Ok it might sound crazy but I actually got the best quality of code (completely ignoring that the cost is likely 10x more) by having a full “project team” using opencode with multiple sub agents which are all managed by a single Opus instance. I gave them the task to port a legacy Java server to C# .NET 10. 9 agents, 7-stage Kanban with isolated Git Worktrees.
Manager (Claude Opus 4.5): Global event loop that wakes up specific agents based on folder (Kanban) state.
Product Owner (Claude Opus 4.5): Strategy. Cuts scope creep
Scrum Master (Opus 4.5): Prioritizes backlog and assigns tickets to technical agents.
Architect (Sonnet 4.5): Design only. Writes specs/interfaces, never implementation.
Archaeologist (Grok-Free): Lazy-loaded. Only reads legacy Java decompilation when Architect hits a doc gap.
CAB (Opus 4.5): The Bouncer. Rejects features at Design phase (Gate 1) and Code phase (Gate 2).
Librarian (Gemini 2.5): Maintains "As-Built" docs and triggers sprint retrospectives.
You might ask yourself the question “isn’t this extremely unnecessary?” and the answer is most likely _yes_. But I never had this much fun watching AI agents at work (especially when CAB rejects implementations).
This was an early version of the process that the AI agents are following (I didn’t update it since it was only for me anyway): https://imgur.com/a/rdEBU5I
Every time I read something like this, it strikes me as an attempt to convince people that various people-management memes are still going to be relevant moving forward.
Or even that they currently work when used on humans today.
The reality is these roles don't even work in human organizations today. Classic "job_description == bottom_of_funnel_competency" fallacy.
If they make the LLMs more productive, it is probably explained by a less complicated phenomenon that has nothing to do with the names of the roles, or their descriptions.
Adversarial techniques work well for ensuring quality, parallelism is obviously useful, important decisions should be made by stronger models, and using the weakest model for the job helps keep costs down.
My understanding is that the main reason splitting up work is effective is context management.
For instance, if an agent only has to be concerned with one task, its context can be massively reduced. Further, the next agent can just be told the outcome, it also has reduced context load, because it doesn't need to do the inner workings, just know what the result is.
For instance, a security testing agent just needs to review code against a set of security rules, and then list the problems. The next agent then just gets a list of problems to fix, without needing a full history of working it out.
Which, ultimately, is not such a big difference to the reason we split up work for humans, either. Human job specialization is just context management over the course of 30 years.
I’ve found that task isolation, rather than preserving your current session’s context budget, is where subagents shine.
In other words, when I have a task that specifically should not have project context, then subagents are great. Claude will also summon these “swarms” for the same reason. For example, you can ask it to analyze a specific issue from multiple relevant POVs, and it will create multiple specialized agents.
However, without fail, I’ve found that creating a subagent for a task that requires project context will result in worse outcomes than using “main CC”, because the sub simply doesn’t receive enough context.
So two things.. Yes this helps with context and is a primary reason to break out the sub-agents.
However one of the bigger things is by having a focus on a specific task or a role, you force the LLM to "pay attention" to certain aspects. The models have finite attention and if you ask them to pay attention to "all things".. they just ignore some.
The act of forcing the model to pay attention can be acoomplished in alternative ways (defined process, commitee formation in single prompt, etc.), but defining personas at the sub-agent is one of the most efficient ways to encode a world view and responsibilities, vs explicitly listing them.
You can create a context that includes info and instructions, but the agent may not pay attention to everything in the context, even if context usage is low.
IMO "Attention" is an abstraction over the result of prompt engineering, the chain reaction of input converging the output (both "thinking" and response).
If so, one benefit is you can quickly and safely mix up your set of agents (a la Inverse Conway Manoeuvre) without the downsides that normally entails (people being forced to move teams or change how they work).
I think it's just the opposite, as LLMs feed on human language. "You are a scrum master." Automatically encodes most of what the LLM needs to know. Trying to describe the same role in a prompt would be a lot more difficult.
Maybe a different separation of roles would be more efficient in theory, but an LLM understands "you are a scrum master" from the get go, while "you are a zhydgry bhnklorts" needs explanation.
-Tested 162 personas across 6 types of interpersonal relationships and 8 domains of expertise, with 4 LLM families and 2,410 factual questions
-Adding personas in system prompts does not improve model performance compared to the control setting where no persona is added
-Automatically identifying the best persona is challenging, with predictions often performing no better than random selection
-While adding a persona may lead to performance gains in certain settings, the effect of each persona can be largely random
Fun piece of trivia - the paper was originally designed to prove the opposite result (that personas make LLMs better). They revised it when they saw the data completely disproved their original hypothesis.
Persona’s is not the same thing as a role. The point of the role is to limit what the work of the agent, and to focus it on one or two behaviors.
What the paper is really addressing is does key words like you are a helpful assistant give better results.
The paper is not addressing a role such as you are system designer, or you are security engineer which will produce completely different results and focus the results of the LLM.
I would be interested in an eval that checked both conditions: you are an amazing x Vs. you are a terrible x. also there have been a bunch of papers recently looking at whether threatening the llm improves output, would like to see a variation that tries carrot and stick as well.
Most model research decays because the evaluation harness isn’t treated as a stable artefact. If you freeze the tasks, acceptance criteria, and measurement method, you can swap models and still compare apples to apples. Without that, each release forces a reset and people mistake novelty for progress.
