[tdesilva]

Good, coding harnesses should be open source and LLMs should be treated as commodities. Minimize switching costs for consumers, and let people understand how they're interacting with the context and the LLM outputs.

The industry has been moving the wrong direction with Claude Code staying closed (despite multiple times leaking the source code!) and the open source Gemini CLI being deprecated in favor of closed source Antigravity CLI.

[tw04]

Why would a company do any of these things? What is their motivation for any of it? That’s like saying cloud providers should be commodity and should open source all of their platforms and eliminate egress fees so customers can easily leave at any point in time.

That’s a charity, not a business model.

[ignoramous]

> That’s a charity, not a business model.

  Joel Spolsky in 2002 identified a major pattern in technology business & economics: The pattern of "commoditizing your complement", an alternative to vertical integration, where companies seek to secure a choke point or quasi-monopoly in products composed of many necessary & sufficient layers by dominating one layer while fostering so much competition in another layer above or below its layer that no competing monopolist can emerge, prices are driven down to marginal costs elsewhere in the stack, total price drops & increases demand, and the majority of the consumer surplus of the final product can be diverted to the quasi-monopolist.

  No matter how valuable the original may be and how much one could charge for it, it can be more valuable to make it free if it increases profits elsewhere.

  This pattern explains many otherwise odd or apparently self-sabotaging ventures by large tech companies into apparently irrelevant fields, such as the high rate of releasing open-source contributions by many Internet companies or the intrusion of advertising companies into smartphone manufacturing & web browser development & statistical software & fiber-optic networks & municipal WiFi & radio spectrum auctions & DNS: they are pre-emptive attempts to commodify another company elsewhere in the stack, or defenses against it being done to them.

https://gwern.net/complement

[kristianp]

Here's the Spolsky article, called "Strategy Letter V": https://www.joelonsoftware.com/2002/06/12/strategy-letter-v/

Here's links to the whole series up to VI: https://lettersremain.com/joel-spolsky-on-strategy/

[idle_zealot]

"Commoditizing your complement" is a convenient market phenomenon when it happens, but it's not a hard rule that everything that should be a commodity lands in the window of incentives that result in this strategy being the self-serving optimum for a company. Usage of "should" here can be interpreted as "would be most beneficial to the largest number of people in terms of cumulative productivity" or "not having this be commoditized leads to terrible incentives that harm a lot of people." There's no guarantee at all that it ends up being most profitable to individual market actors to attempt to commodify LLM compute, and in fact they seem more interested in securing legal moats and protected monopolies. You may think that reality will force this to fail via open source model releases from China or the like, but consider the world in which the US officially bans import and usage of all but specifically licensed and approved LLMs. The most powerful companies and people here seem broadly aligned on wanting that, so it's not implausible. Sure, you as an individual could illegally import and run an unsanctioned LLM yourself, but most people won't do that, and attempts to scale it would be beaten down with the force of law. The companies who strike deals will get 99+% of business and profit handsomely, no need to ever let their golden goose be commodified. It's a strictly better strategy if your government is dysfunctional enough to let it happen.

[bigbadfeline]

> but consider the world in which the US officially bans import and usage of all but specifically licensed and approved LLMs.

Then the US will fall behind China and, more importantly, from the rest of the countries in the world who refuse to lobotomize themselves in the manner you described.

There are also other cultural and political risks which, although delayed, may easily grow into something apocalyptic. Rich people building bunkers isn't just a fad, the outcome they envision isn't fiction, it might actually be a part of the plan, and while this is concerning it does make everything simpler.

[vovavili]

I don't really see how this applies to Anthropic. Claude Code _is_ their integral money maker. Their point above is a three-way neck-to-neck competition where they won't be able to become a near-monopoly any time soon.

[cj]

> it can be more valuable to make it free if it increases profits elsewhere.

That's a big "if" for Claude Code, et al.

