Everyone is already cloning the app layer. There are 30+ claude-code clone, and codex work already cloned cowork, grok and gemini is doing the same.
Curious why your first impulse is not simply to point your favorite agent at a few examples and start brainstorming/planning from there?
Multiple times I’ve built a purpose specific bespoke tool starting this way. In fact, it’s a great way to learn how specialized tools are built.
This preliminary change in cost may not make it worth it for people depending on which is their "favorite agent". Especially as the default harnesses for said agents continue to improve...
No, only Anthropic has that policy (and I think even that is relaxed for an unknown period if you use the Claude Agent SDK: https://support.claude.com/en/articles/15036540-use-the-clau...).
OpenAI, Kimi, Qwen, GLM and Deepseek all allow it.
I'm not sure about Gemini.
The opposite, how easy it is nowadays should be clarified. Codex's app-server quite literally is built to be integrated with, authentication is a breeze and it's quite stable, compared to the TUIs that are around.
It's a real shame Codex is moving in the direction of hiding stuff on local disks from users, hopefully they're revert the decision of encrypting the agent>sub-agent prompts so we get back introspection again...
The alternative of that first impulse should be what exactly? Telling AI to research Medium AI drivel articles (since nobody seems to be posting long/medium form content anywhere else these days) to figure out what’s the best approach to building an AI harness?
BTW just my personal observation but I generally find AI notoriously bad at building any AI applications compared to standard software (which is not surprising given the scarcity of high quality training data)
Ofc I can just fork gemini-cli if I want my own version but like I don't think that's what OP meant by build your own thing and customize it.
https://github.com/99991/MinimalAgent
You only need a single tool to start with. All recent LLMs know how to use bash for reading/writing/editing/executing.
And you add a small twist, that instead of a 1-to-1 back and forth, you instead put it into a loop, where the LLM reply can itself "have a turn", e.g. a tool invocation, where your system is the one that replies (e.g. with the tool invocation's result). That's pretty much it, you have a 1 to potentially many "chat".
The harder part is getting all the "soft" parts right, like how to have well-behaving tool calls, timeouts, prevent huge cycles eating up tokens, but there are no one way to solve these, it's a fundamentally heuristic-heavy area.
For example, the mobile/remote control feature I find very useful but not sure it would be worth cloning. If you aren’t replicating every feature you aren’t exactly “cloning” it, you’ll just end up with a crappier version with a fraction of the features of the real thing.
I found that most things that are “features” in harnesses and tools that don’t relate to UI and UX in those harnesses themselves can be done through skills or just prompts.
A nice features of Claude’s is the remote control through the mobile app. That isn’t just a skill.
We used to have a time where every small business had its own custom accounting software, and discovered that these businesses aren’t actually special snowflakes and are better off just using the generic software and conforming to whatever it does
I’m surprised an open source accounting package hasn’t popped up that handles the accounting well. The backend could largely be the same (or at least modular) while the front legitimately needs to vary based on what type of work, how sales are done, whether they hold inventory, local taxes, etc, etc.
only labs knows what the rl enviroment was so they will always be one step ahead.
They just need to be good enough and just copy.
Sorry american labs - it’s the beginning of the end.
You can also get $10/month opencode go (which officially can be used with any harness) with basically every major open weight model running inference in a place that I deem more trustworthy than the subject-to-Chinese-law inference options (but research this for yourself, there are some asterisks). This is what I use (in opencode) when I want an agent for personal use. Even if I'm being fairly intense with my usage by my standards anyway), I haven't managed to run out (yet), although I usually use the cheaper models (deepseek v4 is my favorite, and I've used both flash and pro).
From the FAQ:
> The plan is designed primarily for international users, with models hosted in the US, EU, and Singapore for stable global access. Our providers follow a zero-retention policy and do not use your data for model training.
And:
> Yes, you can use Go with any agent. Follow the setup instructions in your preferred coding agent.
Opencode Go already seems extremely unprofitable if people use up at least 30% of their monthly tokens so its hard to imagine they’d go with several times more expensive inference provider yet show Deepseek pricing on their page.
And Codex obviously allows that as well. Anthropic is pretty the main exception.
They're at the point where copying can get people 90% of what they want anyway, the moat is drying up.
The speed in which a copier can catch up to a leader is faster than every before, but everything is faster these days. I think the threat is if can these labs catch up and pass American labs while keeping costs low, but I don't see that happening.
Only with some industries, mostly electronics and light manufacturing. The rest of the history of software products doesn't look like this at all. China cloned Amazon and Meta fairly well (but only within their local/captured markets) but not Microsoft or Google and absolutely not Apple. Entertainment and culture exports are still effectively zero (like, what, three blockbuster films?) where the US, Japan and Korea define pop across the globe. Heavy industry is mixed, with decent success in consumer vehicles now but little purchase in stuff like heavy equipment or shipbuilding.
Basically it's complicated, and not really a point well suited to that kind of generalism.
Americans order Chinese products from SHEIN and Aliexpress type websites.
So China did penetrate the US ecommerce market, but they are not taking on Amazon directly.
People are buying from the Chinese and reselling on Amazon, which has a good logistics network that Chinese companies can't possibly build in the USA.
> absolutely not Apple
You need to look up Oneplus to see how many people were using their devices till they announced exit from Western markets.