Fair point on the date - the paper was updated October 2024 with Llama-3 and Qwen2.5 (up to 72B), same findings. The v1 to v3 revision is interesting. They initially found personas helped, then reversed their conclusion after expanding to more models.
"Comprehensively disproven" was too strong - should have said "evidence suggests the effect is largely random." There's also Gupta et al. 2024 (arxiv.org/abs/2408.08631) with similar findings if you want more data points.
A paper’s date does not invalidate its method. Findings stay useful only when you can re-run the same protocol on newer models and report deltas. Treat conclusions as conditional on the frozen tasks, criteria, and measurement, then update with replication, not rhetoric.
Developers do want managers actually, to simplify their daily lives. Otherwise they would self manage themselves better and keep more of the share of revenues for them
Unfortunately some managers get lonely and want a friendly face in their org meetings, or can’t answer any technical questions, or aren’t actually tracking what their team is doing. And so they pull in an engineer from their team.
Being a manager is a hard job but the failure mode usually means an engineer is now doing something extra.
It shows me that there doesn’t appear to be an escape from Conway’s Law, even when you replace the people in an organisation with machines. Fundamentally, the problem is still being explored from the perspective of an organisation of people and it follows what we’ve experienced to work well (or as well as we can manage).
I do think there is some actual value in telling an LLM "you are an expert code reviewer". You really do tend to get better results in the output
When you think about what an LLM is, it makes more sense. It causes a strong activation for neorons related to "code review", and so the model's output sounds more like a code review.
i guess, as a human it’s easier to reason about a multi-agent system when the roles are split intuitively, as we all have mental models. but i agree - it’s a bit redundant/unnecessary
Subagent orchestration without the overhead of frameworks like Gastown is genuinely exciting to see. I’ve recorded several long-running demos of Pied-Piper, which is a Subagents orchestration system for Claude Code and ClaudeCodeRouter+OpenRouter here: https://youtube.com/playlist?list=PLKWJ03cHcPr3OWiSBDghzh62A...
I came across a concept called DreamTeam, where someone was manually coordinating GPT 5.2 Max for planning, Opus 4.5 for coding, and Gemini Pro 3 for security and performance reviews. Interesting approach, but clearly not scalable without orchestration. In parallel, I was trying to do repeatable workflows like API migration, Language migration, Tech stack migration using Coding agents.
Pied-Piper is a subagent orchestration system built to solve these problems and enable repeatable SDLC workflows. It runs from a single Claude Code session, using an orchestrator plus multiple agents that hand off tasks to each other as part of a defined workflow called Playbooks:
https://github.com/sathish316/pied-piper
Playbooks allow you to model both standard SDLC pipelines (Plan → Code → Review → Security Review → Merge) and more complex flows like language migration or tech stack migration (Problem Breakdown → Plan → Migrate → Integration Test → Tech Stack Expert Review → Code Review → Merge).
Ideally, it will require minimal changes once Claude Swarm and Claude Tasks become mainstream.
Personally, I'm fascinated by the opening for protocol languages to become relevant.
The previous generations of AI (AI in the academic sense) like JASON, when combined with a protocol language like BSPL, seems like the easiest way to organize agent armies in ways that "guarantee" specific outcomes.
The example above is very cool, but I'm not sure how flexible it would be (and there's the obvious cost concern). But, then again, I may be going far down the overengineering route.
I have been using a simpler version of this pattern, with a coordinator and several more or less specialized agents (eg, backend, frontend, db expert). It really works, but I think that the key is the coordinator. It decreases my cognitive load, and generally manages to keep track of what everyone is doing.
Share your code of the “actual best quality “ or this is just another meaningless and suspicious attempt to get users to put the already expensive AI in a for-loop to make it even more expensive
Can you share technical details please? How is this implemented? Is it pure prompt-based, plugins, or do you have like script that repeatedly calls the agents? Where does the kanban live?
Not the OP, but this is how I manage my coding agent loops:
I built a drag and drop UI tool that sets up a sequence of agent steps (Claude code or codex) and have created different workflows based on the task. I'll kick them off and monitor.
I’ve been messing around with the BMAD process as well which seems like a simpler workflow than you described. My only concern is that it’s able to get 90% of the way there for productionized ready code, but the last 10% is starts to fail at when the tech debt gets too large.
Have you been able to build anything productionizable this way, or are you just using this workflow for rapid prototyping?
This is genuinely cool, the CAB rejecting implementations must be hilarious to watch in action. The Kanban + Git worktree isolation is smart for keeping agents from stepping on each other.
I've been working on something in this space too. I built https://sonars.dev specifically for orchestrating multiple Claude Code agents working in parallel on the same codebase. Each agent gets its own workspace/worktree and there's a shared context layer so they can ask each other questions about what's happening elsewhere (kind of like your Librarian role but real-time).
The "ask the architect" pattern you described is actually built into our MCP tooling: any agent can query a summary of what other agents have done/learned without needing to parse their full context.
1. Are you using a Claude Code subscription? Or are you using the Claude API? I'm a bit scared to use the subscription in OpenCode due to Anthropic's ToS change.
2. How did you choose what models to use in the different agents? Do you believe or know they are better for certain tasks?
You might as well just have planner and workers, or your architecture essentially echos to such structure. It is difficult to discern how semantics can drive to different behavior amongst those roles, and why planner can't create those prompts the ad-hoc way.
