Have a Coherent AI Policy(brianmeeker.me) |
Have a Coherent AI Policy(brianmeeker.me) |
This frames the argument like a dichotomy. And to be honest, using the Social Media "vibe-coding" as a strawman risks anchoring against something that's a mirage.
There are plenty of good engineers getting good results whilst accepting code-ownership as a continuum.
> If Claude goes down tomorrow, can you still do your job?
This is a valid counterpoint, but doing software is already a tricky set of dependencies. The answer here isn't automatically "you need to be able to do everything". It could simply be also use Codex.
I think the overall point is well made, I just don't agree with the absolute framing. There are things you can hand over AI safely. Even if you start small and increment it'll have a decent impact.
Developers often depend on external libraries like things for Image processing, Numpy, etc. Do they have to "own" the code in those external libraries and review them, in the same way people sometimes insist you have to review all AI-generated code? Do developers have to be able to recreate Numpy by themselves, even if their field isn't necessarily numerical optimization etc?
Those seem like very high and unreasonable bars.
Yeah, I get that it means “you’re responsible for it”, but the phrasing irks me.
It's not a strawman. A lot of people are doing it.
While I agree with the sentiment I fear this won’t last long. I already find myself, when Claude goes down for 15 minutes due to whatever, kind of throwing my hands up and taking a walk assuming it’ll be up by the time I get back. Usually it is.
If it went away for good I’d be able to code, but would I want to? I’d be kind of bummed in a way. Which is odd because I used to tout myself as someone who like programming but I think what I’ve discovered is I like building.
It feels like the arguments here are very "status quo biased" - Ok, sure - I guess for some shops the code is the source of truth for the org and you really have to be prepared/equipped to deeply reason around the code and these so-called-patterns.
Really it seems to me that the era of "agentic coding" has uncovered some uncomfortable truths. This is purely speculative, but it seems like most shops maintain little-to-no specification documentation for defining the code behavior - the code IS the documentation and that feels dangerous from a software engineering / architectural statdpoint.
We all know it's such an insanely gameable metric you'd be insane to actually use it...
Aside from that (and assuming a large enough sample size) I think it's a safe experiment to at least bet on finding profitable use cases. In 1-2 years, after this experiment runs its course, not everyone will have "unlimited" usage.
Absolutely yes.
> You must be able to do your job if your AI tooling disappears
Absolutely not.
Look, I'm an alright programmer. Not good, far from great. Interpreted languages work for me; add all that strong typing and compilation and it starts to go beyond what I'm interested in. Nonetheless, pre-AI, I have shipped many very functional, production-grade applications for many companies.
Now, I can write stuff in Go, and Rust, and it's fantastic. So much faster. The AI likes the strong typing, the test-ability, predictability, it all makes total sense. I'm using this stuff all the time, but I have not learned any Go; I'm too busy focusing on the parts the AI cannot do for me, like real requirements gathering, architecture, fit and finish, engaging stakeholders, etc. that still require the human touch. Maybe I could have learned some Go using that time, but at the end of the day my employer is paying me for results, not for my edification!
There are now huge parts of my job I cannot do without AI. Sure, it's like 800-1200 bucks a month of extra cost; ok; but with that extra low-5-figs a year of cost I am a much better employee in terms of my capabilities. It's easily delivering ROI for me, and therefore for my employer. Instead of sitting around wishing I had a Go developer to ask for help implementing a simple feature in a Terraform provider, I can just fork it and add what I need, try to submit it upstream for inclusion, etc. and the lack of language specific skills is no longer holding me back.
Take away the tool and I can't do that part of the job anymore, sorry. I can still do a lot, but slower, and honestly it would feel like going from a car back to walking, now; walking's fun, I do it recreationally for the sheer joy, but when there's hundreds of kilometres to cover in a short amount of time, the car is clearly the correct choice. So too is it with AI: we've invented the car for computers and only a fool would pretend he can do everything the same without it now.
Imagine reading that version as someone who doesn't know how big companies work. "But then they'll just fire all the mid-level managers, since they don't do any of the actual work!" Haha, boy would you be wrong.
https://dora.dev/capabilities/clear-and-communicated-ai-stan...
Spoiler alert: if you can't do the job, you're not going to be doing the job much longer.
Obviously our ambitions expand due to better tools. I now commit to and deliver much more work than before LLMs, and — before then — ditto for frontend frameworks, generation 4 languages etc.
There are projects I now start without thinking twice that I never would have considered a few years ago.
That's what productivity looks like, and it makes you more valuable, and your job more secure (up until the ASI kills us all...).
"but when there's hundreds of kilometres to cover in a short amount of time, the trebuchet is clearly the correct choice."
you point it in the rough direction and distance you want to go, pull the lever, see if you hit your mark, adjust, pull the lever again, etc.
And once you have dialed in the variables for that particular piece of rock that one time, you write it down in a "skill.md" file and announce to everyone on the team "this trebuchet has been carefully calibrated. Trust it with your other rocks too."
That’s off by about three orders of magnitude. A trebuchet’s range is in the hundreds of meters, not hundreds of kilometers.
Unless you're working in a coding sweatshop, I don't see why you need AI to do what people have been doing for decades just fine without breaking a sweat.
