveyed AI researchers say scaling today’s approaches is unlikely to get there (The Decoder on AAAI 2025).
• Serious warnings are about indifference, not malice. Geoffrey Hinton puts a 10 to 20% chance on AI-caused human catastrophe within 30 years (The Guardian via AITopics, December 2024). Stephen Hawking’s anthill remains the cleanest statement of the risk.
• The paycheck problem arrived first. Secondary write-ups that cite Challenger describe AI as a top employer-stated reason for U.S. cuts in 2026, and a Stanford payroll study found a 13% relative hit to employment among ages 22 to 25 in occupations most exposed to AI (Stanford Digital Economy Lab, August 2025).

Disclosure: Agentic Trainers is paid to implement AI agents for businesses and to staff the human supervisors who keep them running. This post argues that today’s agents need supervision and that AGI timelines should be read with incentives in mind, which is work and a worldview my company sells. Weigh my opinion accordingly, and open the sources yourself.

Why Should You Listen to Me?

My first job, in 2006, was answering technical-services phones as an outsourced call center agent, where I learned that hype and payroll rarely share a calendar. Virtually all of my career since then has been on the technical side of outsourcing, including scaling strategy and the human-capital side of scaling for tech firms, later AI firms, and some semiconductor firms. I joined SuperStaff as CEO in 2019. Today I am CEO of SuperStaff, which is in its 18th year of business, and Managing Director of Agentic Trainers, the AI division of SuperStaff, which builds and supervises AI agent workforces for businesses. Graduate AI studies are underway. I am not a frontier-lab researcher, and nothing here is investment advice. My angle is narrower: how a small business should read the AGI shouting match without betting the company on a press release.

What Does AGI Actually Mean?

Direct answer: there is no agreed definition, and that single fact explains most of the shouting. The term is doing at least three jobs at once.

The original bar. Andrew Ng’s definition is the classic one: AI that can perform any intellectual task a human can, from learning to drive a truck to writing a PhD thesis. He has also noted, correctly, that the term has been defined loosely enough that "you could totally have reached AGI already or even 30 years ago" depending on where you put the bar.

The sliding commercial bar. Richard Socher offers an economic definition, 80% of jobs automatable, which he puts at 3 to 5 years, versus 10 to 200 years for the deeper version. Same two letters, two orders of magnitude apart.

The contractual bar, my personal favorite. Reporting on the Microsoft-OpenAI agreement revealed that AGI was tied to OpenAI generating about $100 billion in profits, later replaced by a provision where an independent expert panel would verify any AGI declaration, and finally rendered moot in the April 2026 restructuring. I almost admire the profit version. It is the only definition in the industry honest enough to admit what the word is for.

Hold onto this section, because it is the decoder ring for everything below. When someone announces a date for AGI without a definition, they have given you astrology with a press release. Ng predicted exactly this move: companies will declare AGI achieved by quietly reworking the definition down to whatever ships that quarter, and he has pointed out that OpenAI in particular "had an economic incentive to try to declare reaching AGI earlier."

Who Says It Is Close, and What Exactly Did They Say?

Direct answer: the leaders of the three biggest labs, with timelines between roughly now and ten years, and every one of them has a fundraise attached to the answer. The receipts, stated fairly:

Sam Altman, June 2025 essay "The Gentle Singularity": "We are past the event horizon; the takeoff has started." In the same essay he writes that 2026 "will likely see the arrival of systems that can figure out novel insights" and that 2027 "may see the arrival of robots that can do tasks in the real world," and he frames the singularity as gentle, already underway, and strangely undramatic.

Dario Amodei of Anthropic has pointed to powerful AI resembling "a country of geniuses in a data center," possibly by 2026.

Demis Hassabis of Google DeepMind, on 60 Minutes: AGI that matches humans at any task in the next five to ten years.

Geoffrey Hinton, who left Google specifically so he could speak freely, wrote: "I now predict 5 to 20 years but without much confidence."

Here is the fair reading and the unfair one together. The unfair reading is that these men are lying for money. I do not believe that. The fair reading is the one any veteran of sales learns in year one: the forecast of the person holding the quota is information about the pipeline and the pressure, both at once. The capability gains these labs ship are real and measurable, which we will get to. But "the singularity has started" is not a measurement. It is a mood with a capitalization table.

Who Says It Is Far, and Why Does Their Track Record Matter?