[mpyne]

Is it? Seems from the outside that it was the thing that finally drove Anthropic into profitability.

[cj]

What drove them to profitability?

[keerthiko]

Anthropic was making moderate revenue from their API tokens, but everything changed when they started bundling crazy amounts of usage credits at $20/$100/$200 claude max subscriptions relative to the API token rate, but only as long as you use their coding harness — despite claude code, back when this started at least, being an inferior coding harness compared to many open source ones.

CC is technically a free product (you can use it with any model). It's got very few popular "opinions" on how a coding harness should work, but it's by far the most popular one (including at competing LLM manufactories like OpenAI and Meta and Google). Why? If it was just that their models are 5% better, most workplaces would optimize on token (aka cost) efficiency.

Anthropic has been winning the usage of their harness, their tokens, while earning significant revenue, by significantly subsiding their token consumption.

This has earned them many things:

- prime data on how software development — simultaneously the leading beneficiary industry from LLM use, and also the most flush with cash to spend — has been using LLMs

- bringing that industry to standardize around their harness concepts. They are essentially establishing themselves as the W3C of LLM interfacing, except as a private organization.

- all dat data

[fnordpiglet]

On the question of language models and periphery tooling -

Open weight models are disruptive to the business models of closed model businesses. An incentive is if your business is built around X but model training is helpful to you, but you don’t expect to meter it specifically. You can release your models and undercut the exclusive moat of a new model company like OpenAI or Anthropic from becoming at some point a competitor, or holding their access as a chip in pricing negotiations. By opening your architectures and weights other competitors can build on them and newer better models emerge faster decoupled from a small number of proprietary models. This lets you focus on X while gaining overall momentum on your model release at no additional cost and no loss in focus on X, while defending against upstarts and monopolies.

This is effectively a lot of the open source world that comes from corporate development as well. It feels odd after this many decades of discussing corporate reasons to participate in open source we keep rehashing it.

[c1sc0]

Maybe these things should be utilities that can be swapped out at will and shouldn’t even be privately owned at all? Heresy, I know!

[vovavili]

Having a five-year plan dictating which publicly owned b2b SaaS AI service you will be provisioned sure sounds like a dream of mine. I wonder what could go wrong.

[amunozo]

Open source is as opposite as state ownership as it is to private ownership.

[ecshafer]

Open Source comes out of Anarchist philosophy not Communist State owned type ideas. There is no direct ownership, its a project built as a common good, and owned by "the public". Its a community garden that people are free to plant and harvest as they see fit.

[vovavili]

The concept of a common-pool resource has nothing to do with awkward radical political ideologies.

[ecshafer]

I am the furthest thing from an Anarchist, but to deny that Anarchism influenced the ideas of Open Source Software is factually wrong.

[zrn900]

The concept of common-pool resource is the basis of the commons, the default state of mankind since its inception to recent times, and the philosophy of commons is called COMMUNism for a reason. In English, it should have been translated more correctly as COMMONism, from the COMMONS.

[zrn900]

Correct. It is therefore COMMUNist with the correct term. That's what communism is - the philosophy of the commons. Except note that the running of that commons, which everyone uses, requires socialist practices (what you mistakenly call communist).

[impulser_]

Because there is literally nothing special about coding hardnesses. The models are doing all the lifting. It just user experience that separates them.

A coding hardness with just bash outperforms Codex, Claude Code, OpenCode, Pi ect. The added features are just user experience features.

[Supermancho]

If harnesses are basically doing nothing, why would these metrics vary so widely?

https://www.endorlabs.com/research/ai-code-security-benchmar...

There's a lot of ways to configure agents and any implicit configuration to harnesses may have a non-trivial effect.

[impulser_]

It's because they do things that is why they score differently. Coding hardness add features for user experience not for agent efficiency. If they did all the coding hardnesses would be using bash and code mode and letting the agents write code to perform tasks but this doesn't work because you want humans in the loop. You want users to be able to approve and deny writes. You want uses to see edits. So you have to build tool for these. It's hard to show diffs when the agent is just using bash.

