Interesting stuff. Between this and Ethan Mollick's post yesterday, I've chosen to set aside time each week to experiment with coding agents, which I hadn't explored previously. I think the implied technical barrier (even if low) turns a lot of folks off, even those working in the policy space. That's certainly a mistake.
I'm curious about this point:
"I would no longer be surprised if we saw AI in the macroeconomic statistics by the end of this year, both on the upside (growth, productivity) and downside (labor market dislocation)."
Is the implication here that you would have been surprised by this, or didn't expect it a year ago? Does this change your policy recommendations or thoughts about near term governmental action?
It doesn't change my policy recommendations especially; I always expected AI to effect macroeconomic statistics eventually, it just would have surprised me if e.g. o3 or Claude 3.5 were noticeable.
I don’t want AGI, nor any controversy about Robot Rights are Human Rights.
I want highly capable, obedient machines serving me.
And maybe writing code.
This was a great intro- inspiration piece. Thanks.
Please consider coining/ using the term aigent to differentiate the ai agents from humans, as we compare the work of travel agents to aigents, for instance. Tho most high quality agents will be efficiently using aigents. English needs this.
Your prediction about upcoming regulation reminds me of when the Internet came to people's homes in the 90s. The regulation that went around was pretty out of touch, with the Communications Decency Act being the most troublesome. Regulators seemed to think that if you run a chat room, it's similar to a physical space and that you're going to have a hall monitor camped out there and watching the people in the room talk to each other. If you actually run a chat room like that, though, then you lose many of the benefits of being able to cheaply poof a chat room into existence out of nothing.
For that matter, the network neutrality discussions, many years later, also seemed a little bonkers. "Neutrality" sounds good but is hard to define when you consider real-world issues like content delivery networks. If Netflix opens a data center and rents some networking bandwidth, then many homes are going to experience better bandwidth to Netflix's data center, for Netflix's data, than they will to other Internet requests they make. What does "neutral" even mean in a context like this?
In both cases, the concept is important, but rather than work out some framework that makes sense, most people seem to repaint it in what they are used to, e.g. corporations versus individuals, or the economy versus equality.
It will not be at all surprising for AI regulation to go the same way.
Great piece. My experience mirrors yours, although you have gone much further afield. I’m a management professor and I’ve been using Claude Opus 4.5 to create interactive learning exercises. The boost to my productivity has been off the charts. Thanks to your article, it will go up another notch.
Thanks for this post, Dean! The last line of code I wrote was HTML in the late 2000s, but with coding agents and their capabilities, I really felt the need to tap my toes into coding again and set up Claude Code over the holidays. When a year ago, Satya Nadella was calling the end of SaaS, I was very skeptical but I think you point of the premium for software moving from usability to interoperability with coding agents is an excellent prediction!
The other thing about the terminal vs a GUI is that the CLI doesn’t guide your thought process like a GUI does. This might be part of what’s so intimidating about the CLI…your thinking is unscaffolded. To your point, you need a mental model of the architecture of what’s happening and what’s possible, but once you do the sky’s the limit.
Love this, such a good piece! Gonna go have a play with Claude Code now
Interesting stuff. Between this and Ethan Mollick's post yesterday, I've chosen to set aside time each week to experiment with coding agents, which I hadn't explored previously. I think the implied technical barrier (even if low) turns a lot of folks off, even those working in the policy space. That's certainly a mistake.
I'm curious about this point:
"I would no longer be surprised if we saw AI in the macroeconomic statistics by the end of this year, both on the upside (growth, productivity) and downside (labor market dislocation)."
Is the implication here that you would have been surprised by this, or didn't expect it a year ago? Does this change your policy recommendations or thoughts about near term governmental action?
It doesn't change my policy recommendations especially; I always expected AI to effect macroeconomic statistics eventually, it just would have surprised me if e.g. o3 or Claude 3.5 were noticeable.
Dean, have you ever by chance read Greg Bear’s “Blood Music”? It’s what popped into my head when I read this post this morning.
I haven't! Will check it out.
Dean, are you able to share Github links to any of these projects that might contain prompt sources or the resulting artifacts?
Are you using git (or something else) to track changes the LLM agents make to the code/files?
I don’t want AGI, nor any controversy about Robot Rights are Human Rights.
I want highly capable, obedient machines serving me.
And maybe writing code.
This was a great intro- inspiration piece. Thanks.
Please consider coining/ using the term aigent to differentiate the ai agents from humans, as we compare the work of travel agents to aigents, for instance. Tho most high quality agents will be efficiently using aigents. English needs this.
Your prediction about upcoming regulation reminds me of when the Internet came to people's homes in the 90s. The regulation that went around was pretty out of touch, with the Communications Decency Act being the most troublesome. Regulators seemed to think that if you run a chat room, it's similar to a physical space and that you're going to have a hall monitor camped out there and watching the people in the room talk to each other. If you actually run a chat room like that, though, then you lose many of the benefits of being able to cheaply poof a chat room into existence out of nothing.
For that matter, the network neutrality discussions, many years later, also seemed a little bonkers. "Neutrality" sounds good but is hard to define when you consider real-world issues like content delivery networks. If Netflix opens a data center and rents some networking bandwidth, then many homes are going to experience better bandwidth to Netflix's data center, for Netflix's data, than they will to other Internet requests they make. What does "neutral" even mean in a context like this?
In both cases, the concept is important, but rather than work out some framework that makes sense, most people seem to repaint it in what they are used to, e.g. corporations versus individuals, or the economy versus equality.
It will not be at all surprising for AI regulation to go the same way.
Great piece. My experience mirrors yours, although you have gone much further afield. I’m a management professor and I’ve been using Claude Opus 4.5 to create interactive learning exercises. The boost to my productivity has been off the charts. Thanks to your article, it will go up another notch.
Thanks for this post, Dean! The last line of code I wrote was HTML in the late 2000s, but with coding agents and their capabilities, I really felt the need to tap my toes into coding again and set up Claude Code over the holidays. When a year ago, Satya Nadella was calling the end of SaaS, I was very skeptical but I think you point of the premium for software moving from usability to interoperability with coding agents is an excellent prediction!
I have no clue how to code but this is an interesting primer on these tools
The other thing about the terminal vs a GUI is that the CLI doesn’t guide your thought process like a GUI does. This might be part of what’s so intimidating about the CLI…your thinking is unscaffolded. To your point, you need a mental model of the architecture of what’s happening and what’s possible, but once you do the sky’s the limit.