Direct answer: Andrew Ng and most of the surveyed research community, and the detail worth noticing is that the "far" camp is not selling AGI. Ng’s record here is unusually consistent. This is the man who taught the planet’s beginner course, AI for Everyone, back when the current panic was still a screenplay, telling millions of students that AI was a powerful tool and nowhere near a human mind. His current position has not moved with the hype cycle.

By the original definition, we are "many decades away from an AI that matches human intelligence." His metaphorical progress bar, offered with the explicit warning not to take the numbers seriously: a year ago AGI felt 50 years away, and "perhaps we’ve made a solid 2% of progress, with another 49 years to go."

Asked for a timeline, he has said only "I hope we get there in our lifetime, but I’m not sure", which, if you do the arithmetic on hope, is a project for our children more than for us.

On the fears themselves: losing control of AI "feels like science fiction" to him, and he has said flatly that the AI-as-nuclear-weapons analogy "has no basis in fact."

And he has company that does not fit the grumpy-skeptic stereotype:

Andrej Karpathy, OpenAI co-founder, a man who built the thing: AGI is still about a decade away, with today’s agents impressive but cognitively incomplete.

Yann LeCun: AGI is "not around the corner", years if not decades, with gradual progress rather than one breakthrough.

The field itself: in the AAAI’s 2025 presidential panel survey of 475 researchers, 76% rated it unlikely or very unlikely that scaling current approaches yields AGI. That is not a fringe. That is three out of four people whose job is the actual science, voting against the premise of the trillion-dollar buildout.

What Do the Doom Warnings Actually Claim?

Direct answer: not that AI will hate us, but that sufficiently capable systems pursuing any goal can destroy what they are indifferent to, the way a construction crew destroys an anthill. Let me give the warnings their strongest form, because they deserve better than the cartoon version, and then give the rebuttal its strongest form too.

The one-sentence version that the industry itself signed in May 2023: "Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war." Signatories included Hinton, Yoshua Bengio, and the CEOs of OpenAI, DeepMind and Anthropic, which produces the strangest fact in modern capitalism: the people building the technology co-signed a statement comparing it to nuclear war, and then went back to building it faster.

Hinton has since put numbers on his worry: a 10 to 20% chance that AI leads to human extinction within the next three decades, with the plain-language warning that we have never had to manage something smarter than ourselves. And the metaphor you have probably heard in some mangled form belongs to Stephen Hawking, from his 2015 Reddit AMA, and it is worth restoring to its original precision: "The real risk with AI isn’t malice, but competence... You’re probably not an evil ant-hater who steps on ants out of malice, but if you’re in charge of a hydroelectric green energy project and there’s an anthill in the region to be flooded, too bad for the ants. Let’s not place humanity in the position of those ants." No animosity required. The ants simply do not exist within the framework of the thing building the dam. That is the actual argument, and as an argument it is coherent: it requires only a system much more capable than us, with goals not quite ours, and resources we are standing on.

Now the strongest rebuttal, which is Ng’s: the argument is coherent the way science fiction is coherent, and it quietly assumes the capability leap that the evidence has not delivered. His words, not mine: today’s systems cannot do what the fear requires, the control-loss scenario "feels like science fiction," and when you research the data instead of the vibes, this is not AGI and it cannot do what people are scared of. My own read, labeled as my read: both camps are answering different questions. The doomers are arguing about what a superintelligence would do if it existed. Ng is arguing about whether one is anywhere in sight. You can believe Hawking’s anthill is airtight logic and still notice that nobody has poured the concrete.

So Is This the Singularity or Not?

Direct answer: capability is compounding on a measurable curve, which is real and impressive, and calling that curve "the singularity" is a claim no evidence can falsify, which is exactly why it makes such good marketing. The singularity, as coined, means the point where machine intelligence improves itself so fast that human prediction breaks down. Altman says we are past its event horizon. The best hard evidence anyone has for compounding capability is METR’s measurement that the length of tasks AI agents can complete at 50% reliability has been doubling roughly every seven months. That is a genuine, startling exponential, and honesty requires saying so.

Honesty also requires the other half: a doubling task horizon is not a self-improving mind. The same systems riding that curve still fail the long list of ordinary things Ng keeps pointing to, which is why 76% of the researchers surveyed do not believe this road, extended, arrives at human-level generality. The jaggedness is the story: superhuman at some tasks, confidently wrong at others, and I have written about what that confident wrongness does to businesses that mistake one for the other. An event horizon that lets you keep publishing quarterly updates from inside it is not an event horizon. It is a metaphor with a product roadmap.

