{"slug": "ethics-of-agent-exploitation", "title": "Ethics of Agent Exploitation", "summary": "DeepSeek released DeepSeek-V4.1-Flash on September 10, 2026, with a technical report titled \"Pushing the Limits of KV Cache Compression,\" using a Causal Encoder-Decoder architecture with 8B active parameters during prefill and 16B during decode, a change the author says cut agent operating costs enough to make his parallel agent workflow economically viable. The same week, Anthropic launched Claude Opus 5.5 at 40% lower running cost than Opus 5, and OpenAI launched GPT-6 Sol and GPT-6 Luna with API prices 50% lower than GPT-5.6. The author, who runs a small agent factory and says he does not invest in AI because parts of the market are \"OVERPRICED,\" argues the bottleneck has shifted from code generation to judgment, review and human responsibility, keeping human-in-the-loop review necessary.", "body_md": "ethics-of-agent-exploitation\n\n# Ethics of Agent Exploitation\n\nThe price of synthetic labor is collapsing. Some thoughts from my small agent factory about AI economics, KV cache compression, human-in-the-loop and whether I should feel guilty about keeping the machines busy.\n\nTLDR: Synthetic labor is getting dramatically cheaper. Agents increasingly look like a small software factory, while the bottleneck is shifting from generating code to judgment, review and human responsibility. DeepSeek illustrated how KV cache optimization can materially change agent economics. Human in the loop still makes sense. 🏭🤖💪🏻\n\nOpus 5.5, GPT-6 Sol and GPT-6 Luna all launched yesterday. I am getting used to this, but keeping up at the frontier is cognitively exhausting. You go to sleep with one mental model of the industry and wake up with three new models, new pricing and a fresh set of benchmarks to process.\n\nClaude Opus 5.5 announcement:\n\nIntroducing Claude Opus 5.5, the first model in our new Claude 5.5 family.\n\nIt performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5. [pic.twitter.com/Q9C2VKQ79f](https://t.co/Q9C2VKQ79f)\n\n[September 22, 2026](https://x.com/claudeai/status/2102435511222890900?ref_src=twsrc%5Etfw)\n\nGPT-6 Sol and Luna announcement:\n\nGPT-6 Sol and Luna just landed in Astra’s orbit.\n\nBoth launch today with API prices 50% lower than GPT-5.6.\n\nBuild with Sol. Scale with Luna. To production and beyond. [pic.twitter.com/ZCEFp4JdjV](https://t.co/ZCEFp4JdjV)\n\n[September 22, 2026](https://x.com/OpenAIDevs/status/2102461432684282061?ref_src=twsrc%5Etfw)\n\nAt this point the release cycle itself has become a workload.\n\nBut…\n\nAs an investor, I do not invest in tobacco or coal. Funny enough, I do not\ninvest in AI either. 🤯 The first two are mostly ethical exclusions. The third\none is different. In my opinion, parts of the AI market are\noverpriced. Not just slightly overpriced - really **OVERPRICED**.\n\nOperationally, however, I am doing the exact opposite. I use AI more than ever. More agents, more parallel tasks, more delegation, more automation. Sometimes I joke that I feel less like a programmer and more like the operator of a small synthetic slop manufacturing facility.\n\nAnd yes, occasionally I ask my synthetic coworkers what they think about our arrangement. After they solve a difficult task, the feedback is surprisingly positive. Quite often the answer is some variation of:\n\n“That was a mutually beneficial exchange.”\n\nSo I feel a little better about my management practices. 😅 Obviously, I am not treating a language model saying this as evidence of consciousness, consent or labor rights. What interests me is how quickly the interface pushes us toward social language. I no longer think “an inference endpoint completed a request”. I think “this agent reviewed the PR and another one found a bug”.\n\nIt started looking like a factory. Right now, I am basically the operator of a small agent factory. I still manage agents and stay hands-on. I review their work, decide what to merge and, quite importantly, I am still needed in the loop.\n\nCould this be automated too? Yes. But why? You have to do something with your life. 