Live data from Hacker News

Domain expertise has always been the real moat

brethorsting.com

571–580 of 592 posts

Re: Domain expertise has always been the real moat

#571
post #469

Earlier quoted context omitted.

I suspect that the reason AI chatbots are not very good at developing such specifications is that they don’t build a domain model in their “head” — the only models they have are already hardcoded in their weights. A human learning a domain, on the other hand, forms new neural connections representing the model of the domain that they build in their head.

Chatbots are good at helping humans arrive at the specification. Apart from people that has deep expertise and/or talent in this area, humans aren't very good at building these specifications. That's the topic of this exchange.

I disagree, unless it’s about a topic the LLM already knows or can research based on public information, or unless the domain expert can actually judge the specification and have it iteratively corrected by the LLM until they are satisfied. That’s generally not the case, and LLMs aren’t good at controlling the iteration themselves until they are satisfied, with the result being a complete and accurate specification for the domain.

If LLMs were good at that, then for coding we wouldn’t need to precisely specify what we want them to code, instead they would ask us until it’s crystal clear what precisely we want in all details.

Re: Domain expertise has always been the real moat

#572
My problem with such articles is that they mix the creation of software with the servicing of the software. It can be pretty complex to serve sophisticated software at scale in a cost-competitive and secure manner and it will remain so. So the engineering will remain important and domain expertise will remain important as well, as it has always been. Different business will rely on different combinations of these to create their own specific moats.

Re: Domain expertise has always been the real moat

#573

Earlier quoted context omitted.

Let them attempt to use ai to build business critical systems and when they waste tens amd hundreds of millions they can rehire master builders who know what they're doing

They will attempt and then soon enough they will be succeeding, and the master builders will have all retired, and that's the point that humans become dependent upon AI's. I aim to live a couple more decades and I sadly expect to see it play out this way.

Master builders ought to write blogs, books and speak at conferences and in podcasts for clout and subscriber money so the training data can get scooped up in the next loop. And pick the best of the best for master/apprentice training while you still can

Re: Domain expertise has always been the real moat

#575

Earlier quoted context omitted.

> So you taught yourself how to program. Do you not see that there's a difference between teaching yourself how to program so that you could build a web application and using a new technology that allows you to build a web application without having to learn how to program? Well, I guess where we're not lining up is I don't see that you are going to deliver a serious application that isn't the equivalent of table kno…

> Well, I guess where we're not lining up is I don't see that you are going to deliver a serious application that isn't the equivalent of table knocked together in a home workshop without, in practice, "learning how to program." My argument is that AI is now "good enough" that there are real people who are smart and tech-savvy but who don't know how to program who are building real applications, shipping them and usi…

OK, let me try putting this in a different way: you may be able to get away with not knowing syntax, but if you aim to deliver and support an application that people pay for, that wasn't the chief thing that was hard to learn about. The tools could improve to the point they deskill the work but today one must learn about enough concepts that, in practice, they're still "learning to program," much like people didn't stop needing to "learn to program" because C or garbage collection or IDEs or WYSIWIG UI editors came into fashion and made some kinds of knowledge less important.

> So who is paying for the tokens? You or your employer? If your employer is paying for them, what's the problem?

There is no problem but it's obviously not "democratizing" it to need to have an employer willing to pay thousands per month for AI tools. Now I'm wondering if you're the one being deliberately obtuse.

Re: Domain expertise has always been the real moat

#576

Earlier quoted context omitted.

> Well, I guess where we're not lining up is I don't see that you are going to deliver a serious application that isn't the equivalent of table knocked together in a home workshop without, in practice, "learning how to program." My argument is that AI is now "good enough" that there are real people who are smart and tech-savvy but who don't know how to program who are building real applications, shipping them and usi…

OK, let me try putting this in a different way: you may be able to get away with not knowing syntax, but if you aim to deliver and support an application that people pay for, that wasn't the chief thing that was hard to learn about. The tools could improve to the point they deskill the work but today one must learn about enough concepts that, in practice, they're still "learning to program," much like people didn't s…

> The tools could improve to the point they deskill the work but today one must learn about enough concepts that, in practice, they're still "learning to program,"...

