Am I an outlier in that every SWE job I've had has expected you to push back on requirements and design the data model for tickets you are handed? Even when the org was large to have an architect, they just were just a figurehead and designed the original system, and ongoing decisions were made by the dev who picks up the ticket.
From that perspective, engineers are still engineers, just without the hands-on-keyboard coding. If that's all your job was, hands-on-keyboard coding with no decision making, yeah you're cooked and need to reskill.
[3 hidden]5 mins ago
It took me several years to realize this wasn't the default. Once I was senior enough to start getting people under me, I would assign tickets like "Scientist Bob would like a better method to view missing data, work with him to develop that over the next 2 weeks", and assume that they, like me, would enjoy the flexibility and the ownership. I didn't care what they came up with; email notifications, enhancing the web tool we have, putting coverage data in our opensearch to query, etc.
After getting complaints, I started getting narrower and narrower in my ticket writing, until I realized many newer software engineers literally want tickets like "modify function x_y_z() in python_file.py so that it also returns..."
[3 hidden]5 mins ago
I would say that I expect senior engineers to have autonomy, and autonomy means not doing strictly what you are told. So if you’re not being told exactly what to do, that means you have to figure it out for yourself. Which leads to steering task direction, which leads to pushback on requirements.
And that’s why junior engineers are fucked at companies that can afford to send all their proprietary information to Anthropic. But since they are public now, they will violate that trust as soon as major shareholders learn to word it so that it sounds like a stranded asset that any responsible C Corp would be expected to rescue. And then everybody’s fucked.
[3 hidden]5 mins ago
I run a development firm so we've had all kinds of clients and interacted with various external dev teams over the last 20+ years.
You aren't an outlier exactly, but there are way more devs than you are probably aware who are just told what to build and then have to go build it. So they don't have or never developed the consulting, planning, and architecture skills and experience.
The more our clients tend towards wanting a process like that, and the less they treat us like valued expert software consultants that can help shape their solution for the better, the worse the outcomes usually are.
[3 hidden]5 mins ago
I'm a self-taught woman who started work in agencies doing front-end work, then moving on to progressively more complex tasks, eventually becoming a lead and doing all of the things you mentioned. Your experience maps to mine 1:1. I've had a lot of imposter syndrome and challenges around this, and I had to go through a similar realization at some point in my career. It sort of sucks because I am just not good at being that kind of blinders-on developer. I also mention my sex/gender because I've had lots of borderline offensive assumptions made about me in certain situations related to what you've described.
[3 hidden]5 mins ago
We call those body shops. Probably because sweatshop sounds overwrought if you know what those are actually like.
[3 hidden]5 mins ago
If the new Engineers are making architectural decisions without a guiding Principal then it's a new job.
[3 hidden]5 mins ago
counterintuitively, i push back less now that im worried about standing out as "in the way of progress" when management is telling me to push their client manager nephew's vibe code to main
[3 hidden]5 mins ago
I'm seeing the Architects who only ever handed off a design based off of requirements to the dev and never implemented anything is equally cooked, in a different way.
There's really no sense in handing off the design to a dev to implement, who just copy pastes it into Claude's prompt. The "Architect" now needs to be fully responsible for what goes into production, instead of being able to blame some poor dev who faithfully implemented a flawed architecture design.
AI is flattening architecture and development into a single role.
[3 hidden]5 mins ago
We used to call that role a Software Engineer.
[3 hidden]5 mins ago
No one is getting promoted. The most likely future is that your job will pay less and will be less fun, as you're increasingly just a "liability sink" for decisions made by an LLM.
Try that with photography, painting, or music: "AI music has promoted musicians to architects".
[3 hidden]5 mins ago
It's weird seeing software devs, a highly educated workforce, get labor relations and labor history so wrong. The idea that our industry will keep paying high six-figure salaries will quickly come to an end unless we collectively act together.
The idea that rationalizing a job more (relying on metrics, measurements, statistics, predictions) will lead to better labor outcomes is foolish. We are only paid well as devs because we have comprehensible jobs (ones that require specialized knowledge and understanding).