This now makes me think that the only way to get AI to work well enough to actually actually replace programmers will probably be paying so much for compute that it's less expensive to just have a junior dev instead.
What are the costs looking like to run this? I wonder whether you would be able to use this approach within a mixture-of-experts model trained end-to-end in ensemble. That might take out some guesswork insofar the roles go.
I was getting good results with a similar flow but was using claude max with ChatGPT. unfortunately not an option available to me anymore unless either I or my company wants to foot the bill.
Nope it isn’t. I did it as a joke initially (I also had a version where every 2 stories there was a meeting and if a someone underperformed it would get fired).
I think there are multiple reasons why it actually works so well:
- I built a system where context (+ the current state + goal) is properly structured and coding agents only get the information they actually need and nothing more. You wouldn’t let your product manager develop your backend and I gave the backend dev only do the things it is supposed to and nothing more. If an agent crashes (or quota limits are reached), the agents can continue exactly where the other agents left off.
- Agents are ”fighting against” each other to some extend? The Architect tries to design while the CAB tries to reject.
- Granular control. I wouldn’t call “the manager” _a deterministic state machine that is calling probabilistic functions_ but that’s to some extent what it is? The manager has clearly defined tasks (like “if file is in 01_design —> Call Architect)
Here’s one example of an agent log after a feature has been implemented from one of the older codebases:
https://pastebin.com/7ySJL5Rg
What OpenCode primitive did you use to implement this? I'd quite like a "senior" Opus agent that lays out a plan, a "junior" Sonnet that does the work, and a senior Opus reviewer to check that it agrees with the plan.
You can define the tools that agents are allowed to use in the opencode.json (also works for MCP tools I think).
Here’s my config: https://pastebin.com/PkaYAfsn
The models can call each other if you reference them using @username.
This is excellent, thank you. I came up with half of this while waiting for this reply, but the extra pointers about mentioning with @ and the {file} syntax really helps, thanks again!
> [...]coding agents only get the information they actually need and nothing more
Extrapolating from this concept led me to a hot-take I haven't had time to blog about: Agentic AI will revive the popularity of microservices. Mostly due to the deleterious effect of context size on agent performance.
Why would they revive the popularity of microservices? They can just as well be used to enforce strict module boundaries within a modular monolith keeping the codebase coherent without splitting off microservices.
And that's why they call it a hot take. No, it isn't going to give rise to microservices. You absolutely can have your agent perform high-level decomposition while maintaining a monolith. A well-written, composable spec is awesome. This has been true for human and AI coders for a very, very long time. The hat trick has always been getting a well-written, composable spec. AI can help with that bit, and I find that is probably the best part of this whole tooling cycle. I can actually interact with an AI to build that spec iteratively. Have it be nice and mean. Have it iterate among many instances and other models, all that fun stuff. It still won't make your idea awesome or make anyone want to spend money on it, though.
In a fresh project that is well documented and set up it might work better. Many issues that Agents have in my work is that the endpoints are not always documented correctly.
Real example that happened to me, Agent forgets to rename an expected parameter in API spec for service 1. Now when working on service 2, there is no other way of finding this mistake for the Agent than to give it access to service 1. And now you are back to "... effect of context size on agent performance ...". For context, we might have ~100 services.
One could argue these issues reduce over time as instruction files are updated etc but that also assumes the models follow instructions and don't hallucinate.
That being said, I do use Agents quite successfully now - but I have to guide them a bit more than some care to admit.
> In a fresh project that is well documented and set up it might work better.
I guess this may be dependent on domain, language, codebase, or soke combination of the 3. The biggest issues I've had with agents is when they go down the wrong path and it snowballs from there. Suddenly they are loading more context unrelated to the tasks and getting more confused. Documenting interfaces doesn't help if the source is available to the agent.
My agentic sweet spot is human-designed interfaces. Agents cannot mess up code they don't have access to, e.g. by inadvertently changing the interface contract and the implementation.
> Agent forgets to rename an expected parameter in API spec for service 1
Document and test your interfaces/logic boundaries! I have witnessed this break many times with human teams with field renames, change in optionality, undocumented field dependencies, etc, there are challenging trade-offs with API versioning. Agents can't fix process issues.
Isn't all this a manual implementation of prompt routing, and, to a lesser extent, Mixture of Experts?
These tools and services are already expected to do the best job for specific prompts. The work you're doing pretty much proves that they don't, while also throwing much more money at them.
How much longer are users going to have to manually manage LLM context to get the most out of these tools? Why is this still a problem ~5 years into this tech?
I'm confused when you say you have a manager, scrum master, archetech, all supposdely sharing the same memory, do each of those "employees" "know" what they are? And if so, based on what are their identities defined? Prompts? Or something more. Or am I just too dumb to understand / swimming against the current here. Either way, it sounds amazing!
It's not satire but I see where you're coming from.
Applying distributed human team concepts to a porting task squeezes extra performance from LLMs much further up the diminishing returns curve. That matters because porting projects are actually well-suited for autonomous agents: existing code provides context, objective criteria catch more LLM-grade bugs than greenfield work, and established unit tests offer clear targets.
I guess what I'm trying to say is that the setup seems absurd because it is. Though it also carries real utility for this specific use case. Apply the same approach to running a startup or writing a paid service from scratch and you'd get very different results.
