AI should only run as fast as we can catch up

I have seen so many projects were people who understood all of it, are just gone. They moved, did something else etc.

As soon as this happens, you no longer have anyone 'getting it'. You have to handle so many people adding/changing very thin lines across all components and you can only hope that the original people had enough foresight adding enough unit tests for core decisions.

So i really don't mind AI here anymore.

One that works particularly well in my case is test-driven development followed by pair programming:

• “given this spec/context/goal/… make test XYZ pass”

• “now that we have a draft solution, is it in the right component? is it efficient? well documented? any corner cases?…”

All the type systems (and model-checkers) for Rust, Ada, OCaml, Haskell, TypeScript, Python, C#, Java, ... are based on such research, and these are all rather weak in comparison to what research has created in the last ~30 years (see Rocq, Idris, Lean).

This goes beyond that, as some of these mechanisms have been applied to mathematics, but also to some aspects of finance and law (I know of at least mechanisms to prove formally implementations of banking contracts and tax management).

So there is lots to do in the domain. Sadly, as every branch of CS other than AI (and in fact pretty much every branch of science other than AI), this branch of computer science is underfunded. But that can change!

  public static double ScoreItem(Span<byte> candidate, Span<byte> target)
  {
     //TODO: Return the normalized Levenshtein distance between the 2 byte sequences.
     //... any additional edge cases here ...
  }

When I give AI a smaller or more focused project, it's magical. I've been using Claude Code to write code for ESP32 projects and it's really impressive. OTOH, it failed to tell me about a standard device driver I could be using instead of a community device driver I found. I think any human who works on ESP-IDF projects would have pointed that out.

AI's failings are always a little weird.

* My 5 years old project: monorepo with backend, 2 front-ends and 2 libraries

* 10+ years old company project: about 20 various packages in monorepo

In both cases I successfully give Claude Code or OpenCode instructions either at package level or monorepo level. Usually I prefer package level.

E.g. just now I gave instructions in my personal project: "Invoice styles in /app/settings/invoice should be localized". It figured out that unlocalized strings comes from library package, added strings to the code and messages files (added missing translations), however has not cleaned up hardcoded strings from library. As I know code I have written extra prompt "Maybe INVOICE_STYLE_CONFIGS can be cleaned-up in such case" and it cleaned-up what I have expected, ran tests and linting.

I've tried vibe coding and usually end up with something subtly or horribly broken, with excessive levels of complexity. Once it digs itself a hole, it's very difficult to extricate it even with explicit instruction.

Another good use case is to use it for knowledge searching within a codebase. I find that to be incredibly useful without much context "engineering"

Also - claude (~the best coding agent currently imo) will make mistakes, sometimes many of them - tell it to test the code it writes and make sure it's working - I've generally found its pretty good at debugging/testing and fixing it's own mistakes.

Instead of dealing with intricacies of directly writing the code, I explain the AI what are we trying to achieve next and what approach I prefer. This way I am still on top of it, I am able to understand the quality of the code it generated and I’m the one who integrates everything.

So far I found the tools that are supposed to be able to edit the whole codebase at once be useless. I instantly loose perspective when the AI IDE fiddles with multiple code blocks and does some magic. The chatbot interface is superior for me as the control stays with me and I still follow the code writing step by step.

I'm in a similar situation, and for the first time ever I'm actually considering if a rewrite to microservices would make sense, with a microservice being something small enough an AI could actually deal with - and maybe even build largely on its own.

You can start there. Does it ever stay that way?

> I work on a large product with two decades of accumulated legacy

Survey says: No.

That’s the typical “claude code writes all my code” setup. That’s my setup.

This does require you to fit your problem to the solution. But when you do, the results are tremendous.

Definitely. I've found Claude at least isn't so good at working in large existing projects, but great at greenfielding.

Most of my use these days is having it write specific functions and tests for them, which in fairness, saves me a ton of time.

Now I use agentic coding a lot with maybe 80-90% success rate.

I’m on greenfield projects (my startup) and maintaining strict Md files with architecture decisions and examples helps a lot.

I barely write code anymore, and mostly code review and maintain the documentation.

In existing codebases pre-ai I think it’s near impossible because I’ve never worked anywhere that maintained documentation. It was always a chore.

This is not the case for most monoliths, unless they are structured into LLM-friendly components that resemble patterns the models have seen millions of times in their training data, such as React components.

And I think it's less about non-deterministic code (the code is actually still deterministic) but more about this new-fangled tool out there that finally allows non-coders to generate something that looks like it works. And in many cases it does.

Like a movie set. Viewed from the right angle it looks just right. Peek behind the curtain and it's all wood, thinly painted, and it's usually easier to rebuild from scratch than to add a layer on top.