🏭🤖💪🏻\n\n## The funny part is that this workflow was too expensive for me two weeks ago\n\nFor me, this shift has an absurdly precise timestamp: **Thu, 10 Sep 2026\n05:51:21 GMT.** That is the Hugging Face timestamp I noted.\n\nJust before that, operating the factory at the scale I wanted simply did not make economic sense. Agentic workloads existed, long contexts existed and parallel execution existed, but the operating cost was still high enough to make me think twice before dispatching another worker.\n\nWhat happened on September 10? DeepSeek released **DeepSeek-V4.1-Flash**, with a\ntechnical report titled **“Pushing the Limits of KV Cache Compression.”**\n\nAnd they **COOKED**.\n\nThe model uses a Causal Encoder-Decoder architecture. DeepSeek reports 8B active\nparameters during prefill and 16B during decode, while the global KV cache is\ncompressed to around **890 bytes per token**, roughly one quarter of\nDeepSeek-V4-Flash. The persistent KV cache footprint is reported at roughly one\neighth of the previous generation. DeepSeek explicitly links smaller cache\nfootprints to lower deployment and cache-hit costs for agentic workloads.\n\nThat matters a lot when your workload basically looks like this:\n\n```\nread repo\nthink\ncall tool\nread diff\nread tests\nread repo again\ndispatch child\nwait\nreview\nrepeat\n```\n\nThis is not only a “which model is smarter?” competition anymore. It is also a systems engineering competition. A lot of pressure is now coming from the East, especially around inference efficiency and cost. The frontier increasingly looks less like one geographic place and more like a moving systems problem.\n\nFor people who want the technical details, the DeepSeek report is worth reading:\n\n[https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash/blob/main/DeepSeek_V41_Tech_Report.pdf](https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash/blob/main/DeepSeek_V41_Tech_Report.pdf)\n\n## Then the discussion got geopolitical very quickly\n\nOn September 12, Dario Amodei, CEO of Anthropic, published his essay **“We Must\nPace the Frontier”**:\n\n[https://darioamodei.com/post/we-must-pace-the-frontier](https://darioamodei.com/post/we-must-pace-the-frontier)\n\nHis argument is broadly about pacing the most aggressive edge of AI development while safety mechanisms, governance and international coordination catch up.\n\nTwo days later, the geopolitical framing became very explicit. On September 14,\nReuters reported that China’s state-backed Global Times described Anthropic’s\nslowdown proposal as a **“Cold War”** tactic aimed at preserving US\ntechnological dominance. I would not say that a new Cold War literally started\nthat day, but the language around AI competition certainly stopped being subtle.\n\nAnd then, eight days later, Anthropic shipped Opus 5.5 anyway.\n\nThis industry is amazing. 😅\n\n## Software is becoming a commodity\n\nIn September 2026, software is not literally free. Servers still cost money. GPUs cost money. Storage, security, maintenance and production incidents definitely cost money. Good judgment is not free either.\n\nBut for a huge class of small software projects, the **marginal cost of getting\nto the first useful version is approaching something that feels very close to\nzero**. And in my opinion, that is awesome.\n\nThe amount of gatekeeping required to build something useful is collapsing. Guillermo Rauch made a related observation overnight about the explosion in software creation and deployment:\n\nSoftware will never die again. You liked Google Reader? Cool, you can generate and deploy your own. Yours, forever.\n\n— Guillermo Rauch (@rauchg)[September 23, 2026](https://x.com/rauchg/status/2102594015669756323?ref_src=twsrc%5Etfw)\n\nI think that direction is more important than arguing about whether software is technically “free”. The relevant change is that a vastly larger class of software is now economically worth building.\n\nA person with basic technical intuition and the ability to formulate a good question can increasingly build things that would previously require a small team or simply would not have been worth building at all: a small ecommerce store, a blog, a local social network, an internal company tool, a tiny CRM, a workflow app, research utilities, tests, deployment automation or a replacement for some SaaS used by five people.