We have over 200 years of history regarding what happens to labor when their industry becomes more rationalized and it has never ever benefitted labor once in the history of capitalism.
[3 hidden]5 mins ago
Yeah, it's incredibly bizarre that people are like "no we will get more valuable, see how much more we can do, we're more productive than ever!!!" when what's actually happening is the thing that you previously had a monopoly on (skilled software engineering) is vaporizing in front of you and the people writing paychecks haven't quite figured out how to value your role. It will leave a very large number of people behind entirely.
[3 hidden]5 mins ago
> unless we collectively act together.
I don't think that's going to work. There's no barrier to entry for vibe coding. Your union would have to include everyone in the world.
[3 hidden]5 mins ago
why do you think helpdesk or call center employees were not able to unionize and protect their jobs?
in the age of AI, the helpdesk/call center is really not different from that of a software engineer. The only diff is Helpdesk operator talks to a human. SWE are now just talking to LLM.
the job is perfectly mobile, can be offshored, outsourced, replaced, and million other ways to destroy any attempt to unionize.
unless, the society-wide labor unites as a single voting block for more redistribution from capital towards labor and universal federal ad-valorem wealth tax, instead of income tax
[3 hidden]5 mins ago
> The idea that rationalizing a job more (relying on metrics, measurements, statistics, predictions) will lead to better labor outcomes is foolish.
The interesting thing is that this exact thing happened before in the software industry. It's what killed the "old" software giants of the 90s. Only really microsoft was able to pivot and recover.
"Big Tech" became big tech by bucking that trend and being much more empowering to "low level" employees. Allowing talented people to build good things.
The current batch of tech CEO is going down the same path of morons like Jack Welch and we're all going to pay the price for it. I guess, if I were them, it's a reasonable choice. Despite leaving a wake of carnage and turning great companies into shells of themselves, they'll be billionaires before anyone figures out how utterly fucking stupid they are.
It's just very odd how, as an industry, we learn nothing from the past. Not at the executive level and not at the worker level. It reeks of arrogance at all levels.
[3 hidden]5 mins ago
you are barking up the wrong tree with that unionization horse apples. Unions plus technology do not go together. i’d rather take my chances on my own than deal with all that nonsense. What’s up with the the streak of communism in the comments lately?
[3 hidden]5 mins ago
Well no, ai music has promoted musicians to publishers, designers to art directors and developers to software engineering managers.
It’s just creating your own team of little robots working for your own personal “department”.
Though what I am seeing is that increasingly I am “setting up teams of agents” which each have their own “lead” and I just talk to these “leads” not individual agent “developers” . Effectively becoming a “VP” of engineering of my own private little “consultancy” … maybe next year (or month?) I will manage a bunch of “vps” as the “ceo” of said mini organisation … a product engineer indeed.
[3 hidden]5 mins ago
I strongly believe good software teams (very rare) require members to play many roles. Sometimes you're a PM, sometimes you're an architect, sometimes you're the systems engineer. Enough people have enough context on the bigger picture that it's stable and there's a shared vision.
More often than not, one person does the majority of this, and does their best to break up the responsibilities. A good CTO does that at the top, and it flows down through the ranks depending on how big the org is. Perhaps it's just a CTO and 5 engineers, perhaps it's a CTO with 5 divisions, each with 10 groups, with many teams.
Flatter structures of engaged members works best from what I've seen, but you need to have a critical mass of people not only willing to take initiative, but empowered by leadership.
[3 hidden]5 mins ago
The problem is that usually each team only needs a single architect.
What is happening is no different when developers got promoted into team lead of the offshoring team.
Which also didn't require that many team leads onsite, and off they were, everyone else that wasn't leading any offshoring team.
[3 hidden]5 mins ago
Agreed, and another point that I haven't seen in the article : 'architect' as a separate software engineering role was a bit of a bad word, even if everybody agreed that architecture was important.
It was mostly because in some companies there was a cast of 'architect' who were seen as above coding, and who mostly delivered out of touch UML diagrams, or ill-fitting data-structure. The core of the issue was that even good programmers would rust and get incapable of understanding the ground truth if the organization discourage them to also be in the implementation side.