[Supermancho]

> The added features are just user experience features.

> It's because they do things that is why they score differently.

That was my point. Regardless of how you feel about UX, it's a value added set of features. The question initially posited, stands. Why would a company do any of these things?

> Coding hardness add features for user experience not for agent efficiency.

Pretending it was always about some metric you just decided was important is moving the goalpost. It's not compelling.

I think it makes more sense that it's Freemium Dominance or they act as Low-Cost Marketing tools.

[avadodin]

A harness(notice the lack of a 'd') is a strap system to gain control over something.

Like the thing people attach a dog lead to so that their kids won't just go kamikaze into a car.

Coding harnesses are named by analogy to that.

They are not hard.

[TurdF3rguson]

The reason I have a dog harness is to distributes weight so I don't choke her when she goes at the other dog that she doesn't like. I'm actually puzzling over kids kamikazeing into cars

[imp0cat]

It's actually only a problem if it's the other way around, isn't it?

If kids run into a car, they will most probably just bounce and continue, perhaps inflicting some minor damage. But if a car mows down a kid, that could well be a fatal injury. Leashes for all the cars! ;)

[cassianoleal]

A kid running into the side of a car that’s moving over a certain speed will be lucky to break only a few bones.

[avadodin]

It is a common fear for parents. Obviously they are not fighting for the emperor but chasing or running away from something.

The strapped kids are often normal with no apparent disabilities(but it is possible they have an ADHD diagnosis).

Never thought about doing it to my own.

[impulser_]

You got to miss spell these days or people assume your ai :)

[IncRnd]

That's very punnyy

Its like yuo're on fire!

[vidarh]

Try Kimi in Kimi CLI and Claude Code and try saying that again. Kimi quickly collapses into tool calling loops without measures in their CLI but not in Claude Code and is largely useless for any long running tasks in harnesses not taking this into account.

With those measures (which are actually quite interesting) it can at times perform at Sonnet level.

[cookiengineer]

I would disagree here.

Building a good and working coding harness with smaller models is really hard. Everything evolves around the limited context size.

Tools must be specification driven to reduce noise and high temp hallucinations, tool call shrinking needs to remove errors and tryouts of different formats of parameters (because LLMs always ignore descriptions in the JSON...), and you have to deal with long running agents because you can't afford them. Planner/orchestrator architecture, agent to agent communication need to be summarized, and then you have the messed up scheduling parts, because you need to prioritize short running agents and give the planner a tool to wait for outputs of spawned contractor agents.

And that's not even talking about sandbox vs playground read/write/access policies of tools.

Harness engineering, if done correctly, is quite hard.

And all of this works 60% of the time, every time.

Anyways, that was somewhat the summary of the last 6 months building my exocomp agentic environment. And it's still not satisfying to work with.

[calgoo]

In my limited experience, the smaller the model, the bigger the harness. Where with something like claude or deepseek the context size etc just let's you give it bash access and step back; small models tends to do better with simple action - response , new context each call. Context management becomes a continuous activity. Its a fun space , and I have found big models decent at building and improving these harnesses for the small ones. Using /loop and just run a continuous test - build - test loop.

[selcuka]

Your reply doesn't answer the question: What is their motivation for any of it?

[devmor]

Public good isn’t a charity, and a business model that doesn’t contribute to the public good should not be allowed to exist.

[digitaltrees]

Is it public good? Or is that a coincidence covering the real motive of an attempt to undermine the viability of American companies?

[devmor]

What?

[digitaltrees]

China is giving away AI for free so it’s harder to make money. The same strategy they did with solar panels. Sell them at a loss long enough until the manufacturers go out of business and you’re the only one surviving. Then flip to extract monopoly profits

[thefounder]

What does it sound so familiar with some US business models? I mean all the big tech does exactly that plus Uber (if you call it a tech company)

[bigbadfeline]

> China is giving away AI for free so it’s harder to make money.

