Otherwise you still end up with balls of spaghetti where each change takes longer and longer to work into the codebase. With a bit of planning, you can keep tackling changes and they keep taking the same amount of time more or less.
Wouldn't count on it. There likely is a ceiling, we might or might not be near it.
These kinds of changes to the way teams are formed and managed, and how products are developed is what really fascinates me about the impact of coding agents.
Even if it works, for some definition of works, to dump a huge spec into a coding agent context window, I suspect this is where a lot of horror stories about agentic coding come from.
Teams, and the management structure above them, are going to have to get used to more collaborative and iterative product definition and development in order to extract the most value from coding agents.
I have code I wrote in 2006 which is still in daily use adding business value at my company. Why would it need replacing?
As is a lot of newer code. Depends on the threat model
> Another reason is that if you can find a better way to accomplish the goal, then you should eventually switch to it.
If it's financially better and the gains outweigh the risks, sure
> If your code from 2006 is already perfect and there is no room for improvement
Or more realistically if the cost of improving it would be higher than the business benefit gained from improving it
> which is that any code from 2006 that you’re running has survived the quality filter of time
Not really, it's terrible code. But it does the job, because it was written to solve a business problem, not written to make beautiful code.
Code has always been fungible, in the sense that it could be refactored.
> you should expect any code you write today to mostly be replaced in the near future
Why on earth do you think this? It does not follow from your other points. Just because you could turn your business into a rube goldberg code juggling machine, why would you?
I find that only works with topics I know very little about.
- tongue in cheek: in topics you know very little about, you don’t have the ability to tell whether the LLM output is trash or not
- very serious: LLMs allow you to create things you could not have created before due to not having the knowledge
Sort of like a Dunning-Kruger or Gell-Mann amnesia.LLMs are allowing people to create things with very little effort compared to before, and those creations are of questionable quality - indeterminable by the LLM-user due to a lack of experience.
I think the quality of the results vary with the rigor of the process. If you one-shot prompt a new service (even with Fable5) "build me an airtable competitor, no bugs", you will get some real code camp quality results, but a well defined set of prompts and a workflow that leverages multiple coding agents reviewing each others work, and implementing test cases at a relatively granular level yields radically different results.
Couple that with a rigorous design/spec session, and multiple checkpoints (say at release milestone) to monitor drift from the original design makes the process a bit more intentional and measurable, while still being far faster than a typical engineer can build.
It's like anything else with LLMs, if you treat it like a magic wand, you haven't really learned or built anything, but using it as a true assistant, it can be a powerful and flexible tool.
The places where you let an agent do its thing is bound by the (implicit) interface your prompt creates. In an ideal situation this interface is a simple isolated thing in your application.
If you just let it do its thing, expect a monolith. You shouldn't rely on it coming up with interfaces that mesh well with what you had in mind, and each of those interfaces represents a decision that it took, restricting future iterations of the code. So a monolith is not only what you can expect, but also the best you can hope for.
If you want to let it cook, create more places for code that does a thing and you don't care how.
The "objective" nature of code is revealing a flaw in the human division of labor.
The trouble starts when the requirements for your software are at odds with the requirements of your organisational structure.
With some up front care you can create an environment where they mostly do what you want when you let them rip without oversight. As the codebase grows it becomes less likely they'll see any one example, and more rarely expressed ones can get missed. And then codebase predating LLMs or designed without this consideration you just have to accept what you get or wrangle them manually.
I know why the US frontier labs push it of course, you can't push a trillion dollar plus idea if you also say it should be a cheap commodity and the best and latest are now the answer, a bazillion mid levels are.
But most work of value I do is with a swarm of "not the latest nor the brightest" models, with a smart one controlling the dance. Most decisions or review is better done through consensus. The myth that "if we keep going it will reach a point where it's always correct" seems so obvious to me, it doesn't work like that, for that matter we don't work like that.
This is not a jab at it. I work with it. I love it. But I think the story imperative are a mismatch for the actual needs of most companies, and it's hiding between the "frontier needs" of special research & co, or the governemental needs.
On the same line of thought: it's pretty obvious as great as these things are, no they're not going to "replace all employee", and no employee without them won't be able to compete either. They're a supercharger, a self multiplying smart assistant with capabilities in any field, and it feels so wrong to me that they've been presented in a complelty different way.
To my surprise, none of them came close to producing an acceptable PR. Quality varied, but even the very best results were missing significant pieces.
My takeaway is that the human in the loop is still massively important for many things. And that makes it really hard to come up with a useful benchmark.
The parallels with self-driving cars are interesting. We demand that agents don’t just produce better code than most humans (which the frontier models do), but that they perform much much much better.
Devs are the PMs structuring the sprints now cause they know when/where/how to pivot
See discourse wrt who Nvy should pick as ceo (if JH doesn't come back anyway); at least from what I seen the theme is if they are not an engineer who can keep up with tech and know why Nvidia needs to change course Nvidia is toast
Same with software; everywhere I worked PMs that were not engineers sucked to be at as the PMs Duning-Kruger called the shots not the demands of technical problems before the devs
You need not just the PMs but the management structure on top of them to get with the idea that product development should be iterative and experimental, not spec driven. But yeah that's how things are going.
All those layers appeared in response to ZIRP to fuel job creation. They did not appear out of real necessity and iteration as you call out. You're blithely rattling off a startup spec sheet.
You need that perhaps, or believe you do from extensive prior experience.
2010s are gone; public opinion on such an economy has effectively turned the corner for good.
Same way devs complain about others getting into their flow, computer users at large are tired of a dev riding shotgun, acting like a helicopter parent. They don't want a subscription to you.
A forced obligation of your having reach into their lives while they have no such leverage feels invasive af.