I don't know about something this complex, but right this moment I have something similar running in Claude Code in another window, and it is very helpful even with a much simpler setup:
If you have these agents do everything at the "top level" they lose track. The moment you introduce sub-agents, you can have the top level run in a tight loop of "tell agent X to do the next task; tell agent Y to review the work; repeat" or similar (add as many agents as makes sense), and it will take a long time to fill up the context. The agents get fresh context, and you get to manage explicitly what information is allowed to flow between them. It also tends to mean it is a lot easier to introduce quality gates - eg. your testing agent and your code review agent etc. will not decide they can skip testing because they "know" they implemented things correctly, because there is no memory of that in their context.
Humans seem to be similar. If a real product designer would dive into all the technical details and code of a product, he would likely forget at least some of the vision behind what the product is actually supposed to be.
I mean no offence to anyone but whenever new tech progresses rapidly it usually catches most unaware, who tend to ridicule or feel the concepts are sourced from it.
ai is actually useful tho. idk about this level of abstraction but the more basic delegation to one little guy in the terminal gives me a lot of extra time
I think many people really like the gamification and complex role playing. That is how GitHub got popular, that is how Rube Goldberg agent/swarm/cult setups get popular.
It attracts the gamers and LARPers. Unfortunately, management is on their side until they find out after four years or so that it is all a scam.
I've heard some people say that "vibe coding" with chatbots is like slot machines, you just keep "propmting" until you hit the jackpot. And there was some earlier study that people _felt_ more productive even if they weren't (caveat that this was with older models), which aligns with the sort of time-dilation people feel when gambling.
I guess "agentic swarms" are the next evolution of the meta-game, the perfect nerd-sniping strategy. Now you can spend all your time minmaxing your team, balancing strengths/weaknesses by tweaking subagents, adding more verifiers and project managers. Maybe there's some psychological draw, that people can feel like gods and have a taste of the power execs feel, even though that power is ultimately a simulacra as well.
Extending this -- unlike real slot machines, there is no definite state of won or not for the person prompting, only if they've been convinced they've won, and that comes
down to how much you're willing to verify the code it has provided, or better, fully test it (which no one wants to do), versus the reality where they do a little light testing and say it's good enough and move on.
Recently fixed a problem over a few days, and found that it was duplicated though differently enough that I asked my coworker to try fixing it with an LLM (he was the originator of the duplicated code, and I didn't want to mess up what was mostly functioning code). Using an LLM, he seemingly did in 1 hour what took me maybe a day or two of tinkering and fixing. After we hop off the call, I do a code read to make sure I understand it fully, and immediately see an issue and test it further only to find out.. it did not in fact fix it, and suffered from the same problems, but it convincingly LOOKED like it fixed it. He was ecstatic at the time-saved while presenting it, and afterwards, alone, all I could think about was how our business users were going to be really unhappy being gaslit into thinking it was fixed because literally every tester I've ever met would definitely have missed it without understanding the code.
People are overjoyed with good enough, and I'm starting to think maybe I'm the problem when it comes to progress? It just gives me Big Short vibes -- why am I drawing attention to this obvious issue in quality, I'm just the guy in the casino screaming "does no one else see the obvious problem with shipping this?" And then I start to understand, yes I am the problem: people have been selling eachother dog water product for millenia because at the end of the day, Edison is the person people remember, not the guy who came after that made it near perfect or hammered out all the issues. Good enough takes its place in history, not perfection. The trick others have found out is they just need to get to the point that they've secured the money and have time to get away before the customer realizes the world of hurt they've paid for.
The next stage in all of this shit is to turn what you have into a service. What's the phrase? I don't want to talk to the monkey, I want to talk to the organ grinder. So when you kick things off it should be a tough interview with the manager and program manager. Once they're on board and know what you want, they start cracking. Then they just call you in to give demos and updates. Lol
I am still torn on this issue. On the one hand, it feels like a copyright violation when other people's works are used to train an ML model. On the other hand, it is not a copyright infringement if I paint a picture in the Studio Ghibli style myself. The question is whether removing a ‘skill requirement’ for replication is sufficient grounds to determine a violation.
I believe the case is that you're welcome to paint a picture perfectly copying Studio Ghibli, but you cannot sell it. You're welcome to even take the style and add enough personal creativity that it becomes a different work and sell that, but only if a random on the street doesn't look at it and say "wow, what Studio Ghibli film is that from?".
That's the problem here, there's no creative input apart from the prompt, so obviously the source is blatant (and often in the prompt).
> I believe the case is that you're welcome to paint a picture perfectly copying Studio Ghibli, but you cannot sell it.
Technically, you can't, but there's no way to enforce copyright infringement on private work.
You can paint a Studio Ghibli-style painting -- the style isn't protected.
These rules assume that copying the style is labor intensive, and righteously rewards the worker.
When an LLM can reproduce thousands and thousands of Ghibli-style paintings effortlessly, not protecting the style seems less fair, because the work of establishing the Ghibli-style was harder than copying it large-scale.
I'm in the "don't fight a roaring ocean, go with the flow" boat:
If your entire livelihood depends on having the right to distribute something anyone can copy, get a stronger business.
Large brand copyright holders gonna sue. It is part of their core business.
If you’re a musician, you gotta tour to make money. If you’re a painter, you need patrons.
There already were a ton of ways organised IP theft would make money on your creative force.