I'm sure lots of people will reply to you stating the opposite, but for what it's worth, I agree. I am not a visual artist... well, not any more, I was really into it as a kid and had it beaten out of me by terrible art teachers, but I digress... I am creative (music), and have a semblance of understanding of the creative process.

I ran a SaaS company for 20 years and would be constantly amazed at how bad the choices of software engineers would be when it came to visual design. I could never quite understand whether they just didn't care or just couldn't see. I always believed (hoped) it was the latter. Even when I explained basic concepts like consistent borders, grid systems, consistent fonts and font-sizing, less visual clutter, etc. they would still make the same mistakes over and over.

To the trained eye they immediately see it and see what's right and what's wrong. And that's why we still need experts. It doesn't matter what is being generated, if you don't have expertise to know whether it's good or not, the chances are glaring errors will be missed (in code and in visual design)

But i'm a software engineere by trade and I do not struggle with telling you that this thing has to move left for reason xy, i would struggle with random tools capable of doing that particular thing for me.

And it does not matter here how i did it if the result is the same result.

In Software Engineering this is just not always the case. Because often enough you would need to verify that what you get is the thing you expect (did the report actually take the right numbers) or Security. Security is the biggest risk to all ai coding out there. Security is already so hard because people don't see it, they ignore it because they don't know.

You have so many non functional requirements in software which just don't exist in art. If i need that image, thats it. Most complex thing here? Perhaps color calibration and color profiles. Resolution.

If we talk about 3D it gets again a little bit more complicated because now we talk the right 3d model, right way to rig, etc.

Also if someone says "i need a picture for x" and is happy about it, the risk is less customers. But if someone needs a new feature and tomorrow all your customer data are exposed or the companies product stops working because of a basic bug, the company might be gone a week later.

Before mechanisation, like 50x more people worked in the agricultural sector, compared to today. So tractors certainly left without work a huge number of people. Our society adapted to this change and sucked these people into industrial sector.

If LLM would work like a tractor, it would force 49 out of 50 programmers (or, more generically, blue-collar workers) to left their industry. Is there a place for them to work instead? I don't know.

For example, Inkscape has this and it is easy to use.

https://www.deere.com/en/autonomous/

You've described AI hype bros in a nutshell, I think.

And I have on occasion found it useful.

If you can make as a rule "no AI for tests", then you can simply make the rule "no AI" or just learn to cope with it.

this is something one can look in further. it is really probabilistic checkable proofs underneath, and we are naturally looking for places where it needs to look right, and use that as a basis of assuming the work is done right.

I still find it sad when people use it for prose though.

I've heard the same claim every year since GPT-3.

It's still just as irrational as it was then.

Bold claim. They said the same thing at the start of this year.

Because you asked the wrong question. The most likely question would be "How do I make a quadrillion dollars and humiliate my super rich peers?".

But realistically, it gave you an answer according to its capacity. A real super intelligent AI, and I mean oh-god-we-are-but-insects-in-its-shadow super intelligence, would give you a roadmap and blueprint, and it would take account for our deep-rooted human flaws, so no one reading it seriously could dismiss it as superficial. in fact, anyone world elite reading it would see it as a chance to humiliate their world elite peers and get all the glory for themselves.

You know how adults can fool little children to do what they don't want to? We would be the toddlers in that scenario. I hope this hypothetical AI has humans in high regard, because that would be the only thing saving us from ourselves.

[0] https://newint.org/features/2018/09/18/10-steps-world-peace

If you are looking for a vision of general AI that confirms a Hobbsian worldview, you might enjoy Lars Doucet's short story, _Four Magic Words_.

[1] https://www.fortressofdoors.com/four-magic-words/

I would kind of feel sorry for a super-intelligent AI having to deal with humans who have their fingers on on/off switch. It would be a very frustrating existence.

A super intelligent ai would have agency, and when incentives are not aligned would be adversarial.

In the caricature scenario, we'd ask, "super ai, how to achieve world peace?" It would answer the same way, but then solve it in a non-human centric approach: reducing humanities autonomy over the world.

Fixed: anthropogenic climate change resolved, inequality and discrimination reduced (by reducing population by 90%, and putting the rest in virtual reality)

It's just fanfiction. They're just making up stories in their heads based on blending sci-fi they've read or watched in the past. There's no theory of power, there's no understanding of history or even the present, it's just a bad Star Trek episode.

"Intelligence" itself isn't even a precise concept. The idea that a "superintelligent" AI is intrinsically going to be obsessed with juvenile power fantasies is just silly. An AI doesn't want to enslave the world, run dictatorial experiments born of childhood frustrations and get all the girls. It doesn't want anything. It's purposeless. Its intelligence won't even be recognized as intelligence if its suggestions aren't pleasing to the powerful. They'll keep tweaking it to keep it precisely as dumb as they themselves are.