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.
> 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?
Yes, that is how most functional organizations operate.
Architects are a bad meme, along the same lines as scrum masters and product managers.
Good software engineers spend more time on design than implementation, and the best are helping their teammates with design, and auditing designs of adjacent teams to ensure integration success.
So architecture is an important part of the job, but that doesn't mean that you can automatically declare a role with the title "architect" and make the whole organization do better design.
Architect roles are a bad meme because the architect always ends up knowing less about the system and how it is used than the engineers and support staff, yet they have more influence over the decisions affecting those things. That's the essence of an organizational anti-pattern/bad-meme any time it shows up: influence over resources anti-correlated with decision quality.
Additionally, architects often report to non-technical folks fairly high up the chain. Those people are bad at screening for such a technical role, sometimes even making the mistake of replacing an architect with an external hire.
I've always treated the architect role as more setting broad goals, defining boundaries for individual ownership, and technical dispute resolution. Sort of what you see with a lot of directors of movies where you get the best work by empowering individuals to completely own a carefully built little box that is a brick in the larger work.
Those could be done by normal management, but I've found it better to separate those discussions from management style discussions. Sort of like how the BDFL role in a large open source project is rarely anyone's actual boss.
Albeit, I'm still recalibrating what all this means in an agentic world. It makes sense with the little swarm I directly command, but the pressures for downstream ICs to ignore the art and taste of what they're doing is more than a bit unfortunate since the core of what I'm saying is to empower them to give a shit and be proud about what they built.
Agree, regardless of your engineering level, you need to own the security/performance within the scope of your change / ownership. Without that ownership, we're back to the old days where we ship things without caring about how they operate. I don't think that's where agentic coding tools are taking us.
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.
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..."
Curious if this is a culture issue or a volume issue.
In the distant past of a couple of years ago, I saw and gave tickets with higher ambiguity and opportunity for ownership and cross-team collaboration. There was breathing room to take that kind of time.
This year, however, as we're all aware, the expectation around speed is much higher. When more spinning plates are added, I want them to be more defined so they can be handled quickly. To be clear, though, I much prefer the former scenario.
Yeah. There are some teams I work with that are exactly like that, and it pisses me off. If I need to write the ticket in such detail then it would be faster for me to write the code myself.
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.
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.
Yeah agreed, I have not experienced this as a change in my job expectations, I just have different (better) tools now. It would have been a change if this hit during my first or maybe second job, but not for a very long time now.
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
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.
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".
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.
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.
Detroit and many other factory and heavy industry jobs collectively acted through the 20th century, which raised costs, slowed production, and simply accelerated their downfall. Introductions of any significant trade protectionist follow-ups ended up being scorned by the public (see the popular reaction to Trump 1, Biden, and Trump 2 tariffs regardless of efficacy).
The medical industry achieved some protection through regulatory capture, licensure, and throttling education. Doctors have very high salaries now but there's a shortage and they're overworked. Innovation is also throttled in that space as a result. Socialized systems aren't faring any better, the doctors I know in Canada and Japan are just as overworked, have to rush cancer patients through 5 minute appointments, and are paid far less.
There is no magic bullet for making a large profession sufficiently protected by mystification (anti-rationalization?).
I reject the notation that only the elites know how to run these society impacting technology companies. These people are anti-human and should be imprisoned. Workers of these companies should be able to elect their own leaders, their voices are worth more than the investors or shareholders.
The economy is the only place in our society that we allow tyrants to openly reign freely. We only have our entire existence as a species to know that monarchies, oligarchies, and authoritarians have hindered our capabilities to thrive as a species. There's no reason why these companies shouldn't be subjected to democratic norms too, they couldn't exist without our democratic societies in the first place; the expectation that only a select few elites should dictate the direction of technology in the US is deeply concerning and should be rejected with something better.
Workplace democracy is better.
---
Fun fact, Congress can impose that corporations become workplace democracies and it's totally legal. Whoa!
Don't let the anti-humans tell us how to run society. Society can be so much better if we rid these anti-humans from our ranks.
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
You'll notice the parent didn't saying anything about unions. Unions cannot help here, but collectively working to establish a legal professional designation (think doctors, lawyers, etc.) can.
> 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. They ditched the measuring productivity with "lines of code" and tickets closed and all that nonsense that was common place before that.
The current batch of tech CEO are 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.
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?
When you look at everything and only see communism, you just don’t actually have a cogent definition of communism. It’s a common problem in American political discourse. Anyone left of Clinton is called a communist in American politics.
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.
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.
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.
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.
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.
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.
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.