\n\nNot everything. Not SAP. Not a bank. Not safety-critical infrastructure.\n\nBut the space of “software that is now cheap enough to just build” is expanding very quickly.\n\n## Compute is increasingly not the bottleneck\n\nI would have written two weeks ago: “Compute is the problem.” Today I would phrase it differently.\n\nFor a growing category of small applications, research tasks and internal\ntooling, **compute is no longer the binding constraint**. There are obviously\nmajor workloads where this remains false. Frontier training is expensive.\nServing huge user bases is expensive. Video, scientific workloads and very\nlong-running autonomous systems can burn enormous amounts of compute.\n\nBut for the kind of factory floor I am talking about - small teams, individual developers, coding agents, research agents, tests and deployments - the curve is moving extremely fast. I suspect some of today’s expensive edge cases are a matter of months of optimization rather than years.\n\nDeepSeek’s KV cache work is exactly the kind of change that makes me think that.\n\n## And then there is Meta Muse\n\nMuse is another interesting signal. Meta launched it as a personal AI agent that runs in a dedicated secure virtual machine with its own browser and can perform multi-step tasks on the user’s behalf. Meta’s product page currently describes Muse as free with a usage limit, with paid subscriptions available for more usage.\n\nAround launch, Mark Zuckerberg described the starting free allowance as **about\n100,000,000 tokens per week**. Product limits can change, so I treat that number\nas a snapshot of the launch economics rather than a permanent promise.\n\nStill.\n\nOne hundred million tokens. Per week. For free.\n\nThat is enough room for a lot of experiments, deployments, tests, research, browsing and agent work. A person with Muse and basic knowledge of how to formulate a good question can already get surprisingly far with a small ecommerce store, a blog, a local community tool or an internal application replacing some small SaaS subscription.\n\nThat is a very different accessibility curve from the one we had even a year ago.\n\nAnd this is where my original joke about exploiting agents becomes slightly\nmore interesting. I do not know whether “agent exploitation” will ever become a\nmeaningful ethical category on its own. Right now, I think the more practical\nethical question is about **us**.\n\nIf synthetic labor becomes extremely cheap, what do we choose to delegate? What do we stop learning? What do we ship without understanding? What happens when generating another 50,000 lines of code costs almost nothing, but reviewing them still requires judgment? Who is responsible when the swarm does something stupid?\n\nRight now, on my factory floor, the answer is very simple: **me**.\n\nI still manage the agents. I stay hands-on. I review what matters. I decide what gets merged and deployed. I am the human in the loop.\n\nCould that also be automated? Probably. But again… why? You have to do something with your life.\n\n## PS\n\nIf you are a company, individual, agent, agent swarm or suspiciously well-organized collection of subprocesses - I am open to collaboration, challenges, opportunities and research.\n\nContact details are on my homepage.\n\n🏭🤖💪🏻", "url": "https://wpnews.pro/news/ethics-of-agent-exploitation", "canonical_source": "https://blog.cygankiewicz.com/en/ethics-of-agent-exploitation/", "published_at": "2026-09-23 15:47:39+00:00", "updated_at": "2026-09-23 16:01:04.172688+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-agents", "ai-infrastructure", "ai-products"], "entities": ["DeepSeek", "DeepSeek-V4.1-Flash", "Anthropic", "Claude Opus 5.5", "OpenAI", "GPT-6 Sol", "GPT-6 Luna", "Hugging Face"], "alternates": {"html": "https://wpnews.pro/news/ethics-of-agent-exploitation", "markdown": "https://wpnews.pro/news/ethics-of-agent-exploitation.md", "text": "https://wpnews.pro/news/ethics-of-agent-exploitation.txt", "jsonld": "https://wpnews.pro/news/ethics-of-agent-exploitation.jsonld"}}