25% code review feels crazy low to me, especially in orgs where token usage is being pushed. Methinks it's closer to 50%, but what's worse are the implications of our how jobs evolve.
They've taken the part we loved about Eng (solving problems), farmed that out to robots, and then made us intellectual bouncers for the bar we call prod. We're looking for bugs that AI missed (and that we will be blamed for), and we're doing that for 10x the volume of code we used to write ourselves. Who wants to do THAT job?
Asking a problem solver / creative to be a gatekeeper is exhausting. People are losing their love of the craft, and they don't see it getting better.
Actually now that I think it over more, it's probably 75% review. I know Eng who haven't written a single line of code in six months. The token pushing is driving some very short-term incentives.
Yep, and IC interview loops are looking more and more like manager loops: additional behavioral rounds like Anthropic's Culture Fit and project deep dive presentations.
Improve your people skills, but also focus on product and business. As you start moving up the stack because of frontier changes, it is important to be able to think multi-dimensionally.
You’re pointing at something that’s already starting to show up in my own work: the need for a new kind of role that sits between AI and human builders. In my case, that’s a Software Architect position I’m actively shaping—someone who doesn’t just write code, but designs systems, manages a team of coders who collaborate with AI, and then reviews and approves what the AI-assisted workflows produce.
What you’re describing here makes me even more convinced that this kind of role will be essential in the very near future: less about brute-force coding, more about high-level thinking, quality control, and making sure AI output actually aligns with business goals, security standards, and long-term maintainability. It feels like we’re moving into a world where the real leverage is in how well we define problems and architect solutions, not just how fast we can type.
Fully aware that software architects exist. The point is they will become THE programming team supported by trained clerical workers who will generate code with AI. It’s going to be a huge thing. It’s already started in some companies.
Interesting. And once todays architects retire, who will replace them? Who will shepherd future AI coders who don’t possess any of the skills, instincts or judgement of the seasoned professional architects we have now? Future generations won’t build these necessary skills because from day one they will be pressing the easy button. So your vision seems pretty temporary, lasting probably only a few years before the entire system suffers from knowledge collapse.
I agree with you that today’s architects bring hard‑won judgment that we absolutely cannot afford to lose. My point isn’t that we’ll replace that experience with people who “just press the easy button,” but that the center of gravity of software teams is shifting.
What I’m describing is a model where the architect becomes the primary producer of understanding and accountability, and AI becomes the primary producer of code. In that world, the architect’s job expands: they design systems, set constraints, define patterns, and also design how AI is used, validated, and governed. The people around them aren’t “non‑technical button‑pushers” – they’re trained operators who can run workflows, read tests, and surface anomalies, but the architect owns the standards and the final decisions.
On the “who replaces today’s architects?” question: every generation has had to learn on top of new abstractions. We went from assembly to high‑level languages to frameworks and cloud services, and each time people said, “future engineers won’t really understand what’s going on under the hood.” Yet we still grow senior talent – we just change what “under the hood” means. The risk with AI is real, but it’s a training and governance problem, not a reason to freeze the model in place. We’ll need curricula and career paths that require people to build systems without AI, then progressively layer AI in, so they develop the mental models you’re talking about instead of skipping straight to vibe‑coding.
So I don’t see this as a brief, unsustainable phase before knowledge collapse. I see it as a transition where we elevate the architect from “person who occasionally reviews pull requests” to “system orchestrator” for humans and machines – and we redesign education and practice so that future architects still do the hard thinking, even if they don’t personally type every line of code. Companies that treat AI‑generated code as untrusted components and keep humans in the loop are already moving in this direction.
In other words, I’m not predicting the death of software craftsmanship – I’m arguing that craftsmanship will live mostly in architecture, verification, and system design, while more and more of the raw code is produced by tools instead of large engineering teams.
Software Architect is a position of the very near future. One where they manage groups of trained personnel using AI to write code. The SA will review the code and approve it for release. It's coming soon!
This all rings true. But are things better or worse now for the average software developer in terms of working conditions and job satisfaction? My impression is that devs are working harder and longer hours than before, and morale is at an all-time low. Every devs is expected to effectively become a tech lead and the EM and their manager may well have been laid off as unnecessary layers, leaving frontline devs accountable for outcomes but without any real leadership to guide and coach them. Devs are working on teams that are only a "team" in name because they're really a silo of one (+ agents) owning a project end-to-end. It all seems pretty dystopian.
I also want to point out the irony that many engineering leaders and blogs like this one in the early 2020's were adamant that EMs should write more code or else become useless, but now it's suddenly normal for nobody to be writing code. All that energy exuded for nothing 😃
25% code review feels crazy low to me, especially in orgs where token usage is being pushed. Methinks it's closer to 50%, but what's worse are the implications of our how jobs evolve.