Here it's not exactly the same thing with AI, since the loop is dramatically shorter, but the (lack of) feedback issue remains.
I really believe that my added value as a programmer was fundamentaly to evaluate technical tradeoff of the different options to solve a business issues, and them to implement them. Now with Claude code I'm infinitely faster at the implementation part, but I'm also eroding my capacity to evaluate those tradeoff.
Programming is a medium, and you also need to work with it from time to time in order to keep some level of competency. And yet in many companies we are basically forbidden to do that.
I feel that we have been given cars, but have also been forbidden to ever walk again; Sure the car is more efficient most of the time, but there's a an almost-vital reason why we have to do jogging to compensate.
[3 hidden]5 mins ago
But during that time the number of software engineering and development jobs and compensation levels continued to increase.
[3 hidden]5 mins ago
> each team only needs a single architect
Are they peer reviewing themselves? Who's pushing back on their ideas and decisions?
[3 hidden]5 mins ago
Usually the developers, which are now a bunch of agents instead.
[3 hidden]5 mins ago
It took me years to learn that an architect is a different job role than a developer. I always thought every developer must understand software architecture deeply to do their job. Later I discovered that there are architects who just draw diagrams. I was blown out.
[3 hidden]5 mins ago
We call them "marketects", and they practice "marketecture".
[3 hidden]5 mins ago
architect with diagrams is like McKinsey consultant with powerpoint slides.
functionally similar
[3 hidden]5 mins ago
I’ve found that the “job” of using AI is mainly foible management. What the author describes here is a habit of coding agents to excessively use and create logging infrastructure, and sloppily leave it in production code even when unnecessary. For instance, I created an IPA IME keyboard (https://ipabet.org/), and I found in review that Claude would leave the debugging tooling around unflagged, which essentially made the program a key-logger.
These foibles come and go: early Claude Code refused to let programs crash, mangled all the comments in your codebase, and then one week you find that it insists on ending all functions with `Of` as in `layoutOf()`, `viewOf()`. Finding and mitigating these annoying habits becomes part of the job. Code review becomes articulating with this entity why certain patterns in the code are silly and should be changed.
The author mentions performance and security issues as the domain of “the architect.” For performance, I’ve found that Claude Code has more or less solved this domain, and can profile and optimize code across a variety of stacks and languages using sophisticated, ad-hoc workflows. For security, it’s just an ever-ending nightmare where random habits and sloppy implementation create random vulernabilities, compounded by the fact that Anthropic has added “safeguard filters” whenever you naively say you’re looking for them. Nevertheless, I’m not sure the architecting role as they describe won’t be automated in the near future.
[3 hidden]5 mins ago
I can generally tell which PRs are AI on the OSS projects I do reviews for because the release notes are a Steinbeck novel instead of written as a one line description meant to be read among a litany of others. But then they sort of do that for everything don’t you think?
[3 hidden]5 mins ago
A dev moving down to architecture is not a "promotion".
Architecture is no more immune to AI than software engineering.
OP is describing somebody using AI poorly, and seems not to have a great handle on the way all jobs are impacted and boundaries being wiped away.
[3 hidden]5 mins ago
OP here. You may be right about architecture being wiped away in the future, but we have to ship with today's tools. Also, architecture is about tradeoffs and could have longer staying power:
> AI coding tools are tuned to implement your functional requirements and two non-functional requirements, i.e. security and backward compatibility to avoid breaking your app. This invariably leads to:
- Architecture that trades off performance for security and slows down the app.
- Multiple versions of datamodel design with redundant tables, columns and code.
[3 hidden]5 mins ago
> A dev moving down to architecture
Heh, at every company I've ever been at, an Architecture role is about the same as a non-banking VP, definitely not a demotion. It's often substantially more pay with substantially less work.
[3 hidden]5 mins ago
Architect very often was a coveted sinecure for failed programmers. It didn't necessarily pay better than a senior IC and they normally had no reports.
[3 hidden]5 mins ago
> Since logging was the first line to execute after installation, the welcome screen took 15 seconds to load
If you go from a (presumably) reasonable load time to one of 15 seconds and it’s not caught before getting merged, you have bigger problems than developer vs. architect and AI coding.