The Chinese Android phones are essentially replacement for a big chunk of Apple ecosystem.
If these cheap mobile devices didn't exist - Apple could possibly be selling many more devices.
> China cloned .... Meta fairly well
TikTok is giving Meta a run for their money in USA.
They had to be restricted via threat of regulation to offload stake in TikTok USA.
> Microsoft
They didn't compete with Microsoft directly as entering Western Enterprise market is hard for the Chinese. Even Google can't beat Microsoft there till now.
The final part of this sentence shows why there is absolutely no need to look up Oneplus.
I'm not sure if you are from the US, but if you are, it really gives you a distorted image of what the worldwide smarphone market is like.
E.g. in many European countries, it is 1/3 iPhone, 1/3 Samsung, 1/3 others (which are mostly Chinese brands).
To get a better idea on the topic that you commented on and views of members of HN
You can read comments on this post - https://news.ycombinator.com/item?id=48932539
And this is really sad, because they have thousands of years of cultural heritage and also plenty of great films. Somehow, they've failed at exporting this completely.
TikTok competes with Meta, not Olivia Rodrigo or BTS.
They even got the car Apple always wanted, but never managed to build.
Uhh, China does over half of the world’s shipbuilding.
China is also responsible for 25% [1] of the world’s heavy equipment manufacturing and is considered to have double[2] the CAGR of the global average for heavy equipment manufacturing.
0: https://www.csis.org/analysis/china-dominates-shipbuilding-i...
1: https://www.grandviewresearch.com/horizon/outlook/constructi...
2: https://www.grandviewresearch.com/industry-analysis/heavy-co...
Yes, and that is how you get a top-heavy k-shaped economy I believe. It is harder and harder to sell premium A+ products, because most people can only afford what works. If my old ice cream vendor charges me 25% more but puts chocolate sprinkles on it i'm going for the non-sprinkle version.
It's like releasing a cutting edge search engine, proclaimed to be on par with Google. Oh yeah, I forgot to mention, you need to actually run it somewhere. One of the core moats for Google has been the cost to scale and compete globally with a search engine, which only a few entities on earth could even attempt. Somehow the HN crowd forgets this every single time. The same claims were endlessly made regarding Uber, as though the app was the hardest part of competing with Uber. Here's a hint: the model isn't the only super hard part of the total process at this point. Try running something akin to GPT 5.6 for a billion users, see how much that costs.
The winners will be the owners of the subscriptions and API usage. That's going to be OpenAI and Anthropic as of now, possibly Google will stay in the fight with their scale advantage focused on the consumer. It will never be the open weight models out of China. Who is going to invest the trillion dollars into infrastructure needed to run the latest version of Kimi 3-5 years out, for a billion users? There's only one entity that would ever consider it, and that's China domestic.
And the real money (for a sustainable business model allowing for constant massive reinvestment & infrastructure maintenance) is in pairing up advertising, which is another spot where the open weight models from China will never compete. It doesn't matter if the HN crowd hates this outcome, nothing will stop it from being the future core of mass consumer usage. And the business use will stay with subscription + API use as well: the best option for that today is OpenAI and Anthropic.
Everybody that is concerned about their data ownership and residency, and how the US political system is behaving right now. There is a huge incentive to cut off ties with API and subscription providers, especially those based in the US.
Frankly I don’t trust any of them. Certainly not Anthropic after they refused to delete my Cowork session. Certainly not OpenAI with their duplicity. I’m a Google workspace subscriber so I use Gemini, but only with logging disabled and for throwaway searches.
I will go at the pace that hardware I control lets me. That happens to be pretty damn slow ATM… waiting for 45 minutes for my 4090 to pp 30k tokens wasn’t much fun the other day, but it got the job done. If speed were a priority I’d propose be renting time on vast.ai or something.
K3 is beating Fable on several benchmarks. So I take it you don't consider Fable 5 to be frontier.
What are the frontier models?
(that's just one, there are others... and several are claiming that those saying "yes sure it's just as good, but it uses more tokens" ain't really true either).
And I don't think anyone is excited about China releasing anything: I think people like the idea of open-weights.
Have you seen Kimi Code ?
It's hard to believe it's a different code than Claude Code.
No one would buy a car steered with pedals; even if that is better (hey planes do it!) - every car maker is "copying" steering wheels and putting them on their cars. Is it controversial? If not why not?
(Before the wheel, ships were steered with tillers - which early cars also tried, interface-wise.)
Actually, Jobs misattributed to Picasso the "Good artists copy, great artists steal." line. Picasso did say:
"When there's anything to steal, I steal."
"What does it mean, an artist copies? It is normal. The important thing is to do it better."
"Art is a series of masterly thefts."I love that quote and like to share it often with people around me
All of his stuff (or most of it) is repurposed and refitted chunks of other classical works. The beautiful part is, he does it so well.
It's clearly a dupe of Claude/Codex products (Codex especially, styling-wise), but I think Kimi's goal here is simply to appear on feature-parity with bigger labs. I doubt they'd want to invest much into designing a UI for "the future of agentic work" from the ground up.
Their bread and butter and claim-to-fame will continue to be low-cost, near-frontier models, for which they need to at least appear to have some app layer to plug into.
And, in some ways, duping is a great way to show people that even the app layer can be commoditized.