That might've crossed their minds but that wouldn't move their hand, not even a finger. Politics is the primary driver here, here's the deal:

AI is the new Internet

China foresaw a world where they'd be blocked from it and Anthropic's ongoing attempt to block their own country form it shows how right the Chinese were.

[randbyte]

Instead we got home solar system become very affordable over the past decade.

And driving out US manufacturers isn’t even the main goal for China. They know their huge risk on reliance on petroleum and was doing everything they can to mitigate that. Building out a huge solar manufacturing base is their answer. Now they are reducing petro imports YoY.

[TurdF3rguson]

Does that apply to US open weights models too? Was Llama an attempt by Meta to destroy America's economy?

[devmor]

Isn’t that the standard economic playbook for all well funded businesses at this point? Walmart, Uber, Amazon, etc

[sumedh]

Who defines what is public good?

[devmor]

Well, the public. It’s in the name.

[blahblaher]

Because they steal everything to train their models. They literally make you pay for the "commons" knowledge

[smgpie]

Yes I do think cloud providers should open source all of their platforms, and this is not charity because it is essentially the hosting that they are providing as a product. Even if, say, google open sources its whole search infrastructure, it does not at all means you can just host your own due to the huge hardware requirements, but you can know(especially after AI which can be utilized to do this) that they are not using your data in a way they shouldnt.

[Someone]

> because it is essentially the hosting that they are providing as a product

That would be true _if_ they were forced to open source all their code, but it isn’t today.

> Even if, say, google open sources its whole search infrastructure, it does not at all means you can just host your own due to the huge hardware requirements

You can’t but Acme Inc and other bigcorps could, and Google’s margins would evaporate overnight.

[smgpie]

Well, chromium the web engine is open source but it is still hard to craft a web engine with equal level of performance and other metrics, even with a team of talents.

Besides that, what you say may be true from the google's perspective. From a user(which I am) perspective, Google and Microsoft and other big corps have proven again and again with their actions that they cannot be trusted, just refer to https://www.gnu.org/proprietary/proprietary.html.

[pstuart]

Good will and trust can ultimately have monetary value, and having a funnel based on open source is a viable play if it leads to a service that is sticky.

[fnordpiglet]

Cloud providers are commodity, and egress pricing is partially cost following because they have to pay peering to their interconnect points for WAN. Internal networks are not charged within the account because the economics of the VPC overlay are optimized for that use, but inter account and VPC and other boundaries carry cost - especially interconnection between accounts because the way VPC treats virtualization requires a relatively expensive routing. Inter AZ and inter region pricing also exists for the same cost following reasons. They also help shape incentives because it allows them to optimize placement of compute within the same AZ to physical buildings or rings.

The case that is largely nonsense is the egress pricing on direct connects since beyond the circuit costs, which the customer pay, there’s no costs for aws not already on the customer regardless. It also makes DC friction weird in that you are incentivized to NOT move storage before compute.

[vidarh]

At least AWS egress pricing is a very high multiple of market rate.

[fnordpiglet]

I will say they do negotiate it if it’s excessive. I have negotiated flat rates on both public and DC egress due to scale and architecture and they were open to it.

[idiotsecant]

The capital motivation isn't the only one that exists. You can say something should be true without having a plan to maximize quarterly revenues.

Even if you consider profit motive, what is the profit motive for corporate contributions to open source? The same applies here.

[jchw]

Because Claude Code is literally nothing particularly special. We don't need their business model. They need their business model, and to that I say, tough shit.

[randbyte]

> cloud providers should be commodity

They are already.

> and should open source all of their platforms

Most of the cloud platforms are open source. Linux, container, k8s… it’s entirely possible for someone to build and deploy their private cloud if they have the resources.