The premise of the post has merit, but the example is a terrible choice.
[3 hidden]5 mins ago
It's a pretty normal failure mode in teams. The abstraction breaks because the developer doesn't know the things that drive their latency.
Is it database requests? RPCs? Is there a function call which is computationally expensive?
Teams then respond by making rules to help make slow things stand out "no RPCs in functions, no synchronous RPCs, only 1 database read per request, don't call this code in a loop", etc.
Caught before merge? Only if you have someone really familiar with all the abstractions _and_ the team's previous failures reviewing it. Otherwise, developers are always finding new ways to write slow code. :)
But, that's why we have feature flags and incremental launches.
[3 hidden]5 mins ago
The role of developer is evolving, and it doesn't means developers scale into architects automatically. They get to learn, which is great.
[3 hidden]5 mins ago
Sorry but Nah. Architecture has always been a part of the job.
If you weren't doing it, you were doing it wrong.
[3 hidden]5 mins ago
That’s not just a continuum but also a timeline. Junior devs need to be encouraged to start thinking about why as much as what, and what more than how.
But they don’t teach you that in school. You get how and a bit of what. Maybe a little why in a sales pitch or an aside from a professor who hasn’t worked in industry except in research which is prototyping and we all know how people who only do prototypes are about rigor.
[3 hidden]5 mins ago
Exactly. You are now PM, Architect, and QA but get paid the same salary
[and your director now is convinced all his stupid ideas are both brilliant and very safe]
[3 hidden]5 mins ago
tell your director he can do it himself with AI.... and see how far he gets
[3 hidden]5 mins ago
If the pace of advancement of the last year continues, how many years until that's exactly what happens in >50% of software creation?
[3 hidden]5 mins ago
All the way to a class action suit and jail.
The thing is that most of us don’t have the patience to wait for the toddlers in ties to fuck around and find out. We have tuitions to pay and retirement funds to fill and we don’t want to be subpoenaed.
[3 hidden]5 mins ago
> Monitoring new customers is the most important part of running a business...
Priorities.
Useful question for "analytics" people. If you only collected data for every thousandth customer, would you have less useful insights?
[3 hidden]5 mins ago
Sample size is king.
[3 hidden]5 mins ago
I think we should not promote existing developers!
If AI is even remotely capable, we should fire all staff, and hire the same number of people in Nigeria! Diverse workforce is by far the most important goal! And it reduces CO2 emissions from commuting! And we save money!
[3 hidden]5 mins ago
It seem I was not only "promoted" to architect but also to PM/BSA. I'm having to gather requirements straight from managers and end-users and document everything, in addition to developing and leading another developer under me. The really interesting part is that my new role (after being pushed out of a Dev role) isn't even in the development/architecture family of jobs. Fuck corporations.
[3 hidden]5 mins ago
Backlog grooming has always felt like devs forcing other people at the company to do their jobs. But then Agile was always that even outside of Scrum.
[3 hidden]5 mins ago
This isn't even backlog grooming. It's up to me to create the backlog and all the analysis that goes with it, plus the stakeholder management that PM are typically supposed to do.
[3 hidden]5 mins ago
In addition to that and reviewing other people's AI slop MRs, I've also been promoted to devops. So many promotions.
[3 hidden]5 mins ago
True, I didn't mention the ops part. But at my company that stuff was lumped in my least 5 years ago, if not more.
[3 hidden]5 mins ago
I don't view becoming an architect as a promotion :/
[3 hidden]5 mins ago
I don’t view it as being a job title either.
Architecture is a responsibility shared between your project leadership, whether you call them leads or principals.
[3 hidden]5 mins ago
> You got promoted. Nobody told you, so I am telling you.
Love that attitude!
[3 hidden]5 mins ago
"Promoted". Some people wanna do.
Jesus, this is gonna end up being middle managers all the way down. Get a grip!
[3 hidden]5 mins ago
> If you are a senior developer or architect who understands the number of API calls behind each method, you would have:
Lol, sounds like something only a junior developer or intern would say.
[3 hidden]5 mins ago
Actually, you should always hill climb security and performance. AI can optimize these things if you ask it to.
