> Are you sure it’s _people_ driving this increase?
Most likely - yes. If Google has been dead for years people wouldn't pour hundreds of billions of dollars into ads there. The Search revenue keeps increasing, even since ChatGPT showed up. It might stagnate soon or even decrease a bit - but "death" ? The numbers don't back this up. One blog saying he stops paying for Google ads conflicts with the reality of around 200 billion yearly revenue from Search.
Exactly this. Businesses decide whether to pay for ads based on clickthru rates and conversions. Bots don't click through. They don't convert. If these rates fall, advertisers will pay proportionally less as their max bid, and Search ads revenue will fall substantially.
That hasn't happened. Google continues to grow with real users.
> It still needs to do learning (RL or otherwise) in order to do new tasks.
Why ? As in - why isn't reading the Brainfuck documentation enough for Gemini to learn Brainfuck ? I'd allow for 3-7 days of a learning curve like perhaps a human would need but why do you need to kinda redo the whole model (or big parts of it) just so it could learn Brainfuck or some other tool? Either the learning (RL or otherwise) need to become way more efficient than it is today (takes today weeks? months? billions of dollars) or it isn't AGI I would say. Not in practical/economic sense and I believe not in the philosophical sense of how we all envisioned true generality.
> Almost anyone can prompt an LLM to generate a thousand-line patch and submit it for code review. That’s no longer valuable. What’s valuable is contributing code that is proven to work.
That's really not a great development for us. If our main point is now reduced to accountability over the result with barely any involvement in the implementation - that's very little moat and doesn't command a high salary. Either we provide real value or we don't ...and from that essay I think it's not totally clear what the value is - it seems like every QA, junior SWE or even product manager can now do the job of prompting and checking the output.
The value is being better at it than any QA or product manager.
Experienced software engineers have such a huge edge over everyone else with this stuff.
If your product manager doesn't understand what a CORS header is good luck having them produce a change that requires cross-domain fetch() call... and first they'll have to know what a "cross-doman fetch() call" means.
And sure they could ask an LLM about that, but they still need the vocabulary and domain knowledge to get to that question.
That's an interesting argument, but from my industry experience, the average experienced QA Engineer and technical Product Manager both have better vocabulary than the average SWE. Indeed, I wonder whether a future curriculum for Vibe Engineering (to borrow your own term) may look more similar to that of present-day QA or Product curricula, than to a typical coding or CS curriculum.
Nah; the only advantage that a software engineer has is that if they are experienced they've probably just a little bit bright. But their role will probably change to something other than a software engineer. Bright valuable people that care and are engaged are rare anyway. They may transition to a different role slowly (e.g. Product, QA, BA, etc) because they still offer value and know the domain, but it isn't traditional SWE. That's been disrupted by AI; I don't want it to be true and I'm hoping for something else; but reality is staring us in the face at the moment and it isn't fair to people to talk platitudes anymore. The fact that you have to write an article like this feels like defensive framing to me + illustrates what happens once a skill is devalued by people/society due to disruption; it proves to me where this is all heading.
My thought on why people especially juniors are just delivering slop: Why bother with quality? Why bother with the craft? When it will be disrupted by the next tool/AI model/etc in the next few years anyway? Just think short term - will this slop get you through the PR and tick a short term box? If so success - might not have a job long term anyway due to all the AI stuff. In fact if I keep ticking boxes I'm more likely to last than the other person given more job incentives. Just get paid today.
In your example a QA that is skilled at testing websites should pick up CORS issues for example. And the models will keep getting better and eventually give them harnesses too - and we SWE will slowly automate everything around this because the only lifeboat left for your career is to cash out hopefully by disrupting yourself (no unions, professional bodies, etc).
LLMs are built based on human language and texts produced by people, and imitate the same exact reasoning patterns that exist in the training data. Sorry for being direct, but this is literally unsurprising. I think it is important to realize it to not anthropomorphize LLM / AI - strictly speaking they do not *become* anything.
> I feel the effects of this are going to take a while to be felt (5 years?);
Who knows if we'll even need senior devs in 5 years. We'll see what happens. I think the role of software development will change so much those years of technical experience as a senior won't be so relevant but that's just my 5 cents.
The way I'm using claude code for personal projects, I feel like most devs will become moreso architects and testers of the output, and reviewers of the output. Which is good, plenty of us have said for ages, devs dont read code enough. Well now you get to read it. ;)
While the work seems to take similar amounts of time, I spend drastically less time fixing bugs, bugs that take me days or God forbid weeks, solved in minutes usually, sometimes maybe an hour if its obscure enough. You just have to feed the model enough context, full stack trace, every time.
Man, I wish this was true. I've given the same feedback on a colleague's clearly LLM-generated PRs. Initially I put effort into explaining why I was flagging the issues, now I just tag them with a sadface and my colleague replies "oh, cursor forgot." Clearly he isn't reading the PRs before they make it to me; so long as it's past lint and our test suite he just sends the PR.