> and eliminate egress fees

What does it mean? If I sign up for cloud service I am only bound to the contract terms. If I am PAYGO I can switch anytime.

[isityettime]

> Most of the cloud platforms are open source. Linux, container, k8s…

Linux isn't a cloud platform and neither is Docker.

Kubernetes was created specifically to create a way in against AWS' de facto public cloud monopoly. The Cloud Native Compute Foundation is a classic "alliance of smaller players uses open-source and interoperability as a wedge against an incumbent that threatens hegemony".

> it’s entirely possible for someone to build and deploy their private cloud if they have the resources.

What is there for that, really? Basically just OpenStack?

[randbyte]

I never say these pieces are the equivalent of a cloud platform. What I said is that with them, it is entirely possible for someone to build their own.

Why someone wants to do that is not my concern. But not being able to is.

[elisbce]

The familiar Chinese recipe for success: Always copy and imitate first, even if it is inferior, always make it cheaper or even free so that the original innovator will be burdened by brutal price competition and much bigger R&D costs and cannot keep up in the long run. Then the copycat will win in the endgame.

[thefounder]

There are very rarely original ideas or original products. It’s more about money put into that idea and the marketing. You could say that Apple copies and imitates just like the chinese. After all they wait for the others to pour money into the ideas and they just deliver a polished experience on top of the hard work others did. Yet people don’t accuse them of being imitators.I think people are mad at the Chinese because they do it for cheap and at an industrial scale on so many products/industries(i.e even food). To me they are a force for good competition.

[tw1984]

did you just time travelled from 2010?

show me from which country or company did the Chinese copied their EV techs, batteries, drones, robotics etc. for their military gears, show me show they copied their J-36 and J-50 fighter jets, Type-055 destroyers, YJ-21 missile and the most recent 09x sub.

time to stop living your socially isolated life and start reading stuff other than whatever fed to you by your brainwashing media.

[downrightmike]

They did steal all of our written knowledge

[femiagbabiaka]

The cloud provider isn't the harness, Terraform/OpenTofu/Pelumi and the abstractions you build using them are. The cloud provider is the LLM. It's not as fungible as the LLM and there's no direct comparison to egress costs of course, but that's moreso a problem with the metaphor.

[sally_glance]

Well, until you established monopoly you need to build trust. Open Source is one way of doing just that. One of the better ways I would say even...

[sharts]

Do you think Internet Explorer 6.0 was a good decision?

[digitaltrees]

There is a difference between a loss leader that drives a monopoly. Microsoft gave away a browser to make their monopoly OS more valuable and deepen their anti competitive moat. They were almost broken up as a result and honestly should have been if we had any anti trust enforcement.

How does open sourcing a Claude code clone drive adoption of anything that is a monopoly or even commercially related? Instead it seems like an attempt to undermine US AI companies.

That being said I am increasingly skeptical of how the US leaders are converging on creating monopolies and going deeper into the app layer that means they will end up owning everything rather than being the substrate of a competitive and flourishing ecosystem.

If it were up to me I would implement a regulation that 1) AI labs can’t own inference hardware, data centers would be a regulated utility like electricity and the internet required to provide open access third party safety, guardrails and audit, 2) inference providers can’t build apps beyond serving API requests 3) training data sets are required to be open sourced within 3 years of training a model.

What we are doing now is allowing vertical and horizontal integration of the hardware, training, inference and application layer. Last time we did that standard oil ended up owning the rail network, pipelines, oil fields, gas stations and refineries. Go see how that worked out for society.

[bellowsgulch]

After 16 years on Hacker News, I've come to associate its readership with cheap bastards who think everything should be free while simultaneously wanting to keep their 6-figure jobs.

There's a very strong overlap with male gamers, who also think everything involving sophisticated engineering and design should be cheaper than a cup of coffee.

Just call it out and maybe we can collectively choose to towards a culture that doesn't encourage such shameless behavior or perverted values.