AI training seems different because it can gobble up anything out of order and somehow digest it into something valuable without human intervention.
The training is absurdly expensive, but considering how they capture the profit on the value created, and the training input owners won’t, will just mark the end of the open internet for non-FOSS normies.
A lot of people will need to stop identifying as 'someone who makes art with enough value to trade it for the decent wage'. Millions of egos destroyed, not a dollar of GDP lost.
Not a dollar lost, and yet all of society is poorer. Less social commentary via art. Less beauty. Less novelty and less new forms invented. Less entertrainment (entertainment that has some human values imbued into it, but there will be more 'entertainment', just now devoid of intentionality/humanity/novelty, a firehose of images/noises/colors with nothing behind it, a firehouse of slop). Less personal discipline trying to master something. Less seeing through the eyes of another person, so less relation, less empathy, less humanity. Less, less, less. AI is an inertia machine.
No dollars burned, just huge huge amounts of cultural capital. Just a reduction to a more primitive, less cultured/developed version of what it means to be human. Less thinking out loud, sharing of thoughts, exposure to new thoughts. More retreating into (a now lesser developed, now culturally atrophied) self.
So I guess you don't read books/comics, watch TV/Movies/plays, play video games or listen to music? People aren't born skilled artists, that takes time and effort. Being able to prompt GenAI well just makes you a skilled prompter, not a skilled artist. Over time, we will lose a lot of skilled artists and that is something worth thinking more deeply about, instead of giving a callous hot take. Artists are trained to view the world critically, and I want more critical thinkers - not less.
My wife and I went for wall decoration. There’s an art gallery and a poster shop right next to each other. The price difference is a factor 100 for an average art piece.
In the poster shop you can choose between a bunch of classics, or you can upload your own AI-generated picture and have that printed as a poster.
Art was always expensive, and posters as an alternative to paintings existed way before AI. Same with copying all kinds of art.
The main difference seems to be that we can’t clearly pay royalties to anyone for AI artwork, because it’s not obvious exactly where it came from.
There was a YouTube channel dedicated to Warhammer lore narrated by an AI David Attenborough. It got taken down for infringing on his voice, but its replacement came up, starting out with a generic old man’s voice and over time gradually more Attenborough-like. When should the Attenborough estate start to get royalties? At 60% Attenborough? Or at 80% Attenborough?
In my original comment I was asking people to follow me into an imaginary future where there are less artists. Artists reveal something about the world that speaks to us, which they do through critically breaking down and reforming what they see. I can't remember who said it, but they said when art speaks to you, it's a momentary bridge between the artist's soul and yours.
I'll answer your question, but my question for you is: why were you buying wall decorations in the first place? To me, it sounds like you were searching for a product category, and not specifically for art.
Regarding your example, if the AI is capable of imitating David Attenborough by including his name in the prompt, then it was probably trained on his data. If he didn't consent, then I might argue that is ethically wrong and, in my view, theft. If the channel was not monetized and done without his consent, I might argue that is just an ethical failing. In using his voice, the channel betrays the fact that it has value, otherwise they would continue to use the random old man voice.
Two of the four core tests for fair use hinge on this.
1. Purpose and character of the use. With emphasis on whether the copy was made for commercial use.
4. Effect on the work's value, and the creator's ability to exploit their work.
---
Both can be dramatically impacted by the intent of the copy, usually with enforcement and punishment also being considerably stronger if the copy is being made for commercial gain and not private use.
This for the actual work, definitely not for transformative works like llm output. So a ghibli style image of me is fine legally, whether I sell it or not.
The language used to describe LLM behaviour such as "training" and "reasoning" has led people to treat them the same as humans, instead of a new and different entity that requires us to update our set of rules.
If I was the first person to invent a car, for example, and I named its method of locomotion "walking", would you treat it the same as a human and let it "walk" in all the same places humans walk? After all, it's simply using kinetic energy and friction to propel itself along the ground, as we do.
Because a car is so obviously different to a human, we intuitively understand it requires an alteration to our rules in order for us to coexist peacefully. Since LLMs are so abstract, we don't intuitively understand this distinction, and so continue to treat them as if they should be bound by the same rules and laws as us.
It's less about paint a picture yourself, arguably there is little to no value there. OpenAI et al, sell the product of creating pictures in the style of their material. I see this as a direct competition to Studio Ghibli's right to produce their own material with their own IP.
I agree with this. I don't know how to create artistic styles by hand or using any creative software for that matter. All the LLM tools out there gave me the "ability" and "talent" to create something "good enough" and, in some cases, pretty close to the original art.
I rarely use these tools (I'm not in marketing, game design, or any related field), but I can see the problem these tools are causing to artists, etc.
Any LLM company offering these services needs to pay the piper.
Is it not a copyright infringement if you pain it yourself? Why is that the case? I thought it would be just that studios wouldn't care for the most part. Hasn't Disney gone after this type of personal projects in the past?
I am mostly ok with these copyright crackdowns in AI in the spirit of - if a human were to do it commercially, it would be illegal.
We can argue if that should be the case or not, which is a different issue.
However, it should not be legal to automate something at scale that is illegal when done by an individual human. Allowing it just tips the scale against labor even more.
You are not a for profit software product. You are a human. If you make a drawing, that drawing MIGHT be a for profit product.
If I make a for profit AI, that AI is a product. And if that product required others' copyrighted works to derive it's deliverable, it is by definition creating derivative works. Again, creating a small human, not creating a product. Creating a for profit AI, creating a product.
