Part 6: What Comes Next

Is AGI already here, or still far off?

You could say both are right. Depending on where you set the bar for 'how much does it have to do to count as AGI,' today's AI either passes or falls short.

Set the bar low — 'if you can talk with it like a person, it's AGI' — and it has already passed. Turing's 1950 test was close to that, and AI today has even had cases of solving hard test questions better than the average person.

Set the bar high — 'if it can learn a task it's never seen on its own and finish it, it's AGI' — and it's not there yet. On a new test that came out in March 2026, AI got almost none of the problems people got all right.

Try it — Where do you set the 'AGI bar'? — move the slider and the answer changes

Slide left and the bar is low; slide right and the bar is high. The bars aren't survey results. They just show how the answer splits as you move the bar.

Setting the bar low vs setting it high

The verdict on the same AI splits because people are using different rulers.

So when someone says 'AGI is here' or 'it isn't,' there's one thing to ask first — what's your bar?

1. Q1What's the biggest reason people disagree about AGI?

   1. ① Different bars for 'how much counts as AGI'
   2. ② Nobody can measure how good AI is
   3. ③ AI changes every day
   4. ④ Someone is hiding the numbers on purpose

2. Q2If you raise the bar to 'learn a new task on its own and finish it,' what about today's AI?

   1. ① It still passes easily
   2. ② Far more say it's not there yet
   3. ③ It can't be judged at all
   4. ④ It's always better than people

3. Q3When you hear 'AGI is here,' what should you ask first?

   1. ① Who said it?
   2. ② How much does it cost?
   3. ③ What's the bar?
   4. ④ When did it come out?

Whether AGI is here isn't a question to ask the AI, it's a question for us: where do we set the bar?

Will my job disappear completely?

It's less that whole jobs disappear and more that one job splits into pieces, and only a few of those pieces move over to AI. That's closer to what's been observed so far.

AI is amazing at a single piece, but it's still weak at carrying a job from start to finish on its own. In a 2025 test that handed AI whole real-world jobs, the best AI finished just 2.5% of them.

So the pieces left for people change. The weight shifts toward deciding what's right, checking it, and taking responsibility, and sometimes brand-new job titles appear.

Try it — Whose piece is this?

Don't look at the job title, look at 'the piece of the work' and choose. Even inside the same job, the answer changes piece by piece.

Split one job into four pieces

The orange pieces are the ones AI is quickly taking over right now.

In 2023, a brand-new job title even showed up: 'prompt engineer' — when the pieces change, the names change too.

1. Q1Which best describes what's been seen so far between AI and work?

   1. ① Whole jobs disappear
   2. ② Work splits into pieces and a few move over
   3. ③ Nothing changes
   4. ④ Nothing is left for people to do

2. Q2What happened in the 2025 test that handed AI whole real-world jobs?

   1. ① It finished most of them
   2. ② It was twice as fast as people
   3. ③ Even the best AI finished only 2.5%
   4. ④ It couldn't even get started

3. Q3Which pieces mostly stay with people?

   1. ① Repeating things fast
   2. ② Taking notes
   3. ③ Memorizing a lot
   4. ④ Deciding and owning it

What AI takes isn't jobs, it's pieces. So become someone who's great at the pieces that are left.

Who teaches AI to say 'no'?

AI safety isn't magic that makes AI good. It's safety guards you add on top of the model, one at a time. There are three big ones — refusing, asking before acting, and checking afterward.

Adding the guards really does make a difference. In one 2025 study, the success rate of trick questions that get it to break the rules fell from 86% to 4.4%.

But it's not free. It takes more computing (about 24% more in that study), and you get 'over-refusal,' where it turns down perfectly fine requests too. So these days people measure refusing well and over-refusal together.

Try it — Taking the safety guards off and putting them on

It's the same model. Just take the guards on top off and put them back on, and watch how its behavior changes.

Same model, guards off vs guards on

What changed isn't the model, it's the guards around the model.

So a better question than 'Is this AI safe?' is 'What guards does it have?'

1. Q1What are the three AI safety guards from today?

   1. ① Fast, cheap, lots
   2. ② On, off, on again
   3. ③ Memorize, solve, get it right
   4. ④ Refuse, ask first, check afterward

2. Q2With the guards on, how did the success rate of trick questions change?

   1. ① It dropped from 86% to 4.4%
   2. ② It hit 0%
   3. ③ It stayed the same
   4. ④ It actually went up

3. Q3What price do safety guards come with?

   1. ① It gets faster
   2. ② More compute, over-refusal
   3. ③ It gets more accurate
   4. ④ No price at all

Safety isn't AI having a nice personality, it's guards someone paid a price to put on.

Which pieces of a robot have AI inside?

Robots look smart these days, but not because their bodies got better. It's because the same kind of AI we use got plugged into the robot's 'brain' spot.

Split it up and you'll see AI went into the eyes, the ears for instructions and the decision-making, while the pieces that actually move the hands and legs are still mechanical engineering. The two fields have different roots, and they started really coming together around 2023.

So the truly hard part isn't the thinking, it's the body. Balance has to be adjusted hundreds to thousands of times a second, and thinking can't go that fast, so robots run fast circuits and slow circuits separately.

Try it — Split a robot into pieces — where did AI go in?

Tap the buttons to see the same robot split up by era, and then all the way inside. Orange pieces have AI in them, green pieces run on mechanical engineering.

What it takes for a robot to make one move

The orange pieces are AI, and the last green piece is mechanical engineering.

So it's more accurate to say 'there's AI inside the robot' than 'the robot is AI.'

1. Q1What's the biggest reason robots look smart these days?

   1. ① AI like ours got plugged into the 'brain' spot
   2. ② Their bodies got much tougher
   3. ③ Their batteries got bigger
   4. ④ A person is controlling them from behind

2. Q2What was the biggest weakness of older robots?

   1. ① They were weak
   2. ② Moved only as told, so small shifts made it miss
   3. ③ They were too slow
   4. ④ They were too noisy

3. Q3Why do robots split work between fast circuits and slow circuits?

   1. ① To save electricity
   2. ② To be quieter
   3. ③ Balance needs 100s a second
   4. ④ To weigh less

Robots didn't turn into AI, AI went into the robot's first three pieces.