I've found that on some projects maybe 70-80% of what can be done with Sonnet 4.6 in OpenCode can be done with a cheaper model like MiMo V2 Pro or similar. On others Sonnet completely outperforms. I'm not sure why. I only find Opus to be worth the extra cost maybe 5% of the time.
I also find OpenCode to be drastically better than Claude Code, to the extent that I'm buying OpenRouter API credits rather than Claude Max because Claude Code just isn't good enough.
I'm frankly amazed at what OpenCode can do with a few custom commands (just for common things like doing a quality review, etc.), and maybe an extra "agent" definition or two. For many projects even most of this isn't necessary. Often I just ask it to write an AGENTS.md that encapsulates a good development workflow, git branch/commit policy, testing and quality standards, and ROADMAP.md plus per milestone markdown files with phases and task tracking, and this is enough.
I'm somewhat interested in these more involved harnesses that automated or enforce more, but I don't know that they'd give me much that I don't have and I think they'd be tough to keep up with the state of the art compared to something less specific.
This code matters for exactly one reason: they’re playing stupid DRM games restricting what subscriber tokens can be used for to force you to use their front ends and harnesses or buy more expensive API credits.
Claude Code is strictly worse than e.g. OpenCode in my experience. Not much to see in the app’s code except how it authenticates itself…
Sure I try and use all my subscription allowance with CC on side tasks, etc. but I still end up burning a bunch of API tokens (via OpenRouter) for more serious work (even the UI and ability to quickly review what the agent has done/is doing is vastly inferior in CC).
What they have done is got me experimenting with cheaper models from other providers with those API credits.
I've not struggled to find the things I need at https://open-vsx.org (usually by searching directly within VSCodium), but then I only use it for editing things like markdown docs and presentations, LaTeX/Typst, rather than coding, which I prefer to do in a terminal and with a modal editor.
I like designing data, algorithms, and systems. I like picking the right tools for the job. I like making architectural and user interface (CLI, configuration format, GUI, whatever) decisions.
Actually typing code is pretty dull. To the extent that I rarely do it full time (basically only when prototyping or making very simple scripts etc.), even though I love making things.
So for me, personally, LLMs are great. I'm making more software (and hardware) than ever, mostly just to scratch an itch.
Those people that really love it should be fine. Hobbies aren't supposed to make you money anyway.
I don't have much interest in maintaining the existence of software development/engineering (or anything else) as a profession if it turns out it's not necessary. Not that I think that's really what's happening. Software engineering will continue as a profession. Many developers have been doing barely useful glue work (often as a result of bad/overcomplicated abstractions and tooling in the first place, IMO) and perhaps that won't be needed, but plenty more engineers will continue to design and build things just more effectively and with better tools.
The assembly line has been mass producing ready-made products for over 100 years and yet product quality, material stability, aesthetic trends, and function design still dominate the purchasing decisions of the general public.
Being tapped into fickle human preference and changing utility landscape will be necessary for a long time still. It may get faster and easier to build, but tastemakers and craftsmen still have heavy sway over markets than can mass-produce vanilla products.
I would generally put “stability” and “quality” as attributes of mass production far more than that of handmade things. Yes, an expert can make a quality product by hand, but MOST handmade things are far more likely to be shoddy.
The whole point of mass production was that suddenly you could make a million identical perfect products.
I think reducing what LLMs do to « typing » is misleading. If it was just typing, you could simply use speech-to-text. But LLMs do far more than that, they shape the code itself. And I think we lose something when we delegate that work to LLMs
We do lose something, but really I still see it as an extension of autocomplete.
I had some pieces of code I wrote I was quite proud of: well documented, clear code yet clever designs and algorithm.
But really what always mattered most to me was designing the solution. Then the coding part, even though I take some pride in the code I write, was most a mean to an end. Especially once I start having to add things like data validation and API layers, plotting analysis results and so many other things that are time consuming but easy and imho not very rewarding
To me its just a natural evolution of the search engine.
And now looking back its an obvious outcome. The search engine of the time was the best way to find websites. But those websites grew in quantity and volume over time with information.. that information as we know is/was a vital input for LLMs.
Agreed. Also, it takes time to understand a domain properly- so the innate slowness of coding helps with letting things “simmer” in the back of the mind.
It’s not that they replace the act of typing, so much as figuring out how to express the specific algorithm or data structure in a given programming language, typing that, debugging it, etc.