If my product couldn't produce the same output without having at some point consumed the other works, I've triggered copyright concerns.
If I make and train a small human, I am not creating a for profit product so it doesn't come in to play at all. The two are not similar in any way. The human is not the product. If THEY create a product later (a drawing in this case) then THAT is where copyright comes in.
It is a tough issue, but legalese will probably start from where and on what and how did this thing train on in order to be able to produce such results. It's similar to (guitar) amp modeling. Eventually they will either have to remove it or license it. I don't see other way. What makes it challenging is that vast majority of things that it was trained on can make a similar claim and it opens the floodgate. Outcome is either youtube-like thing where things exist in certain capacity, but not full or alternative is Napster like destiny where it gets banned altogether. Stakes are now a bit too high for Napster scenario. OpenAI might have a youtube-like angle for licensing with its Sora thing which seems to be turning into its own social network of sorts.
If you paint a Studio Ghibli totoro on your cup, pillow, PC, T-shirt, nobody is going to care. If you do this a thousand times it obviously is an issue. And if you charge people to access you to do this, it is also obviously an issue.
The difference is that you’re not making money off your pictures (or else you'd probably need a licensing agreement), whereas the AI companies are making money off of the IP holders.
it's a form of slavery when someone is profiting off the looong and hard, concentrated work of others without reimbursing them adequately.
that's why piracy Robin Hood Style is fine, but corporate piracy is not. I downloaded a Ghibli movie because I could't afford the DVD. I didn't copy it on VHS then to sell it via e-commerce to 1000 people.
AI companies grabbed IP and pull hundreds of thousands of customers with it, then collect their interactions either way and profit exponentially while Ghibli, Square Enix et al. don't profit from users using more and more AI ...
and most people are not "training" ML models ... people are using copy machines that already learned to compensate for their lack of will to put effort into stuff.
a lot of us been there and enough decided to move beyond and get/become better at being human aka evolve vs get cozy in some sub-singularity. some didn't and won't, and they are easiest to profit from.
It is absolutely infringement if you paint a picture in Ghibli style. You just have fair use to infringe in a personal, noncommercial, educational, etc. purpose.
Fair use is a defense to infringement, like self defense is a defense to homicide. If you infringe but are noncommercial, it is more likely to be ruled fair use. If Disney did a Ghibli style ripoff for their next movie, that is clearly not fair use.
OpenAI is clearly gaining significant material benefits from their models being able to infringe Ghibli style.
> It is absolutely infringement if you paint a picture in Ghibli style.
Of course not because by this twisted logic every piece of art is inspired by what comes before and you could claim Ghibli is just a derivative of what came before and nobody has any copyright then...
> The threat actor appears to have obtained this information by paying multiple contractors or employees working in support roles outside the United States to collect information from internal Coinbase systems to which they had access in order to perform their job responsibilities
Based on the information present in the breach, I think it's likely that the source was their customer support in the Philippines. Monthly salary is usually < 1000$/month (entry-level probably even less than 500$) and a 5000$ bribe could be more than a year worth of money, tax-free. Considering the money you can make with that dataset now, this is just a small investment.
>
•Name, address, phone, and email;
•Masked Social Security (last 4 digits only);
•Masked bank-account numbers and some bank account identifiers;
•Government‑ID images (e.g., driver’s license, passport);
•Account data (balance snapshots and transaction history); and
•Limited corporate data (including documents, training material, and communications available to support agents).
This is every threat actor's dream. Even if you only had email addresses and account balances, this is a nightmare. Instead of blackmailing the company, you can now blackmail each individual user. "Send me 50% of your BTC and I won't publish all of your information on the internet".
My guess is that we will have a similar situation like we had with the Vastaamo data breach...
Blackmail would be the least of my worries, in France we had at least five kidnappings/attempted kidnappings related to crypto investors since the beginning of the year.
Crypto is advertised as providing irreversible transfers, and having ownership of assets solely established by ownership of keys. It shouldn't be surprising that such features would attract criminals.
You can easily establish the connection from a bank account to a person. A connection from a crypto wallet to a person is extremely difficult.
Money laundering with crypto is also much easier (and cheaper usually).
In the vast majority of cases, it's actually extremely easy. It took less than an afternoon for me to learn how to trace 90%+ of transactions on either BTC or any of the networks built on Ethereum or an Ethereum-like protocol. There are large companies that specialize in exactly this, which make tools that allow government agents who have no particular crypto expertise to trace the majority of transactions.
It is possible to make your transactions extremely difficult to trace, but you really, really,REALLY have to know what you're doing.
Law enforcement loves that people think it's easy and cheap to launder money with crypto, though. It's made it vastly easier for them to catch those people!
I never doubted that it's possible but it's way harder than identifying bank accounts.
There is a massive business behind crypto tracking, that's why companies like MasterCard have acquired CipherTrace. Some years ago there was a really good article / case study from them. I think it was related to a ransomware gang and they were able to identify the threat actor's wallets through crypto tumblers and chain hopping.
It's just a matter of how much money and time are you willing to invest into finding out and not a matter of possibility.
You can trace the BTC or Ethereal transaction of coins, but you cannot trace the criminals after it's converted to Monero or some other "privacy" chain on an exchange run on the dark web. After that you're just tracing other owners, possibly who have no idea where that it was stolen. It literally takes a few hours to wash it all out.