I'd worry less if the LLMs weren't prone to modifying the preconditions of the test whenever they fail such that the tests get neutered, rather than correctly resolving the logic issues.
We need to develop new etiquette around submitting AI-generated code for review. Using AI for code generation is one thing, but asking other people review something that you neither wrote nor read is inconsiderate of their time.
I'm getting AI generated product requirements that they haven't read themselves. It is so frustrating. Random requirements like "this service must have a response time of 5s or less" - "A retry mechanism must be present". We have a specific SLA already for response time and the designs don't have a retry mechanism built.
The bad product managers have become 10x worse because they just generate AI garbage to spray at the engineering team. We are now writing AI review process for our user stories to counter the AI generation of the product team. I'd much rather spend my time building things than having AI wars between teams.
Oof. My general principle is "sending AI-authored prose to another human without at least editing it is rude". Getting an AI-generated message from someone at all feels rude to me, kind of like an extreme version of "dictated but not read" being in a letter in the old days.
At least they're running the test suite? I'm working with guys who don't even do that! I've also heard "I've fixed the tests" only to discover, yes, the tests pass now, but the behavior is no longer correct...
> I feel like most devs will become moreso architects and testers of the output
which means either devs will take over architectural roles (which already exist and are filled) or architects will take over dev roles. same goes for testing/QA - these are already positions within the industry in addition to being hats that we sometimes put on out of necessity or personal interest.
I've seen Product Manager / Technical Program Manager types leaning into using AI to research what's involved in a solution, or even fix small bugs themselves. Many of these people have significant software experience already.
This is mostly a good thing provided you have a clear separation between solution exploration and actually shipping software - as the extra work put into productionizing a solution may not be obvious or familiar to someone who can use AI to identify a bugfix candidate, but might not know how we go about doing pre-release verification.
> I feel like most devs will become moreso architects and testers of the output
Which stands to reason you'll need less of them. I'm really hoping this somehow leads to an explosion of new companies being built and hiring workers , otherwise - not good for us.
> Which stands to reason you'll need less of them.
Depends on how much demand there would be for somewhat-cheaper software. Human hours taken could well remain the same.
Also depends on whether this approach leads to a whole lot of badly-fucked projects that companies can’t do without and have to hire human teams to fix…
This is what I'm doing, Opus 4.5 for personal projects and to learn the flow and what's needed. Only thing I'll disagree with is how the work takes similar amount of time because I'm finding it unbelievably faster. It's crazy how with smart planning and documentation that we can do with the agents, getting markdown files etc, they can write the code better and faster than I can as a senior dev. No question.
I've found Opus 4.5 as a big upgrade compared to any of the other models. Big step up and the minor issues that were annoying and I needed to watch out for with Sonnet and GPT5.1.
It's to the point where I'm on the side of, if the models are offline or I run out of tokens for the 5 hour window or the week (with what I'm paying now), there's kind of no use of doing work. I can use other models to do planning or some review, but then wait until I'm back with Opus 4.5 to do the code.
It still absolutely requires review from me and planning before writing the code, and this is why there can be some slop that goes by, but it's the same as if you have a junior and they put in weak PRs. Difference is much quicker planning which the models help with, better implementation with basic conventions compared to juniors, and much easier to tell a model to make changes compared to a human.
> This is what I'm doing, Opus 4.5 for personal projects and to learn the flow and what's needed. Only thing I'll disagree with is how the work takes similar amount of time because I'm finding it unbelievably faster.
I guess it depends on the project type, in some cases like you're saying way faster. I definitely recognize I've shaved weeks off a project, and I get really nuanced and Claude just updates and adjusts.
I'm impressed by this. You know in the beginning I was like hey why doesn't this look like counterstrike ? yeah I had the exepectation this things can one shot an industry leading computer game. Of course that's not yet possible.
But still, this is pretty damn impressive for me.
In a way, they really condensed perfectly a lot of what's silly currently around AI.
> Codex, Opus, Gemini try to build Counter Strike
Even though the prompt mentions Counter Strike, it actually asks to build the basics of a generic FPS, and with a few iterations ends up with some sort of minecraft-looking generic FPS with code that would never make it to prod anywhere sane.
It's technically impressive. But functionally very dubious (and not at all anything remotely close to Counter-Strike besides "being an FPS").
The same can be said about hucksters of all stripes, yes.
But maybe not contrarians/non-contrarians? They are just the agree/disagree commentators. And much of the most valuable commentary is nuanced with support for and against their own position. But generally for.
Most likely - yes. If Google has been dead for years people wouldn't pour hundreds of billions of dollars into ads there. The Search revenue keeps increasing, even since ChatGPT showed up. It might stagnate soon or even decrease a bit - but "death" ? The numbers don't back this up. One blog saying he stops paying for Google ads conflicts with the reality of around 200 billion yearly revenue from Search.
reply