Once I can describe something well, that’s most of the interesting part (to me) done.
with an LLM, you can have an ill formed idea, and let the LLM mold it into a shape to your liking, without having the investment required to learn how to do molding first.
Not OP but I would certainly ban adding gambling "features" to other products or services. Either you can be a gambling or betting shop/platform (regulated and restricted to adults) or something else, but not both.
I agree with most of the article, right up to the point where the assumption is that AI will make things worse.
We have reached a point of complexity and short-termism where it's standard practice to shove a huge, barely tested mass of Python, JavaScript, shell scripts, and who knows what else inside a docker container and call it done. Complete with hundreds or thousands of intractable dependencies, many of which we know ~nothing about and thousands of lines of esoteric configurations for servers we have barely any hope of even getting to run optimally, let alone securely.
Most software has been awful for a while.
Already, with AI:
- We can build everything in a statically typed, analysable, and memory safe language like Rust[0], even bits that really have to interact with the shell or wider OS and would previously have been shell scripts
- We can insist on positively deranged amounts of testing, at every level
- We can easily cut the number of dependencies in our code by >50%, in many cases more like 90%
- We can do the refactor as soon as it becomes obvious that it would be a good idea
- We can implement quality of life admin and monitoring features that would otherwise sit on the backlog for eternity
- We can educate ourselves about what we've built[1] by asking questions about our codebase, build tools to understand the behaviour of our systems, etc.
So yes, I agree that "The Future Belongs to Whoever Understands What They Shipped", but unlike the author I am somewhat optimistic[2]. There is more opportunity than ever to build and understand extremely high quality software that does not accept technical debt, corner cutting to meet deadlines, or poor quality (in design or implementation), for those that engineers who are knowledgeable enough and willing to embrace the new tools.
And AI, and the tooling around it, is only getting better.
[0] or Go or even TypeScript, but there's precious little reason not to pick Rust for most use cases now
[1] of course we need to choose to, and many won't…
[2] of course, there'll also be near-infinite valueless slop and some people will get sucked into that, but this seems little different to regurgitated SEO spam, short form video, and all the non-AI enshittification we already put up with, and perhaps AI will help more of us do a better job of avoiding it
In the UK at least, banking apps are how you give people cash when you owe them for dinner, drinks, whatever. It's also needed to authorise online payments. And for travel, location services is often used by the better banks as an alternative to immediately blocking your card every time you go anywhere. Then there are account perks[0] like airport lounges, co-working spaces, exercise classes, etc. that all use the app for access.
It'd be more than just a bit inconvenient to lose all of these things…
Luckily, all of my personal and business banking apps work fine on Graphene. Even the apps for the crusty old "bricks and mortar" banks that I still have backup accounts with.
[0] As an aside, Revolut Ultra in the UK costs less than the FT Digital subscription it includes so if you're an FT subscriber, all the other stuff that comes with the account is cheaper than free.
> And for travel, location services is often used by the better banks as an alternative to immediately blocking your card every time you go anywhere.
I routinely use my (U.S.) credit card abroad and never had issues. I don't have any banking app on my phone, which runs Graphene, because I do not need access to banking on the go. Things such as airport lounges, co-working spaces, exercise classes, may be valuable, but none of these things is more important than my freedom. I do not tolerate the thought that some company may track which services I use and where I go without my consent. I therefore do not use proprietary apps on my phone. If something does not work in the browser (Vanadium), I will do without it.
Better for it to grow layers that are new and exciting, accessible only to the cultures that create them (and whatever comes after) and those who make the effort to continue learning
I also find OpenCode to be drastically better than Claude Code, to the extent that I'm buying OpenRouter API credits rather than Claude Max because Claude Code just isn't good enough.
I'm frankly amazed at what OpenCode can do with a few custom commands (just for common things like doing a quality review, etc.), and maybe an extra "agent" definition or two. For many projects even most of this isn't necessary. Often I just ask it to write an AGENTS.md that encapsulates a good development workflow, git branch/commit policy, testing and quality standards, and ROADMAP.md plus per milestone markdown files with phases and task tracking, and this is enough.
I'm somewhat interested in these more involved harnesses that automated or enforce more, but I don't know that they'd give me much that I don't have and I think they'd be tough to keep up with the state of the art compared to something less specific.
reply