[3 hidden]5 mins ago
Architecture is not hill climbing. Confusing the two is why software becomes slow and bloated - you believe low hanging fruit can fix all problems. They all end in local minima that I suppose is fine if nobody knows any better but users aren’t stupid. They can see that a similar app works 3x faster.
From that perspective, engineers are still engineers, just without the hands-on-keyboard coding. If that's all your job was, hands-on-keyboard coding with no decision making, yeah you're cooked and need to reskill.
After getting complaints, I started getting narrower and narrower in my ticket writing, until I realized many newer software engineers literally want tickets like "modify function x_y_z() in python_file.py so that it also returns..."
And that’s why junior engineers are fucked at companies that can afford to send all their proprietary information to Anthropic. But since they are public now, they will violate that trust as soon as major shareholders learn to word it so that it sounds like a stranded asset that any responsible C Corp would be expected to rescue. And then everybody’s fucked.
You aren't an outlier exactly, but there are way more devs than you are probably aware who are just told what to build and then have to go build it. So they don't have or never developed the consulting, planning, and architecture skills and experience.
The more our clients tend towards wanting a process like that, and the less they treat us like valued expert software consultants that can help shape their solution for the better, the worse the outcomes usually are.
There's really no sense in handing off the design to a dev to implement, who just copy pastes it into Claude's prompt. The "Architect" now needs to be fully responsible for what goes into production, instead of being able to blame some poor dev who faithfully implemented a flawed architecture design.
AI is flattening architecture and development into a single role.
Try that with photography, painting, or music: "AI music has promoted musicians to architects".
The idea that rationalizing a job more (relying on metrics, measurements, statistics, predictions) will lead to better labor outcomes is foolish. We are only paid well as devs because we have comprehensible jobs (ones that require specialized knowledge and understanding).
We have over 200 years of history regarding what happens to labor when their industry becomes more rationalized and it has never ever benefitted labor once in the history of capitalism.
I don't think that's going to work. There's no barrier to entry for vibe coding. Your union would have to include everyone in the world.
in the age of AI, the helpdesk/call center is really not different from that of a software engineer. The only diff is Helpdesk operator talks to a human. SWE are now just talking to LLM.
the job is perfectly mobile, can be offshored, outsourced, replaced, and million other ways to destroy any attempt to unionize.
unless, the society-wide labor unites as a single voting block for more redistribution from capital towards labor and universal federal ad-valorem wealth tax, instead of income tax
The interesting thing is that this exact thing happened before in the software industry. It's what killed the "old" software giants of the 90s. Only really microsoft was able to pivot and recover.
"Big Tech" became big tech by bucking that trend and being much more empowering to "low level" employees. Allowing talented people to build good things.
The current batch of tech CEO is going down the same path of morons like Jack Welch and we're all going to pay the price for it. I guess, if I were them, it's a reasonable choice. Despite leaving a wake of carnage and turning great companies into shells of themselves, they'll be billionaires before anyone figures out how utterly fucking stupid they are.
It's just very odd how, as an industry, we learn nothing from the past. Not at the executive level and not at the worker level. It reeks of arrogance at all levels.
It’s just creating your own team of little robots working for your own personal “department”.
Though what I am seeing is that increasingly I am “setting up teams of agents” which each have their own “lead” and I just talk to these “leads” not individual agent “developers” . Effectively becoming a “VP” of engineering of my own private little “consultancy” … maybe next year (or month?) I will manage a bunch of “vps” as the “ceo” of said mini organisation … a product engineer indeed.
More often than not, one person does the majority of this, and does their best to break up the responsibilities. A good CTO does that at the top, and it flows down through the ranks depending on how big the org is. Perhaps it's just a CTO and 5 engineers, perhaps it's a CTO with 5 divisions, each with 10 groups, with many teams.
Flatter structures of engaged members works best from what I've seen, but you need to have a critical mass of people not only willing to take initiative, but empowered by leadership.
What is happening is no different when developers got promoted into team lead of the offshoring team.
Which also didn't require that many team leads onsite, and off they were, everyone else that wasn't leading any offshoring team.