What are you working on?
If other companies are able to tolerate larger amounts of tech debt while shipping new features faster then you'll be out of a job at some point when your company loses market share.
It's fine if you disagree with the idea that AI lets established companies ship faster. I'm not here to argue that. But I think it's pretty easy to empathize with "why might one need to change their behavior due to this new technology?"
You are obviously unaware of what the silicon valley companies are asking for and commiting to.
> There are projects I now start without thinking twice that I never would have considered a few years ago.
I'm sick of seeing this argument because it's not as persuasive as you think. If you were incapable of doing it before, why would I ever trust that you could properly evaluate the result? Even if I did, it's still like saying, "I never would've been able to do this project without a subordinate that knew how to do it, now look at me!" Okay? So why would I choose you when it sounds like I could pick anyone with basic programming knowledge to manage the subordinate since I clearly don't need someone with the know-how to do the thing, just someone capable of wrangling a coding agent? Might as well get the cheapest college CS graduate I can find.
Is not working in SV enough of a moat?
> If other companies are able to tolerate larger amounts of tech debt while shipping new features faster then you'll be out of a job at some point when your company loses market share.
I'm saying that B2B services are very common outside of SV and more focused on stability, compliance, long-term maintenance, and the operational knowhow that comes with all that rather than just shipping new features. It's not that there isn't some competition, but that the business is built on much more comprehensive partnerships than just being a software vendor. I can't believe I'm saying this, but "synergy" sometimes isn't just a meaningless buzzword.
When you try to jam "AI" into the mix, the disruption harms the business value. Many including myself would like to be enlightened if you think otherwise.
We're B2B SaaS in the Ed Tech space. It's very sales-driven. There's only so many players in the space, customers come with a laundry list of things they've seen others do and expect you to have those features, too. There are basic expectations that need to be met, some of those are compliance, but, sadly, a lot of what actually drives sales is just... flashy shit that looks good to those signing the checks not those using the underlying software. We lost a sale recently because someone was upset we didn't have the ability to give digital stickers to children - seriously.
You're more than welcome to tell the customer they're wrong and not give them their stickers. Or you can ask Claude to build stickers for you in two days and keep up with the Joneses.
Don't get me wrong. Customers aren't retained long-term with flashy shit. People churn out because of poor UX, security fears, pricing hikes, etc. Those frustrations tend to build over years and pain has to get pretty high because it's effortful to shift software providers. But, for getting new customers, sales is driven by flashy features and, at least in our experience, we need to be able to build those as quickly as our competitors or we lose out.
Even knowledge of how to drop to C is fairly rare in much of development, and you know what? That’s okay. We all specialize in our own areas of this beast of a field.
The software that gets used is the metric that really matters.
You can write 'perfect' software that runs at 100% efficient and never makes a mistake, but if no one ever downloads it and uses it, you've just engaged in a bout of intellectual masturbation.
And honestly I've seen the 'write better software' people complain for years as Microsoft just absolutely financially decimates them. And yea, Microsoft loves writing bloated electron crap. And "one of these days Alice, people are going to rise up and use less bloated software and Microsoft is going to die", lol, just fucking kidding, people will never do that.
I don't really know why you brought Microsoft up. I don't know anyone who thinks writing better software can displace Windows. Windows has absolute ecosystem capture, notably on the hardware front -- you can't write a better OS even if you want to because hardware vendors simply won't work with you, and even if you did write a better OS you have to contend with not having 30 years of software developed for it. Computers are increasingly falling into the domain of professionals (with smartphones displacing casual consumer usage), who need Photoshop, Excel, and all of the rest of professional tools and would put up with an inferior OS out of necessity, because the tools are more important than the OS.
Even then, Windows is an excellent piece of software, technically. It is handicapped by explicit anti-consumer decisions shoving things users don't want into it, but the kernel is hands down better than Linux's kernel, and the userspace is superior on technical merits if not user-friendliness merits. Windows has been going downhill in more recent years, but the gap between it and Linux is still massive.
Really, I don't know how Microsoft is your go-to example, when they are actually a producer of excellent software. Excel, the modern .NET ecosystem, C#, and Typescript are top-tier, and VSCode is perhaps the only software I've used that justifies being an Electron application, because it actually exposes the capability to completely customise every aspect of it and extend it with sandboxed extensions. To the extent people have grievances with Microsoft, it is largely because of deliberate monopolistic practices rather than the technical quality of their software.
I guess you are conceding that LLMs can't write good software, but are suggesting that good software doesn't matter. I think it does matter very much. In cases of monopoly control, people will begrudgingly use bad software, but they won't be loyal and you will bleed users over time as the frustrations build. But I think most critically, in cases where you don't already have monopoly control, nobody will use your bad software. This is why we haven't seen any vibe coded applications really taking the world by storm despite all the LLM hype. OpenAI and Anthropic can make you use their bad software because it's the gate to a useful proprietary tool, which is their real attraction, but Random Startup #482942 cannot make you use their bad software. Creating good software doesn't guarantee that Random Startup #482942 will succeed, given other market factors, but creating bad software guarantees they won't succeed.