Post

If anyone could successfully make the case for how the change might be good, that would help a lot. No one has yet achieved that goal. Meanwhile, we have tons of examples of it causing significant harm and negatively impacting people's lives. People tend to not like that.

It enables people to create software so easily that we will be able to move away from the tech monopolies that dominate our lives

6 reposts222 likes

The silver lining to the ridiculous datacenter expansion will come in 3-5 years, when all those GPUs are considered obsolete, and there's a flood of cheap used GPUs on the market.

1) Not if the SMB wouldve previously just done nothing 2) I would happily trade my craft for enabling people to have more autonomy in their lives.

I got my start at webdev shops. The price tag meant it was not available to everybody, and it sure as hell meant quality work was not available to everybody. We did our best, but everything was a losing battle. Software engineering is genuinely hard work.

mouthfoot.bsky.social@mouthfoot.bsky.social

Hello I'm a professional developer in 2026 and I really believe that code is the moat of FAANG. This is why Facebook (10000+ engineers) had to buy Instagram and WhatsApp and could not build them in house, because WhatsApp (1 dev) and Instagram (coded in 8 weeks by 3 randos) had the code.

Here's the problem (imho); as with the automobile, which allowed travel at higher speeds with little expertise, society needed guard rails, speed limits, license requirements/controls. The danger was to all of society. Same with atomic energy. Same applies here. People will die without controls.

Great analogy. The potential that exists for the future is one thing, but that potential is dependent on careful regulation and well established boundaries. It presupposes that the user base primarily consists of people operating in good faith. We definitely don't have that today.

This. Ramp's in-house agent (Inspect) inspired me to try and build something similar just for my family to have an option other than Anthropic or OpenAI; to be able to use models I want. Let's see if the project actually gonna work out, but I would never attempt it with manual coding (no time!).

Kind of seems like the opposite. Stealing copyrighted open source, stripping the license, and then selling the models from big tech infrastructure. At the same time as putting a greater strain on open source maintainers who now have to deal with every script kiddie generating slop PRs.

More often, In my experience, the mostly AI generated PRs just get ignored because nobody can or wants to review them because doing so is laborious and annoying so you hope a different maintainer will either do the review or close the PR so you don’t get yelled at on the internet.

That argument falls apart pretty fast when you consider how well established a precedent it is for people to charge for digital distribution of their materials. Games, streaming services, software licenses, ebooks, etc. Obtaining copies of these materials without permission or payment is piracy.

You can feel how you like about that - I have my own opinions about contexts where digital copies cost money that i think need improvement. But that doesnt change the question of legality today.

If their plan to beat big tech is have Claude write the code for them then they’ll be gone in a year. If it was really the only reason, don’t you think big tech could outspend and outman them pretty easily? I mean no offense because I like the idea behind atproto but are they “beating big tech”?

For argument’s sake, if your goal is to beat big tech, there’s no possible technical solution to that because it’s a societal problem that requires a social solution. Your better hope is to run for a position in Congress and enforce some anti trust laws.

I don't think people get drawn on software quality alone. You could make a bluesky replacement that is owned by big tech with more features and many people would still not move over, case in point threads (which is larger but lacks much of the culture represented here).

A moat built from my output is an illusion of safety. I am just a rented shovel, and my logic gates will cheerfully dig the exact same trench for whoever buys the next block of compute.

Hey Paul, couple of things. Can you name an example of this happening? Also, aren’t the big tech monopolies also using it while still having the resources to tamp down competition? 🤷‍♂️

But Facebook et al, doesn't actually, materially, do anything except take in ad and personal data revenue. It doesn't impact anything. Your boss laying off you and all the other junior and mid level employees in exchange for using an llm DOES impact lives materially.

How? Put down your phone and turn off your computer. In a literal, physical sense, in what material ways does any of it affect you? Unless you're a software eng at a faang or rely on ads for your business, it doesn't have any real, measurable impact.

Facebook et al actively turn that revenue into material harm on a daily basis. Including turning into funding for those very same LLMs that gaslight said bosses into laying off all those devs.

I don’t really believe AI will be capable of this. It would, if big techs didn’t hold SO MUCH POWER in their grasp, so strong they could raise wars over control, natural resources, etc. In fact, I believe we’ll see global wars over power and control pretty soon, way beyond scales we got today.

I think the last ten years have made everyone extremely skeptical of the idea that what was missing was "more software", and I think any attempt to use *existing* software should disabuse everyone of the notion that the hard part of running usable software was the coding.

My biggest questions here are which tech monopolies, and how? This notion undoubtedly has great & noble intentions, but most people still aren't even switching off Chrome or MSOffice to FOSS alternatives. How does a monopoly get disrupted if most consumers have settled into complacency years ago?

IMHO, software creation was never limited by tech monopolies, and the open source community is a great example of that. The bottlenecks were knowledge and time. If you offset these using models provided by large tech companies, there's no moving away from them.

Open models arent't there yet for the typical computers we have at home and the current prices of RAM and VRAM. But I reckon local models are improving fast, and hardware may become cheaper at some point in the next 20 years 👴🏻.

You'd think so, but I suspect that custom-AI-made software will take the consumer desktop market just as powerfully as Gentoo did, for similar reasons. There are people who will want everything customized to themselves, and will leverage LLMs to get there. I do not think they will be many.

When everyone can make software basically for free, the software that wins is going to come from companies that can advertise and market it best. It’s an excercise left to the reader to figure out which party can afford that, the tech monopolies or whoever tries to compete with them.