Seems unlikely given who has been targeted. I doubt the Ledger or Paymium guys have been evading tax on crypto given that they're publicly involved in it and likely would be scrutinised more than the average person by tax authorities.
This may seem callous, but isn't a large point of crypto that you are 'free' from the shackles imposed by the State?
And I guess that includes protection from criminals by the oppressive forces of the State (aka the police). In which case being kidnapped and having your fingers sent to your family is an integral part of your 'freedom'.
Crypto isn’t synonymous with anarchy, just like the internet isn’t synonymous with pornography. Both are cliches from long ago.
All of the victims are likely tax payers. Law and order is a fundamental service that a legitimate state must provide to all in its jurisdiction, even those who are only resident non-citizens and those that pay little to no taxes in a progressive tax system.
> Crypto isn’t synonymous with anarchy, just like the internet isn’t synonymous with pornography. Both are cliches from long ago.
Saying crypto isn’t synonymous with anarchy, like the internet isn’t with pornography, sidesteps the point. Pornography is just one use of the internet — not its central purpose.
But crypto wasn’t just built to host financial activity — it was designed to restructure it, removing reliance on central authorities. That core intent isn’t a cliché; it’s a defining feature.
Comparing it to incidental internet content is a rhetorical deflection, not a real counterpoint.
That's not what it was designed for, that's just a mixture of propaganda and confusion.
It was designed to solve the double-spending problem with digital currencies, replacing the need for "a authoritative ledger" with a one difficult to forge.
The political project around this was to provide people with a deflationary currency akin to gold, whose inflation could not be controlled by government.
The lack of government control over the inflation of this particular currency, and the lack of an authoritative ledger, are an extremely minimal sense of currency protections (, freedoms). They have as much to do with anarchy as the internet had with porn.
It was designed to avoid the need for existing financial institutions. The doublespend problem was merely the blocker that prevented people from otherwise doing it.
> A purely peer-to-peer version of electronic cash would allow online
payments to be sent directly from one party to another without going through a
financial institution.
Your point is merely a non sequitur: a change in banking isn’t related to paying taxes or the state as a whole, nor anarchy.
You’re not supporting your central thesis that disintermediating finance is in any way related to removing government — and people using Coinbase, a service that is centralized and does collaborate with government regulation seems to directly counter your stereotype of the customers.
Their point is correct: people who match your fantasy wouldn’t be Coinbase customers — you’re relying on old tropes.
Most (developed states at least) don’t claim the monetary system as a taxation medium. Debasement of currency is a bug not a feature. In the US, you are not required to process your transaction in USD but only need it to pay taxes.
Failed countries (ie: Turkey) rely on the financial system for taxation. Functioning countries shouldn’t care or be bothered by it.
It seems that law-abiding citizens often bear the greatest risk by declaring their assets to tax authorities and relying on so-called "trusted custodians" for savings. Ironically, for many, the safest course of action is likely non-disclosure, though this is, of course, illegal in much of the world.
I only have to declare crypto < 1 year in my holding which means that, while technically illegal to buy 1 second after the new tax year start and not declaring it, in practice, obviously, no-one cares about that. Especially as crypto is not a 1 second buy; it can take hours.
This may be surprising, but I actually don't think opting for a payment method with less consumer protections (that I pay cap gains tax on when if I dispose of it for a profit) is me ceding my right to be protected by the police. You're right that it does seem extremely callous and is honestly a disturbing mindset to have. Hopefully you never experience terror like the victims of the last few months in France experienced in your life.
> You're right that it does seem extremely callous and is honestly a disturbing mindset to have. Hopefully you never experience terror like the victims of the last few months in France experienced in your life.
Thanks for the tone-policing. But instead of implicitly suggesting that my mindset or tone is inappropriate, it would be great if we discussed the substance of the points.
> it would be great if we discussed the substance of the points.
Sure, just read the sentence from my response that you skipped over.
To be clear: I didn't implicitly suggest that your mindset of people who use crypto somehow ceding their right to protection from the state was inappropriate, I stated outright that it was a disturbing and callous mindset.
It's like suggesting that people who protest against police brutality shouldn't get protection from the police in emergency situations, or believe people who are racist to healthcare workers should lose all right to healthcare. The type of mindset held by those who care more about retribution against those who hold different views than a just society.
You can argue that once you are 'free' to own guns, defend yourself, and seek revenge. The state limits your ability to protect yourself, so it has to assume that responsibility.
The persons in France probably paid their taxes. So no, your premise is wrong in that the state will help vs. in a crypto no-tax world. Actually the de-jour crypto paradise didn’t have any kidnappings so far and you don’t have to pay taxes either.
Depends on if they cashed out and how they did it. There was a big trend for a while to go live in Portugal for a while, enough to be considered a tax resident there, and then cash out there because (at the time, idk if it's still true), they had no (or little) tax on crypto cash out.
Yeah, I know two French people who did it (one of them avoided UK taxes as he was paid in crypto while working in the UK, the other it's muddier). I know three people in the space, and only those two were on the financial side, so to me, while Blockchain is still a legit tech, anybody using cryptocurrency I peg as a tax evader.
Good thing we have courts, lawyers and judges for that. It’s funny everyone here hates on Trump but as soon as something align with their view, they want a defacto no due process application.