[bigbadfeline]

> on Hacker News, I've come to associate its readership with cheap bastards

They might be cheap but they surely aren't stupid. They sure know who not to trust, security beats imaginary profits promised by cheap used-car salesmen who did move software jobs to wherever they could spare a dime.

And you want us to defend their current profits which they swear to use for the removal of the remaining software jobs...

[sjbzbeiks]

You’re not totally wrong but I think the motivations are varied. Using myself as an example, I have paid huge sums for software, films, news, and yes open source software, but I hate paying 5 bucks for a sub or buying certain products because the deal just isn’t a fair trade. I’d say most modern software is a bad deal.

That said I am a cheap bastard.

[randbyte]

If you see this as people just want free stuffs you are missing the point. There is an entire different level of danger to allow a few entities to completely own a day to day technology.

Imagine personal computers and open source operating systems completely don’t exist. What you can do on a pc today, you have to always do it on a thin client always connected to a main frame run by a selected few. Everything you do is recorded and subjected to at least 30 days of retention and inspection.

Imagine every car is completely not customer serviceable and have to be connected to one of the three manufacturers in order to operate. Everywhere you go is recorded. The manufacturer may decided the place you want to go is inappropriate for you to go.

[vovavili]

I'd gladly pay for software that demonstrates craftsmanship, quality and care for the user. Not a very common phenomenon unfortunately.

[galaxyLogic]

Note it is easy to confuse "free" with "paid by advertisements".

[arizen]

opensourcing software may enable leverage of wider network of contributors to given piece of software,hence software can evolve much more quickly and efficiently.

[tw1984]

there is no proof of that whatsoever.

the internet is full of semi skilled "developers" who should never be developers in the first place.

[jrflowers]

“A business that does things that customers actually like is a charity” lmao

[jnovek]

> Why would a company do any of these things?

What? It’s actually insane that they haven’t yet.

I don’t like changing tools. What engineer does? I want to learn one tool and tune it to my exact preferences. Proprietary vendor tools are not portable and I avoid them.

Either Anthropic or OpenAI could drop the first-to-market open coding harness tomorrow and it would be as big as VSCode, it would be the standard platform everyone builds stuff on.

[sroussey]

MS releasing VS code led to Cursor reaping $60B in value.

[jnovek]

What do you mean by $60b in value? Revenue? Valuation? Cursor’s most recent raise was at a post valuation of $29b.

Also, leadership at MSFT has talked quite a bit about their strategy with VS Code and it’s exactly what I was describing above (which is why I used it as an example). They give it away for free so that they can control the ecosystem.

[sroussey]

SpaceX is buying it for $60B

[_HMCB_]

This ^

[tdesilva]

ah nevermind it's just a fork of OpenCode

[baq]

The jagged frontier of frontier models means treating tokens as fungible between providers is naive at the limit of capability, but also will work for solved problems far from the boundary. The problem is you need to keep evaluating all models to know where your use case lies on the frontier map.

As a concrete example, you’ll get very different results for the same prompt for sonnet, opus, fable, gemini, gpt 5.5, …

[bellowsgulch]

The complements ARE the LLM AND the harness. The actual products we're all consuming are GPUs. Memory being expensive is a second-order effect.

The platform is the GPU, and doing cool shit with it IS the complement, which requires more memory. And demand is so high and will stay high, that it looks like the platform itself.

[bigbadfeline]

> And demand is so high and will stay high,

The question is why supply is restricted, primarily by sanctions and tariffs to China, and the expressed refusal of RAM makers to even think about increasing supply, they are actually all sweaty about China taking a bit of the unrestricted market.

[up2isomorphism]

There is really nothing free in terms of money, there are only things really free in term of spirit. But AI coding assistant are not those things related to spiritual freedom.

[dominotw]

what do you mean by should?

[miroljub]

As much as their propaganda wants us to believe, Anthropic is not 'the industry'.