AI Is Not a Dev
22 comments
I’ve been in the industry for 30 years professionally and before that 10 years as a hobbyist assembly language (65C02, 68K, PPC and finally x86) programmer. To pretend that the previous “4GL” tools are analogous to what LLMs can do is silly.
I am always using the anecdote that there are projects I scoped pre-2023 that I would have had to have at least one or two junior devs to do help me do the work after I wrote out all of the design specs and now I can do myself in the same amount of time with ChatGPT. That’s along with my lead/architect role of talking to the client, a lot of project management work, design documents, helping sales etc.
It’s a step change.
While I hated every single 4GL tool that I’ve ever encountered (well HyperCard has a soft spot in my heart), because it was limiting, I can use ChatGPT, to create my infrastructure (CDK, Cloudformation, Terraform), and my code based on talking to it like I would a good junior developer.
I can’t tell a junior developer I need IAC done in Node (the CDK), and code written in Python using the SDK and expect them to immediately know how to use both across 130+ different services.
I have never once in my decade plus of being over projects and having the ear (or direct authority) to hire someone said “what I really need are some junior developers who don’t have any real world experience. That would definitely help me ship faster”.
Why would I do that instead of poaching a mid level developer who can hit the ground running without doing a year of negative work (ie take time away from the team asking dumb questions).
I am always using the anecdote that there are projects I scoped pre-2023 that I would have had to have at least one or two junior devs to do help me do the work after I wrote out all of the design specs and now I can do myself in the same amount of time with ChatGPT. That’s along with my lead/architect role of talking to the client, a lot of project management work, design documents, helping sales etc.
It’s a step change.
While I hated every single 4GL tool that I’ve ever encountered (well HyperCard has a soft spot in my heart), because it was limiting, I can use ChatGPT, to create my infrastructure (CDK, Cloudformation, Terraform), and my code based on talking to it like I would a good junior developer.
I can’t tell a junior developer I need IAC done in Node (the CDK), and code written in Python using the SDK and expect them to immediately know how to use both across 130+ different services.
I have never once in my decade plus of being over projects and having the ear (or direct authority) to hire someone said “what I really need are some junior developers who don’t have any real world experience. That would definitely help me ship faster”.
Why would I do that instead of poaching a mid level developer who can hit the ground running without doing a year of negative work (ie take time away from the team asking dumb questions).
> What's the benefit of over-anthropomorphizing a hammer?
Lowering dev salaries. Not a benefit to most of us, tho
Lowering dev salaries. Not a benefit to most of us, tho
If the myth of the "copy-paste dev" has any truth to it, then salaries are inflated.
On the flip-side, lists like "here are 80 agentic tools for your start-up" sounds like new opportunities for devs on quite a few dimensions, no?
On the flip-side, lists like "here are 80 agentic tools for your start-up" sounds like new opportunities for devs on quite a few dimensions, no?
have you tried those 80 agentic tools? They are not opportunities for devs, they are opportunities for the people selling them. We are 2 years into vibecoding and nobody can point to one great piece of software created using it.
I have tried quite a few and would defo agree with your conclusion. Also not fully buying the "you're holding it wrong" answers.
I point at long such lists of tools only to indicate a certain level of complexity, which will most likely fall in the realm of "oh, this is too technical for me. I should delegate this to Alice, because she good with tech". This is only shifting the problems and problems mean opportunity.
I point at long such lists of tools only to indicate a certain level of complexity, which will most likely fall in the realm of "oh, this is too technical for me. I should delegate this to Alice, because she good with tech". This is only shifting the problems and problems mean opportunity.
It's a new hammer
but one that is improving at an exponential pace and is developing capabilities to use itself with increasing reliability
It's easy to look at AI and draw a simple analogy to existing tools, because in most cases it is used as a tool, but the properties of intelligence and its ability to make things in the world is very unique and not comparable to any other tool.
All tools are useful because they require intelligence to use, and the tool magnifies the aim of intelligence. When the tools become intelligent themselves, certain recursive feedback loops will start to appear. Simply look at the quality of AI code outputs from 2 years ago compared to today.
but one that is improving at an exponential pace and is developing capabilities to use itself with increasing reliability
It's easy to look at AI and draw a simple analogy to existing tools, because in most cases it is used as a tool, but the properties of intelligence and its ability to make things in the world is very unique and not comparable to any other tool.
All tools are useful because they require intelligence to use, and the tool magnifies the aim of intelligence. When the tools become intelligent themselves, certain recursive feedback loops will start to appear. Simply look at the quality of AI code outputs from 2 years ago compared to today.
