The muscle point is the one that survives contact with practice. I run a public experiment on exactly this: 498 drafts through an AI detector so far, and the only ones that read as human are the ones where a human wrote the majority of the source material. The tool did not decide the outcome. The reps did. Ng is describing the same curve from the learning side: skip the reps and you rent the muscle, do the reps and you own it.
This is an important distinction. AI is not necessarily replacing entire jobs. It is taking over specific, repetitive tasks and making human judgment, initiative, context, and relationship skills even more valuable. This closely reflects what I’m building with FounderFlow: AI handles the information overload, identifies what matters, and recommends the next action, while the founder remains responsible for the decisions. The future is not simply humans versus AI. It belongs to people who learn how to work intelligently with AI while retaining ownership of the outcome.
I feel like this was spot on. AI isn’t just eliminating everyone’s work but it’s a tool that’s reshaping how we get things done. My husband is an AI engineer and still spends time coding but he uses AI to write better and faster. Great read!
The part where Ng ships code and forgets it six months later is the clue. A tutor that is rewarded for giving the most satisfying answer has no reason to leave the difficulty in place, and the difficulty is what encodes the skill as yours. Retention drops because the tool did the one step that learning required you to do.
The 'cognitive offloading' point is equally relevant outside of the classroom, and something I've learnt in my career as an entrepreneur. While AI can make so many workflows more efficient, it does pay sometimes to get your hands dirty the old fashioned way. Really understanding your business down to the finest details is crucial if you want to sell it to customers/investors!
In the EU the staff-competence question is already a legal duty rather than a training preference. Article 4 of Regulation (EU) 2024/1689 requires providers and deployers of AI systems to take measures to ensure, to their best extent, a sufficient level of AI literacy among their staff and other persons operating the systems on their behalf, judged against those people's technical knowledge, experience, education and the context of use. A company with no AI team becomes a deployer the moment it puts a tool in front of employees. The Act sets no format, no hours and no certificate, which leaves the standard open.
Correction to my own comment above. Article 4 was replaced by Regulation (EU) 2026/1744, the Digital Omnibus on AI, published on 24 July 2026. The duty now reads take measures to support the development of AI literacy of staff, and the same paragraph states that it does not require providers or deployers to guarantee any specific level of AI literacy of any individual. Softer than I quoted, and the standard is even more open than I said.
Ng pointing at regulatory capture is the rare AI warning that doesn't flatter the incumbents. Most 'scary AI' panic does exactly the opposite — it hands the giants a moat and calls it safety.
I loved this for so many reasons. My top two are that we’ve seen this in language education for years. Machine learning built products and virtual classrooms had horrible pedagogical outcomes. It was all we had during COVID but it was the best we had at a very specific period of time. Nothing beats in person, learning from humans, scaffolded, task-based learning (or any other learning with a purpose). We also see this in language testing: test takers get better at taking the test but struggle in authentic communication with other humans.
I also loved how you laid out the argument of the strength of the narrative. I’ve tracked the corporate narrative over several posts and this could not be more clear about how much control this narrative has, and how woefully unprepared the average consumer is about the very scary headlines. Thank you for pulling all of this together
Point 5 is the one worth sitting with. Homework scores up while retention drops matches the older cognitive offloading research, though most of that work looked at students learning defined material, not an engineer asking a model how their own code works and shipping it. Different setting, same suspicion. Point 9 is the one teams actually skip, and the hard part is not writing the rule, it is that people cannot reliably tell what counts as sensitive while they are mid task, so the line has to be drawn by data class in advance rather than by judgment at the keyboard.
The muscle point is the one that survives contact with practice. I run a public experiment on exactly this: 498 drafts through an AI detector so far, and the only ones that read as human are the ones where a human wrote the majority of the source material. The tool did not decide the outcome. The reps did. Ng is describing the same curve from the learning side: skip the reps and you rent the muscle, do the reps and you own it.
This is an important distinction. AI is not necessarily replacing entire jobs. It is taking over specific, repetitive tasks and making human judgment, initiative, context, and relationship skills even more valuable. This closely reflects what I’m building with FounderFlow: AI handles the information overload, identifies what matters, and recommends the next action, while the founder remains responsible for the decisions. The future is not simply humans versus AI. It belongs to people who learn how to work intelligently with AI while retaining ownership of the outcome.
I feel like this was spot on. AI isn’t just eliminating everyone’s work but it’s a tool that’s reshaping how we get things done. My husband is an AI engineer and still spends time coding but he uses AI to write better and faster. Great read!
The part where Ng ships code and forgets it six months later is the clue. A tutor that is rewarded for giving the most satisfying answer has no reason to leave the difficulty in place, and the difficulty is what encodes the skill as yours. Retention drops because the tool did the one step that learning required you to do.
The 'cognitive offloading' point is equally relevant outside of the classroom, and something I've learnt in my career as an entrepreneur. While AI can make so many workflows more efficient, it does pay sometimes to get your hands dirty the old fashioned way. Really understanding your business down to the finest details is crucial if you want to sell it to customers/investors!
In the EU the staff-competence question is already a legal duty rather than a training preference. Article 4 of Regulation (EU) 2024/1689 requires providers and deployers of AI systems to take measures to ensure, to their best extent, a sufficient level of AI literacy among their staff and other persons operating the systems on their behalf, judged against those people's technical knowledge, experience, education and the context of use. A company with no AI team becomes a deployer the moment it puts a tool in front of employees. The Act sets no format, no hours and no certificate, which leaves the standard open.
Correction to my own comment above. Article 4 was replaced by Regulation (EU) 2026/1744, the Digital Omnibus on AI, published on 24 July 2026. The duty now reads take measures to support the development of AI literacy of staff, and the same paragraph states that it does not require providers or deployers to guarantee any specific level of AI literacy of any individual. Softer than I quoted, and the standard is even more open than I said.
Ng pointing at regulatory capture is the rare AI warning that doesn't flatter the incumbents. Most 'scary AI' panic does exactly the opposite — it hands the giants a moat and calls it safety.
I loved this for so many reasons. My top two are that we’ve seen this in language education for years. Machine learning built products and virtual classrooms had horrible pedagogical outcomes. It was all we had during COVID but it was the best we had at a very specific period of time. Nothing beats in person, learning from humans, scaffolded, task-based learning (or any other learning with a purpose). We also see this in language testing: test takers get better at taking the test but struggle in authentic communication with other humans.
I also loved how you laid out the argument of the strength of the narrative. I’ve tracked the corporate narrative over several posts and this could not be more clear about how much control this narrative has, and how woefully unprepared the average consumer is about the very scary headlines. Thank you for pulling all of this together
Point 5 is the one worth sitting with. Homework scores up while retention drops matches the older cognitive offloading research, though most of that work looked at students learning defined material, not an engineer asking a model how their own code works and shipping it. Different setting, same suspicion. Point 9 is the one teams actually skip, and the hard part is not writing the rule, it is that people cannot reliably tell what counts as sensitive while they are mid task, so the line has to be drawn by data class in advance rather than by judgment at the keyboard.