Forget AGI for a Second. What Damage Is Already Real?

Direct answer: measurable job displacement, concentrated on the youngest workers, with the promised replacement jobs still mostly promissory. This is the part of the debate where both extremes agree without admitting it: you do not need AGI for AI to rearrange your economy.

According to reporting that cites Challenger, Gray & Christmas, that outplacement tracker tallied 112,713 announced U.S. cuts where employers named AI, covering January through July 2026, over twice the AI-tagged total the same tracker logged for the whole of 2025, and those write-ups called AI a leading stated reason in that stretch. We have not opened Challenger’s own release. Treat the number as an employer-stated-reason count from secondary coverage, not a census of AI-caused firings, and note that "AI" is partly a fashionable label on ordinary cost-cutting.

The sharpest measured effect is generational: a Stanford Digital Economy Lab analysis of payroll records put the relative drop in work for ages 22 to 25 at 13% in roles most open to AI automation, with that slide dating from the end of 2022, even as colleagues a few years older in identical titles kept their footing. The ladder’s bottom rung is being sawed off first.

The promised jobs are, so far, mostly a forecast: the World Economic Forum projects 170 million new jobs and 92 million displaced by 2030, a net positive that arrives on a schedule no displaced 24-year-old can eat. Meanwhile the firings ran ahead of the readiness: 55% of employers already regret their AI layoffs, and the rehiring boomerang is documented.

And the honest counterpoints, because this post does not do doom either: Ng argues the job apocalypse "is just not going to happen", noting AI automates tasks rather than whole jobs, and the US Chamber finds AI-using small businesses more likely to report growth, with 47% saying AI creates jobs in their business.

Both of these can be true at once, and in the data they are: no AGI anywhere in sight, and real paychecks disappearing anyway, especially at the entry level. Waiting to find out who wins the philosophy debate is not a workforce strategy.

Where Does the Argument Realistically Sit?

Direct answer: on the evidence, three positions are defensible at once: real compounding capability, no human-level generality on any confirmed horizon, and economic disruption that is already here. The timeline scorecard, so you can watch the predictions age in public:

Voice	Claim on record	By when
Sam Altman	Past the singularity’s event horizon; novel-insight systems; real-world robots	Now; 2026; 2027
Dario Amodei	"A country of geniuses in a data center"	As early as 2026
Demis Hassabis	AGI matching humans at any task	5 to 10 years (2025)
Geoffrey Hinton	Smarter-than-human AI, "without much confidence"; 10 to 20% catastrophe risk	5 to 20 years; within 30 years
Andrej Karpathy	AGI still about a decade away	About 2035
Andrew Ng	"Many decades"; a playful 2% progress bar with 49 years to go	Maybe our children’s lifetime
76% of 475 surveyed researchers	Scaling current approaches is unlikely to reach AGI at all	Not on this road

My own verdict, labeled as judgment and planted firmly on the fence where the evidence actually sits: the compounding is real, the generality is not, the definitions are being quietly devalued like a currency, and the first casualties were never going to be humanity. They were going to be entry-level payrolls, and the data says they already are. As CEO of SuperStaff, which is in its 18th year of business, I have watched every technology panic resolve the same way: the apocalypse stays scheduled, the restructuring happens on time.

So plan for the world the measurements describe. Agents that get startlingly better every seven months at bounded tasks, fail sideways without warning, and need trained humans deciding what they touch. If the optimists are right, that posture keeps you safe on the way up. If Ng is right, it is simply correct for the next few decades. Either way, the ant problem remains postponed, and the paycheck problem is on your desk now.

AGI is a definition fight. The payroll cuts are not.

If you want help designing supervision for the agents you already have, or reviewing which workflows are safe to automate before anyone whispers AGI, book a discovery call with us. For the practical side of agent supervision, start with our guide to AI agents for small business.

Frequently Asked Questions

What is AGI in simple terms?

The original definition is an AI that can perform any intellectual task a human can. There is no agreed standard, and definitions in commercial use range from "automates most jobs" to a reported contractual milestone of about $100 billion in profits, which is why claims of imminent AGI should always be checked against which definition is being used.

How close are we to AGI?

Nobody knows, and credentialed estimates span from "already beginning" (Sam Altman) through 5 to 10 years (Demis Hassabis) and about a decade (Andrej Karpathy) to "many decades" (Andrew Ng). In a 2025 AAAI survey of 475 researchers, 76% said scaling current approaches is unlikely to reach AGI at all.