> one that is improving at an exponential rate
I don't know what AI you've been looking at but GPT-5 is not twice as good as GPT-4 which wasn't twice as good as GPT-3
I don't know what AI you've been looking at but GPT-5 is not twice as good as GPT-4 which wasn't twice as good as GPT-3
I'm referring to the long-horizon task benchmark which has been exponential since GPT-2
https://metr.org/blog/2025-03-19-measuring-ai-ability-to-com...
https://metr.org/blog/2025-03-19-measuring-ai-ability-to-com...
As long as you're not going too far down the "the hammer will start using itself" path.
Recursive feedback loops and fast pace of improvements are priced in.
Recursive feedback loops and fast pace of improvements are priced in.
They're not priced in inasmuch as many people consider it fundamentally impossible for AGI to be reached
From the junior devs I've seen? Even GPT-4 is better than 80% of junior devs. This is not to say "amazeballs", this is the lowest bar where someone actually gets paid.
Those juniors quickly stop being juniors, but while they are juniors, GPT-4 passes this minimal-chargeable-bar.
> What's the benefit of over-anthropomorphizing a hammer?
I go with what (IIRC) @TeMPOraL says on occasion: anthropomorphising them alerts you to the categories of error to expect, that you need to mitigate.
(This is separate to "why do people anthropomorphise this hammer?", to which the answer is IMO "this hammer loudly anthropomorphises itself whenever anyone so much as touches it").
Those juniors quickly stop being juniors, but while they are juniors, GPT-4 passes this minimal-chargeable-bar.
> What's the benefit of over-anthropomorphizing a hammer?
I go with what (IIRC) @TeMPOraL says on occasion: anthropomorphising them alerts you to the categories of error to expect, that you need to mitigate.
(This is separate to "why do people anthropomorphise this hammer?", to which the answer is IMO "this hammer loudly anthropomorphises itself whenever anyone so much as touches it").
Most of my personal beef with AI comes from the way that it's being framed as an end-all tool for everything, despite the fact that it really can only do a small subset of things right now. It's horrid at math, it's wrong about basic information given to it; I've had it give me code that just used a library that doesn't exist (granted, it only did this once).
It's a very new, shiny hammer, but a hammer isn't going to be able to carve a hole into wood. I look forward to it being regulated and finding a real, approachable niche for it to work into.
It's a very new, shiny hammer, but a hammer isn't going to be able to carve a hole into wood. I look forward to it being regulated and finding a real, approachable niche for it to work into.
With every technological leap we have pushback, it's natural.
I'm sure people complained that hammers were a useless invention and why would anyone not want to keep using rocks.
I'm sure people complained that hammers were a useless invention and why would anyone not want to keep using rocks.
Absolutely.
When complexity grows and lines between boundries of what's what blur, opportunity for misunderstanding sneaks in.
We should welcome scrutiny, though.
When complexity grows and lines between boundries of what's what blur, opportunity for misunderstanding sneaks in.
We should welcome scrutiny, though.
If I have to get a job again it will be nice to deal with less people.
I like people and working with them.
I absolutely hate corporate colleagues who have to play games and be lames instead of good engineers. People who operate out of fear because they genuinely suck.
Hopefully we can get rid of the managers asap, and also increase barrier to entry because we don’t need these roided out SWE whose whole identity is TC OR GTFO we need nerds.
I like people and working with them.
I absolutely hate corporate colleagues who have to play games and be lames instead of good engineers. People who operate out of fear because they genuinely suck.
Hopefully we can get rid of the managers asap, and also increase barrier to entry because we don’t need these roided out SWE whose whole identity is TC OR GTFO we need nerds.
It’s not a human. But it is useful. Heck sometimes it’s even been a senior dev to some of my queries, because I need the knowledge I don’t gave.
But it’s not a human. I can’t coach it and it’s not my friend. Boundaries and purpose
But it’s not a human. I can’t coach it and it’s not my friend. Boundaries and purpose
> Boundaries and purpose
This can't be emphasized enough!
This can't be emphasized enough!
Bro it was analogy. Stop taking things too seriously.
GenAI is fits in the same space, but with extra steps, benefits and drawbacks. It's not "a junior dev". It's a new hammer.
The craftsman enjoys the new hammer, pushes its limits, nerds out about the intricacies. Tools have limits in terms of wear and tear, plus cost.
Junior devs are humans, looking to survive and flourish. They pick up new tools faster than most. The only barrier the same that has always been: access to said tools and visibility over the outcome.
What's the benefit of over-anthropomorphizing a hammer?