I think that the more you use LLMs to work over time, the better you get at intuiting how and why the harness, training processes, or minor version changes cause these mercurial changes in performance.
And I completely agree about performance optimization btw, though I don't think I would say LLMs are better than humans at security yet because of how poor their epistemics/short their context is.
They're both highly verifiable domains that you could easily train to be superhuman because they're also easy to generate and verify against synthetic tasks. And they probably bring a lot more knowledge about software's internals in-distribution while doing it, so I don't expect to ever be able to out-optimize frontier AI at squeezing performance out of hardware or into software.
But I would be cautious to deferring to their judgment in long-horizon architecture or the more procedural, threat-modelly/incentives side of security, and for other large scale judgment calls because they simply have no experience with them.
RL is turning them into Idiot Savants so jagged in understanding and performance that they defy human expectations for what kind of knowledge and behavior correlate (causally). A human engineer superhuman at hill-climbing low level performance would be at least 95%ile at at general architecture, since they're the same problem just at larger and slower scale.
There's no good hyperbaric benchmaxxing chamber devised for tasks at this level of scope/ambiguity yet. they fall back to platitudes and best practices (the "Reddit"/HN part of the weights). In my opinion, they are just kinda "ok" at architecture. Once you're in the pocket of what they've spent more time learning in parallel than any (or all) humans could, they just let their freak flag fly without hedging or asking about it.
To demonstrate: I had Claude earlier this morning spending 3+hrs pruning a pure x86 ASM implementation of symmetric DTLS/symmetric UDP, down to a 3.8KB binary, completely from scratch. It flew off and ripped. No human alive would or will ever be able do this kind of task as quickly - expand those "66 steps" and see for yourself.
But! Any human capable of doing this task would know to add support for certificates before spending an ungodly amount of effort optimizing a security protocol implementation that needs them but doesn't have them yet! Claude has just never done anything greater in scope than 1M tokens so: https://claude.ai/share/90de9350-81f9-46ea-a628-be625e8b9a7a
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?
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.
> 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.
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.
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.
Depends what "it" is, but for a lot of web SaaS companies, probably quite far. Think about thinks like Trello, Workday, Greenhouse, Zendesk, Jira etc. Definitely doesn't take much technical expertise to vibe code replacements for those.
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.
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.
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.
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!
Yesterday, I was debugging a performance problem with our Google Forms addon. It turned out my colleague had vibe coded the logging part, so that he could monitor the new installs every week. ... All of these are one line of code, but internally each of these makes multiple API calls to Google servers. Since logging was the first line to execute after installation, the welcome screen took 15 seconds to load.
AI is great at writing code. But it doesn’t know if you are launching the app or have millions of users, which it needs to make the right architecture call to log a timestamp vs email. It is also sycophantic and tries to execute your requirements exactly without pushing back, which means you end up with a slow app that takes 15 seconds to load a static welcome screen.
It is so paranoid about security that it will never write code to accept an unauthenticated payload. Even if you write it yourself, it will rewrite it into an authenticated payload and screw up performance or force users to sign in again on your app.
Just incredible seeing this post at the top next to "Unforced Navier-Stokes singularity, and the Generalized Riemann Hypothesis". We're at the "Japanese soldier who kept fighting 29 years after WW2" stage of AI denialism.
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.
Yes, that is how most functional organizations operate.
Architects are a bad meme, along the same lines as scrum masters and product managers. Good software engineers spend more time on design than implementation, and the best are helping their teammates with design, and auditing designs of adjacent teams to ensure integration success. So architecture is an important part of the job, but that doesn't mean that you can automatically declare a role with the title "architect" and make the whole organization do better design.
Architect roles are a bad meme because the architect always ends up knowing less about the system and how it is used than the engineers and support staff, yet they have more influence over the decisions affecting those things. That's the essence of an organizational anti-pattern/bad-meme any time it shows up: influence over resources anti-correlated with decision quality. Additionally, architects often report to non-technical folks fairly high up the chain. Those people are bad at screening for such a technical role, sometimes even making the mistake of replacing an architect with an external hire.
Those could be done by normal management, but I've found it better to separate those discussions from management style discussions. Sort of like how the BDFL role in a large open source project is rarely anyone's actual boss.
Albeit, I'm still recalibrating what all this means in an agentic world. It makes sense with the little swarm I directly command, but the pressures for downstream ICs to ignore the art and taste of what they're doing is more than a bit unfortunate since the core of what I'm saying is to empower them to give a shit and be proud about what they built.
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.
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..."