It was mostly because in some companies there was a cast of 'architect' who were seen as above coding, and who mostly delivered out of touch UML diagrams, or ill-fitting data-structure. The core of the issue was that even good programmers would rust and get incapable of understanding the ground truth if the organization discourage them to also be in the implementation side.
Here it's not exactly the same thing with AI, since the loop is dramatically shorter, but the (lack of) feedback issue remains.
I really believe that my added value as a programmer was fundamentaly to evaluate technical tradeoff of the different options to solve a business issues, and them to implement them. Now with Claude code I'm infinitely faster at the implementation part, but I'm also eroding my capacity to evaluate those tradeoff.
Programming is a medium, and you also need to work with it from time to time in order to keep some level of competency. And yet in many companies we are basically forbidden to do that.
I feel that we have been given cars, but have also been forbidden to ever walk again; Sure the car is more efficient most of the time, but there's a an almost-vital reason why we have to do jogging to compensate.
Are they peer reviewing themselves? Who's pushing back on their ideas and decisions?
functionally similar
These foibles come and go: early Claude Code refused to let programs crash, mangled all the comments in your codebase, and then one week you find that it insists on ending all functions with `Of` as in `layoutOf()`, `viewOf()`. Finding and mitigating these annoying habits becomes part of the job. Code review becomes articulating with this entity why certain patterns in the code are silly and should be changed.
The author mentions performance and security issues as the domain of “the architect.” For performance, I’ve found that Claude Code has more or less solved this domain, and can profile and optimize code across a variety of stacks and languages using sophisticated, ad-hoc workflows. For security, it’s just an ever-ending nightmare where random habits and sloppy implementation create random vulernabilities, compounded by the fact that Anthropic has added “safeguard filters” whenever you naively say you’re looking for them. Nevertheless, I’m not sure the architecting role as they describe won’t be automated in the near future.
Architecture is no more immune to AI than software engineering.
OP is describing somebody using AI poorly, and seems not to have a great handle on the way all jobs are impacted and boundaries being wiped away.
> AI coding tools are tuned to implement your functional requirements and two non-functional requirements, i.e. security and backward compatibility to avoid breaking your app. This invariably leads to:
- Architecture that trades off performance for security and slows down the app.
- Multiple versions of datamodel design with redundant tables, columns and code.
Heh, at every company I've ever been at, an Architecture role is about the same as a non-banking VP, definitely not a demotion. It's often substantially more pay with substantially less work.
If you go from a (presumably) reasonable load time to one of 15 seconds and it’s not caught before getting merged, you have bigger problems than developer vs. architect and AI coding.
The premise of the post has merit, but the example is a terrible choice.
Is it database requests? RPCs? Is there a function call which is computationally expensive?
Teams then respond by making rules to help make slow things stand out "no RPCs in functions, no synchronous RPCs, only 1 database read per request, don't call this code in a loop", etc.
Caught before merge? Only if you have someone really familiar with all the abstractions _and_ the team's previous failures reviewing it. Otherwise, developers are always finding new ways to write slow code. :)
But, that's why we have feature flags and incremental launches.
If you weren't doing it, you were doing it wrong.
But they don’t teach you that in school. You get how and a bit of what. Maybe a little why in a sales pitch or an aside from a professor who hasn’t worked in industry except in research which is prototyping and we all know how people who only do prototypes are about rigor.
[and your director now is convinced all his stupid ideas are both brilliant and very safe]
The thing is that most of us don’t have the patience to wait for the toddlers in ties to fuck around and find out. We have tuitions to pay and retirement funds to fill and we don’t want to be subpoenaed.
Priorities.
Useful question for "analytics" people. If you only collected data for every thousandth customer, would you have less useful insights?
If AI is even remotely capable, we should fire all staff, and hire the same number of people in Nigeria! Diverse workforce is by far the most important goal! And it reduces CO2 emissions from commuting! And we save money!
Architecture is a responsibility shared between your project leadership, whether you call them leads or principals.
Love that attitude!
Jesus, this is gonna end up being middle managers all the way down. Get a grip!
Lol, sounds like something only a junior developer or intern would say.