They've taken the part we loved about Eng (solving problems), farmed that out to robots, and then made us intellectual bouncers for the bar we call prod. We're looking for bugs that AI missed (and that we will be blamed for), and we're doing that for 10x the volume of code we used to write ourselves. Who wants to do THAT job?
Asking a problem solver / creative to be a gatekeeper is exhausting. People are losing their love of the craft, and they don't see it getting better.
Actually now that I think it over more, it's probably 75% review. I know Eng who haven't written a single line of code in six months. The token pushing is driving some very short-term incentives.
Yep, and IC interview loops are looking more and more like manager loops: additional behavioral rounds like Anthropic's Culture Fit and project deep dive presentations.
Improve your people skills, but also focus on product and business. As you start moving up the stack because of frontier changes, it is important to be able to think multi-dimensionally.
You’re pointing at something that’s already starting to show up in my own work: the need for a new kind of role that sits between AI and human builders. In my case, that’s a Software Architect position I’m actively shaping—someone who doesn’t just write code, but designs systems, manages a team of coders who collaborate with AI, and then reviews and approves what the AI-assisted workflows produce.
What you’re describing here makes me even more convinced that this kind of role will be essential in the very near future: less about brute-force coding, more about high-level thinking, quality control, and making sure AI output actually aligns with business goals, security standards, and long-term maintainability. It feels like we’re moving into a world where the real leverage is in how well we define problems and architect solutions, not just how fast we can type.
Software architects already exist and until now have never needed an algorithm to perform their thinking.
I’m
Fully aware that software architects exist. The point is they will become THE programming team supported by trained clerical workers who will generate code with AI. It’s going to be a huge thing. It’s already started in some companies.
Interesting. And once todays architects retire, who will replace them? Who will shepherd future AI coders who don’t possess any of the skills, instincts or judgement of the seasoned professional architects we have now? Future generations won’t build these necessary skills because from day one they will be pressing the easy button. So your vision seems pretty temporary, lasting probably only a few years before the entire system suffers from knowledge collapse.
I agree with you that today’s architects bring hard‑won judgment that we absolutely cannot afford to lose. My point isn’t that we’ll replace that experience with people who “just press the easy button,” but that the center of gravity of software teams is shifting.
What I’m describing is a model where the architect becomes the primary producer of understanding and accountability, and AI becomes the primary producer of code. In that world, the architect’s job expands: they design systems, set constraints, define patterns, and also design how AI is used, validated, and governed. The people around them aren’t “non‑technical button‑pushers” – they’re trained operators who can run workflows, read tests, and surface anomalies, but the architect owns the standards and the final decisions.
On the “who replaces today’s architects?” question: every generation has had to learn on top of new abstractions. We went from assembly to high‑level languages to frameworks and cloud services, and each time people said, “future engineers won’t really understand what’s going on under the hood.” Yet we still grow senior talent – we just change what “under the hood” means. The risk with AI is real, but it’s a training and governance problem, not a reason to freeze the model in place. We’ll need curricula and career paths that require people to build systems without AI, then progressively layer AI in, so they develop the mental models you’re talking about instead of skipping straight to vibe‑coding.
So I don’t see this as a brief, unsustainable phase before knowledge collapse. I see it as a transition where we elevate the architect from “person who occasionally reviews pull requests” to “system orchestrator” for humans and machines – and we redesign education and practice so that future architects still do the hard thinking, even if they don’t personally type every line of code. Companies that treat AI‑generated code as untrusted components and keep humans in the loop are already moving in this direction.
In other words, I’m not predicting the death of software craftsmanship – I’m arguing that craftsmanship will live mostly in architecture, verification, and system design, while more and more of the raw code is produced by tools instead of large engineering teams.
Software Architect is a position of the very near future. One where they manage groups of trained personnel using AI to write code. The SA will review the code and approve it for release. It's coming soon!
This all rings true. But are things better or worse now for the average software developer in terms of working conditions and job satisfaction? My impression is that devs are working harder and longer hours than before, and morale is at an all-time low. Every devs is expected to effectively become a tech lead and the EM and their manager may well have been laid off as unnecessary layers, leaving frontline devs accountable for outcomes but without any real leadership to guide and coach them. Devs are working on teams that are only a "team" in name because they're really a silo of one (+ agents) owning a project end-to-end. It all seems pretty dystopian.
I also want to point out the irony that many engineering leaders and blogs like this one in the early 2020's were adamant that EMs should write more code or else become useless, but now it's suddenly normal for nobody to be writing code. All that energy exuded for nothing 😃