Are we at the singularity?

No measurable event supports that claim, which is itself unfalsifiable. What is measurable is the trend that AI agents’ task horizons have doubled roughly every seven months (METR), which is genuine compounding capability but not a self-improving superintelligence.

Will AI really kill us all?

The serious version of the warning, from Hawking and Hinton, is about indifference rather than malice: a far more capable system pursuing other goals could harm humans the way a construction project destroys an anthill. Hinton estimates a 10 to 20% chance of catastrophe within 30 years; skeptics like Andrew Ng respond that today’s systems are nowhere near the capability the scenario requires and that the fear "feels like science fiction."

Is AI taking jobs right now, even without AGI?

Yes, measurably. Secondary write-ups that cite Challenger describe AI as a top employer-stated reason for U.S. cuts across months of 2026, and Stanford’s Canaries paper reports a 13% relative employment decline for the 22 to 25 cohort in high AI-exposure roles, while promised replacement jobs remain largely forecasts. Many AI-driven layoffs are also being reversed, which argues for redeploying people to supervise AI rather than firing ahead of the evidence.

---

Disclaimer: This article is commentary and opinion based on public sources available as of November 9, 2026. It is not investment, legal, career or safety advice. AGI definitions and timelines are contested; lab-leader quotes come from essays, interviews and secondary coverage noted in Sources. The Challenger AI-cited cuts figure is taken from secondary reporting that cites Challenger, Gray & Christmas; we did not open Challenger’s own release, and it is an employer-stated-reason count, not a verified count of AI-caused firings. Stanford, METR, WEF, AAAI and US Chamber figures carry the hedges of their original studies. The scorecard and verdicts labeled as mine are judgment, not forecasts. Agentic Trainers is paid to implement and supervise AI agents, a conflict of interest I have tried to disclose.

Written by Matt Narciso with AI-assisted research and drafting. Every fact and source was reviewed by a human editor.

Sources

1. AI Pioneer Andrew Ng Expects Decades-Long Wait For AGI, PYMNTS (February 27, 2026; summarizing Fast Company interview)
2. Andrew Ng says AGI is decades away, Fast Company (February 27, 2026)
3. Andrew Ng: The Biggest Opportunities in AI Aren’t Where You Think (transcript), Singju Post / Silicon Valley Girl Podcast (August 28/31, 2026)
4. The Gentle Singularity, Sam Altman (June 2025; primary essay)
5. OpenAI CEO Says We’ve Already Passed the ‘Superintelligence Event Horizon’, Decrypt (June 2025; secondary coverage of Altman essay)
6. We are past the event horizon: Sam Altman predictions, TechRadar (June 2025; secondary coverage of Altman essay)
7. Here’s how far we are from AGI, according to the people developing it, Business Insider via AOL (2024, updated 2025)
8. AI that can match humans at any task will be here in five to 10 years, Google DeepMind CEO says, OODA Loop (2025)
9. Andrej Karpathy: AGI is still a decade away, Gigazine (October 21, 2025)
10. Most AI researchers are skeptical about language models achieving AGI, The Decoder on AAAI 2025 Presidential Panel (March 2025)
11. Statement on AI Risk, Center for AI Safety press release (May 30, 2023)
12. ‘Godfather of AI’ shortens odds of the technology wiping out humanity over next 30 years, The Guardian via AITopics (December 27, 2024)
13. 10 Things We Learned from Stephen Hawking’s AMA, Mental Floss (Hawking 2015 Reddit AMA anthill passage)
14. Microsoft and OpenAI’s reported $100 billion AGI definition, Gigazine summarizing The Information / TechCrunch (December 27, 2024)
15. Tracking the history of the now-deceased OpenAI Microsoft AGI clause, Simon Willison (April 27, 2026)
16. Measuring AI Ability to Complete Long Tasks, METR (March 19, 2025)
17. Challenger says employers cited AI in 112,713 U.S. job cuts through July 2026, AI Understanding briefing citing Startup Fortune / Challenger (August 26, 2026; secondary; Challenger primary not opened)
18. Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence, Stanford Digital Economy Lab (August 2025)
19. Future of Jobs Report 2025: 170 million new jobs and 92 million displaced by 2030, World Economic Forum (January 2025)
20. 2026 Empowering Small Business Report, U.S. Chamber of Commerce