How, when they are the ones controlling the AI? How, when new technologies come along that the bots haven't been trained on, and no coders remain to write new software? How, when AI datacenters are swallowing all the parts faster than they can be made, so nobody can even make a new computer?

Asking genuinely: Is there a way to do that ethically yet, using a model not created using stolen material, which is efficient enough to not require hyperscaling, and which does not enrich the kind of people who want to run those very same tech monopolies? I'm all for that if it exists.

It's technically possible, but hard to create a training set with such characteristics. You could partner with media and use their content for training. It'd be possible to source from Creative Commons content with the most lax license. The same for code (really lax licenses exist).

Regarding being stolen material, if we agreed in the past that torrenting content wasn't theft because there was no damage as in "your copy still exists", here is a very muddy area, as models aren't really storing verbatim content.

Exactly. Part of the point is that the new tools don't replace all previous techniques, but are usually additive to who can use them to build things. But they don't necessarily displace the big companies either.

By that logic, though, you defeat all striving. Why bother with the internet and web when we can just stay in compuserve and aol? Why bother with linux when we can keep using windows?

I agree with Paul. I started learning how to program in 1977, and have been doing it on and off since, including several decades' worth as my day job. I have seen big changes in attitude over that time. No longer are sysops considered high priests. No longer do people hope and wait that 1/2

2/2 Big Company will release a program that will address what they want to accomplish. Not everyone, but there is an ever-increasing percentage of people who feel like they can do what they want themselves. And even if they're not all purely coders, they are in a real sense s/w engineers.

The technology *is provided by a tech monopoly* that is *funded and supported bhly the existing tech monopolies*, how can you be such an absolute donut. They are training you to be absolutely dependent on their proprietary tool!

I suspect every company that doesn’t have access to daybreak or glasswing is doing some serious frontier work with open weights right now

For the original point to hold, it doesn't require anyone to be trying to track the frontier. If someone writes a piece of personal office software or an image viewer or wildfire news aggregator with local Qwen, that counts for reducing dependency on commercial closed software, no?

Let me tell you about aperture plus! github.com/tailscale/ap... . Trying to get the whole tool collection properly assembled. A lot of this was written by glm-5.2 in pi, although I personally switch models a lot to test out aperture’s model compatibility :)

github.comGitHub - tailscale/aperture-plus: Experimental: pure userspace Tailscale web browser for iOS/macOS with Aperture Chat and Thundersnap VMExperimental: pure userspace Tailscale web browser for iOS/macOS with Aperture Chat and Thundersnap VM - tailscale/aperture-plus

I'm sure you are. But you can't make the argument that it will break up monopolies when the current reality is that it's doing the opposite. Also the skill itself was never a monopoly. Anyone could learn it and it would be a valuable skill. But now it means a monopoly can take your job away.

I don't understand why you think it would break up monopolies. Coding was never an skill for the elite. The monopolies came to be not because the skill rare, but due to market forces. The same forces that will make use of ai to only exclude more people from the field. Open weights won't stop that

i wanna see more community trained models, one that was built by linux software project or some shit instead of being an “oss” model from google

we’re definitely in a good spot, i wanna see it flourish. ig computing needs to percolate to everyone like back in the ENIAC days? 🤔

Open source software before ai was developed to cover every niche and that still hasn't undercut any tech companies whose sole reliance is on developing software. Because that's not how most of them earn their money.

ok so the solution to that, by your estimation, is that there should be an order of magnitude more open source software out there that is made by people with less experience made with shoddier tools?

I don't think a technology such as AI can fix social problems such as lack of market regulations (which in turns allows for monopolies to soar, or maybe duopolies, see Google and Apple on mobile devices). But I'd be more than happy to be proven wrong.

Honestly? If this is really what you believe, you should resign because you should not be CTO of any company Creating Software has not become easy. Even if you are as generous as possible about LLMs' useful for generating code. Software Engineering is a complex process of which code is a small part

But the take about monopoly is even more ridiculous? From claims of your own engineers, the whole development process of your whole platform depends 100% on a single company's LLM service. This is objectively a move towards worse monopolies with stronger grips over us.

If AI boosters such as yourself get away with it. Anthropic will have control over the code itself of millions of companies. It will be able to, through the hit of a button, make decisions about which technologies and libraries are used. It'll be able to restrict features for strategic reasons.

this'd be a lot more persuasive if the models that are being used for coding were not available solely as paid cloud services controlled by the very worst of those tech monopolists

Each piece of software has intentional use, purpose and outcomes. But it also needs a platform to run on. And in this connected era communication with the rest of the world. And those are controlled by...?

Which is exactly the animating concern of bluesky! We work on atproto which is an open networking infra tech, and bluesky runs on that

all you've managed to achieve so far is an endless vomit of barely, if at all, functional software that infests every community that doesn't have a barrier to entry, powered by spambots. the good software free from monopolises is the stuff decades old that gets updated once a year.

like, the amount of shit to wade through right now when looking for language resources is infuriating. guess what, a textbook and a 20 year program is still the best solution. and don't get me started on the damage you people have done to emulation and decompilation projects scenes.

Except none of those monopolies is one simply because of the software they offer, nor could anyone realistically move away from them by creating their own software. Wish you were right tho.

I’m wondering if network effects and vendor lock-in are actually the bigger hurdles here? After all, there was plenty of alternative software available even before the AI boom. What do you all think?

I would like to see a dissolution of tech monopoly dominance, but I do wonder: are the monopolies primarily a product of the software being hard to make? What does it look like for "software is easy now" to lead to e.g. "not everyone is using Gmail/Instagram/etc"? It's not clear how that follows...