In the distant past of a couple of years ago, I saw and gave tickets with higher ambiguity and opportunity for ownership and cross-team collaboration. There was breathing room to take that kind of time.
This year, however, as we're all aware, the expectation around speed is much higher. When more spinning plates are added, I want them to be more defined so they can be handled quickly. To be clear, though, I much prefer the former scenario.
Now with AI coding, it pisses me off even more.
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.
The medical industry achieved some protection through regulatory capture, licensure, and throttling education. Doctors have very high salaries now but there's a shortage and they're overworked. Innovation is also throttled in that space as a result. Socialized systems aren't faring any better, the doctors I know in Canada and Japan are just as overworked, have to rush cancer patients through 5 minute appointments, and are paid far less.
There is no magic bullet for making a large profession sufficiently protected by mystification (anti-rationalization?).
https://en.wikipedia.org/wiki/Workplace_democracy
I reject the notation that only the elites know how to run these society impacting technology companies. These people are anti-human and should be imprisoned. Workers of these companies should be able to elect their own leaders, their voices are worth more than the investors or shareholders.
The economy is the only place in our society that we allow tyrants to openly reign freely. We only have our entire existence as a species to know that monarchies, oligarchies, and authoritarians have hindered our capabilities to thrive as a species. There's no reason why these companies shouldn't be subjected to democratic norms too, they couldn't exist without our democratic societies in the first place; the expectation that only a select few elites should dictate the direction of technology in the US is deeply concerning and should be rejected with something better.
Workplace democracy is better.
---
Fun fact, Congress can impose that corporations become workplace democracies and it's totally legal. Whoa!
Don't let the anti-humans tell us how to run society. Society can be so much better if we rid these anti-humans from our ranks.
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
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.
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. They ditched the measuring productivity with "lines of code" and tickets closed and all that nonsense that was common place before that.
The current batch of tech CEO are 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.
And yes walking is refreshing and should not be taken away.
Are they peer reviewing themselves? Who's pushing back on their ideas and decisions?
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.
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.
And I completely agree about performance optimization btw, though I don't think I would say LLMs are better than humans at security yet because of how poor their epistemics/short their context is.
They're both highly verifiable domains that you could easily train to be superhuman because they're also easy to generate and verify against synthetic tasks. And they probably bring a lot more knowledge about software's internals in-distribution while doing it, so I don't expect to ever be able to out-optimize frontier AI at squeezing performance out of hardware or into software.
But I would be cautious to deferring to their judgment in long-horizon architecture or the more procedural, threat-modelly/incentives side of security, and for other large scale judgment calls because they simply have no experience with them.
RL is turning them into Idiot Savants so jagged in understanding and performance that they defy human expectations for what kind of knowledge and behavior correlate (causally). A human engineer superhuman at hill-climbing low level performance would be at least 95%ile at at general architecture, since they're the same problem just at larger and slower scale.
There's no good hyperbaric benchmaxxing chamber devised for tasks at this level of scope/ambiguity yet. they fall back to platitudes and best practices (the "Reddit"/HN part of the weights). In my opinion, they are just kinda "ok" at architecture. Once you're in the pocket of what they've spent more time learning in parallel than any (or all) humans could, they just let their freak flag fly without hedging or asking about it.
To demonstrate: I had Claude earlier this morning spending 3+hrs pruning a pure x86 ASM implementation of symmetric DTLS/symmetric UDP, down to a 3.8KB binary, completely from scratch. It flew off and ripped. No human alive would or will ever be able do this kind of task as quickly - expand those "66 steps" and see for yourself.
But! Any human capable of doing this task would know to add support for certificates before spending an ungodly amount of effort optimizing a security protocol implementation that needs them but doesn't have them yet! Claude has just never done anything greater in scope than 1M tokens so: https://claude.ai/share/90de9350-81f9-46ea-a628-be625e8b9a7a
functionally similar
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!
AI is great at writing code. But it doesn’t know if you are launching the app or have millions of users, which it needs to make the right architecture call to log a timestamp vs email. It is also sycophantic and tries to execute your requirements exactly without pushing back, which means you end up with a slow app that takes 15 seconds to load a static welcome screen.
It is so paranoid about security that it will never write code to accept an unauthenticated payload. Even if you write it yourself, it will rewrite it into an authenticated payload and screw up performance or force users to sign in again on your app.
Just incredible seeing this post at the top next to "Unforced Navier-Stokes singularity, and the Generalized Riemann Hypothesis". We're at the "Japanese soldier who kept fighting 29 years after WW2" stage of AI denialism.
Architecture is a responsibility shared between your project leadership, whether you call them leads or principals.
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.
Love that attitude!