I think this is also what can happen in the teaching profession as well - though many teachers currently use AI in a simple enough way to be seen as a time-saver (creating individualized worksheets for students, generating lesson plans, etc...), I think as more and more begin to experiment with agentic systems, this will also start to happen. An increase in possibilities means you can do things you never considered before which means spending more time not less on aiding students. This may have a net benefit overall, but I agree with the general principle that, though counterintuitive, AI may not make necessarily make us more efficient. I wrote about this last month.
Indeed ! Just read your piece I feel we are 200% in agreement … I was looking at this paradox from the beginner vs advanced usage as well in the link below.
In the end we converge on this idea that this “extra time” should convert in reality into higher quality higher intensity work. And so AI if well used may be a great multiplier of knowledge and progress (notwithstanding all the issues about politics, distribution of this extra value etc..)
"The bottleneck moves from producing to choosing." That's great. I'd add one thing: choosing well is a skill that atrophies when you stop practicing it. Most people using AI are automating the iteration, meaning the back and forth, the rejection and the refinement, which is exactly where creative judgment develops. The efficiency gains are real. The skill loss is quieter and slower to show up. By the time you notice it, you've forgotten what good felt like.
Very interesting point. Over the past few months, I have had the feeling that working is becoming more intense and ... tiring. AI is probably the reason.
AI takes over most of the easy tasks and this does not mean less work but more harder problems to solve. The risk is that organisations will just fill the recovered time with more complex work. This won't work on the long term. Work will need to be redesign so that people are not overloaded with complex task to burnout.
Jean-Paul, very insightful point that some of the work agents absorb was the inspection layer, the place where someone noticed the numbers felt wrong. That is the cost that does not show up on the efficiency slide, because you only see it the first time a wrong number sails through unchallenged. Removing the friction also removes the moment of noticing. Do you think that noticing can be designed back in deliberately, or is it something that only lives in a human doing the slow work?
Oh didn’t see your comment until now. Tbh I am not sure you can really design this check step efficiently. Like in airports studies show that agents are pretty inefficient to spot dangerous stuff because the job of checking on a monitor is super boring and attention killing. So yes I think designing for “human in the loop” is an open research problem ;)
I think this is also what can happen in the teaching profession as well - though many teachers currently use AI in a simple enough way to be seen as a time-saver (creating individualized worksheets for students, generating lesson plans, etc...), I think as more and more begin to experiment with agentic systems, this will also start to happen. An increase in possibilities means you can do things you never considered before which means spending more time not less on aiding students. This may have a net benefit overall, but I agree with the general principle that, though counterintuitive, AI may not make necessarily make us more efficient. I wrote about this last month.
https://fitzyhistory.substack.com/p/the-myth-of-ai-time-savings
Indeed ! Just read your piece I feel we are 200% in agreement … I was looking at this paradox from the beginner vs advanced usage as well in the link below.
https://theintelligencefabric.substack.com/p/why-ai-disappoints-at-productivity
In the end we converge on this idea that this “extra time” should convert in reality into higher quality higher intensity work. And so AI if well used may be a great multiplier of knowledge and progress (notwithstanding all the issues about politics, distribution of this extra value etc..)
"The bottleneck moves from producing to choosing." That's great. I'd add one thing: choosing well is a skill that atrophies when you stop practicing it. Most people using AI are automating the iteration, meaning the back and forth, the rejection and the refinement, which is exactly where creative judgment develops. The efficiency gains are real. The skill loss is quieter and slower to show up. By the time you notice it, you've forgotten what good felt like.
You are right that there is a strong temptation to set up 2 agents and let them optimize… and when you do so you loose your ability to judge clearly…
Very interesting point. Over the past few months, I have had the feeling that working is becoming more intense and ... tiring. AI is probably the reason.
AI takes over most of the easy tasks and this does not mean less work but more harder problems to solve. The risk is that organisations will just fill the recovered time with more complex work. This won't work on the long term. Work will need to be redesign so that people are not overloaded with complex task to burnout.
Add a heat wave to this... I need a nap.
Indeed heat wave is the (hot) cherry on the cake
Jean-Paul, very insightful point that some of the work agents absorb was the inspection layer, the place where someone noticed the numbers felt wrong. That is the cost that does not show up on the efficiency slide, because you only see it the first time a wrong number sails through unchallenged. Removing the friction also removes the moment of noticing. Do you think that noticing can be designed back in deliberately, or is it something that only lives in a human doing the slow work?
Oh didn’t see your comment until now. Tbh I am not sure you can really design this check step efficiently. Like in airports studies show that agents are pretty inefficient to spot dangerous stuff because the job of checking on a monitor is super boring and attention killing. So yes I think designing for “human in the loop” is an open research problem ;)