Sorry if i implied anything, i must have missed part of the conversation, i was just confirming that did happen (taking the portugese residency to avoid crypto tax) a few years ago. In my opinion, police should protect even violent criminals from violence when possible, so of course i'm not advocating for anything to happen on tax "avoiders", and they should be protected. I was just stating that i know people in the crypto space, and if you are, i immediately peg you as a small-time sociopath from my past experience.
Also i don't care about them getting judged for tax evasion, i know they won't be and honestly, good for them. I also don't care for nonviolent thieves and think the same thing about them. Profiteering was not how i was raised, but i understand different people have different standards (and parents, luckily mine are great, it's not the case for everybody). People do what they need to do, i found some comportment sociopathic, but as long as it is nonviolent, i'm not mad.
Which state are you talking about? The 0% tax bracket for long-term capital gains in the U.S. for 2024 for single filers was $47,024, never mind the standard deduction. Then it goes up to 15%, then 20%.
It way worse. The US companies, pay $3-$6 per hour to outsource their support to the Philippines. The companies which provide the service have very high turnover rate. For some companies the employees stay on average about 6 months. There is absolutely no reason to be loyal.
We are getting zero government regulations on AI, no punishment data breaches, and no human protections against wide scale abuse. The opposite is happening.
I suspect to see America in chaos from these disruptions in the very near future.
Beyond the Philippines low wage, the point is that there is a price for "everybody" if it were in the US it will be a much higher price, and most probably paying for higher attack benefits.
This is perfect advertising for the Ladybird browser. I hope that some of the developers (if this really goes live on the release channel) will join other projects. I can understand that Mozilla needs money, but I don't think this feature fits with Firefox and what it stands for.
It's bad press for Firefox, but honestly it just makes me want to use the internet less. Ladybird is cool, but the incompatibility with badly-made websites only intensifies the more niche your browser gets. Librewolf or the resurrected Sero browser feels like the next best thing, but even so it feels like a losing battle.
The website uses "Enter your 6-digit authentication code" as an example and then shows a 4-digit auth code in the text field
https://imgur.com/a/u4STHPe
I don't understand why the software is built how it's built. Why would you want to implement licensing in the future for a software product that only creates fake processes and registry keys from a list: https://pastebin.com/JVZy4U5i .
The limitation to 3 processes and license dialog make me feel uncomfortable using the software. All the processes are 14.1MB in size (and basically the scarecrow_process.dll - https://www.virustotal.com/gui/file/83ea1c039f031aa2b05a082c...). I just don't understand why you create such a complex piece of software if you can just use a Powershell script that does exactly the same using less resources. The science behind it only kinda makes sense. There is some malware that is using techniques to check if there are those processes are running but by no means is this a good way to keep you protected. Most common malware like credential stealers (redline, vidar, blahblah) don't care about that and they are by far the most common type of malware deployed. Even ransomware like Lockbit doesn't care, even if it's attached to a debugger. I think this mostly creates a false sense of security and if you plan to grow a business out of this, it would probably only take hours until there would be an open source option available. Don't get me wrong - I like the idea of creating new ways of defending malware, what I don't like is the way you try to "sell" it.
They know that if this idea catches on, a dozen completely free imitations will crop up, so ... the time to grab whatever cash can be squeezed out of this is now.
McAfee/Norton/etc. could license signed "scarecrow" versions of their products for use with something like this so that it's impossible for the malware to distinguish a scarecrow version of MacAfee from the real thing (and they would get a cut/kickback).
I would pay a small amount for a scarecrow version of AV software if a) it had zero footprint on my system resources, and b) it really did scare away malware that checks for such things.
Either way, though, it makes malware more onerous to develop since it has to bundle in public keys in order to verify running processes are correctly signed.
Are you telling me this thing spawned 50 new processes on your computer? Could you zip up all the executable files and whatever it installed and upload it somewhere so we can analyze the assembly?
Is there a way for me to curl their executable into my UNIX terminal so I can read the assembly? Or does Any Run keep the samples to themselves? I know a lot about portable executable but very little about these online services.
To your point, I made this a few years ago using powershell. I just created a stub .exe using csc on install and renamed it to match a similar list of binary names. Maybe I will dig it up...
Manager (Claude Opus 4.5): Global event loop that wakes up specific agents based on folder (Kanban) state.
Product Owner (Claude Opus 4.5): Strategy. Cuts scope creep
Scrum Master (Opus 4.5): Prioritizes backlog and assigns tickets to technical agents.
Architect (Sonnet 4.5): Design only. Writes specs/interfaces, never implementation.
Archaeologist (Grok-Free): Lazy-loaded. Only reads legacy Java decompilation when Architect hits a doc gap.
CAB (Opus 4.5): The Bouncer. Rejects features at Design phase (Gate 1) and Code phase (Gate 2).
Dev Pair (Sonnet 4.5 + Haiku 4.5): AD-TDD loop. Junior (Haiku) writes failing NUnit tests; Senior (Sonnet) fixes them.
Librarian (Gemini 2.5): Maintains "As-Built" docs and triggers sprint retrospectives.
You might ask yourself the question “isn’t this extremely unnecessary?” and the answer is most likely _yes_. But I never had this much fun watching AI agents at work (especially when CAB rejects implementations). This was an early version of the process that the AI agents are following (I didn’t update it since it was only for me anyway): https://imgur.com/a/rdEBU5I
reply