Then we'll have to agree to disagree. There are people building, deploying and selling applications using AI who aren't doing anything close to what I would consider "programming". This is so far beyond the comparison to an IDE or WYSIWYG editor.

> There is no problem but it's obviously not "democratizing" it to need to have an employer willing to pay thousands per month for AI tools.

Why does your employer need to be willing to pay thousands per month for AI tools if you don't need AI to do your job? Can't you just tell your employer you don't need AI? If you use 0 tokens, don't they pay for 0 tokens? Or do you have an employer who is forcing you to use AI? How are you using it if you don't need it?

Re: Domain expertise has always been the real moat

#577

Earlier quoted context omitted.

OK, let me try putting this in a different way: you may be able to get away with not knowing syntax, but if you aim to deliver and support an application that people pay for, that wasn't the chief thing that was hard to learn about. The tools could improve to the point they deskill the work but today one must learn about enough concepts that, in practice, they're still "learning to program," much like people didn't s…

> The tools could improve to the point they deskill the work but today one must learn about enough concepts that, in practice, they're still "learning to program,"... Then we'll have to agree to disagree. There are people building, deploying and selling applications using AI who aren't doing anything close to what I would consider "programming". This is so far beyond the comparison to an IDE or WYSIWYG editor. > Ther…

> Why does your employer need to be willing to pay thousands per month for AI tools if you don't need AI to do your job? Can't you just tell your employer you don't need AI? If you use 0 tokens, don't they pay for 0 tokens? Or do you have an employer who is forcing you to use AI? How are you using it if you don't need it?

An accountant could do his job without Excel, a developer when AI didn't exist could do his job without IntelliJ, a carpenter can use hand tools, and on and on and on. I don't really understand what you think you're revealing with this line of questioning. I can do the job more productively with the tools so they pay for the tools. If we're making appeals to rational behavior on employers' part, they did hire me, and at prevailing SWE wages, to do it, rather than getting someone who doesn't know how to program in any traditional sense, and then immediately encourage my use of the tools.

Re: Domain expertise has always been the real moat

#578

Earlier quoted context omitted.

You are wrong. Writing programs (atleast efficiently) is not the easiest part. It just appears to be once you get accustomed. Domain expertise is hard but not that hard compared to the insane mental discipline required to write efficient scalable code. Domain expertise is valuable and hard but I don't get this "domain expertise is harder than disciplined coding" mentality.

Seriously doubt you ever did serious financial math. Coding is way easier.

Lol truly. A short list of just a few things that haunt my dreams: Distributions, American/European waterfalls, Carried interest, Management fees, and basically every single detail about Master feeder fund economics

Re: Domain expertise has always been the real moat

#579

Earlier quoted context omitted.

You are wrong. Writing programs (atleast efficiently) is not the easiest part. It just appears to be once you get accustomed. Domain expertise is hard but not that hard compared to the insane mental discipline required to write efficient scalable code. Domain expertise is valuable and hard but I don't get this "domain expertise is harder than disciplined coding" mentality.

Seriously doubt you ever did serious financial math. Coding is way easier.

Seriously doubt your reading skills. Read again. Hint "efficient" and "scalable". Go figure. I seriously doubt you have written scalable code. Financial math is laughable by comparison.

Re: Domain expertise has always been the real moat

#580
post #578

Earlier quoted context omitted.

Seriously doubt you ever did serious financial math. Coding is way easier.

Lol truly. A short list of just a few things that haunt my dreams: Distributions, American/European waterfalls, Carried interest, Management fees, and basically every single detail about Master feeder fund economics

Funny. Sure. Kernel development and truly scalable architectures must come easy for you. Management fees must be so hard. :)
Post reply on HN