{"slug": "i-combined-3-ai-models-using-hermes-agents-mixture-of-agents-heres-what-happened", "title": "I Combined 3 AI Models Using Hermes Agent’s Mixture of Agents! Here’s What Happened", "summary": "A developer combined three AI models using Hermes Agent's Mixture of Agents (MoA) to build a Kanban board from a single prompt. The system sent the prompt to multiple reference models in parallel, then used an aggregator model to synthesize their perspectives before taking action. The experiment resulted in a functional single-page Kanban application after about 14 minutes.", "body_md": "We spend a lot of time asking one question:\n\nWhich AI model is the smartest?\n\nGPT? Claude? DeepSeek? MiniMax?\n\nBut after experimenting with Hermes Agent’s Mixture of Agents (MoA), I started wondering whether that’s the wrong question.\n\nWhat if, instead of choosing one AI model, you could make several models analyze the same problem and let another model combine their best ideas?\n\nIn other words:\n\n3 AI Models → 1 Smarter Decision → Better Results\n\nThat’s the idea behind Mixture of Agents.\n\nAnd I decided to put it to the test.\n\nMost AI agents work with a single model.\n\nYou send a prompt.\n\nThe model reasons about it, uses tools, and gives you an answer.\n\nMixture of Agents introduces another layer.\n\nYour prompt is first sent to multiple reference models in parallel.\n\nEach model independently analyzes the same problem and provides its perspective.\n\nTheir responses are then given to an aggregator model, which reviews the different approaches, combines the strongest ideas, and decides what to do next.\n\nThe reference models act like advisors.\n\nThe aggregator is the decision-maker.\n\nOnly the aggregator can:\n\nSo from your perspective, you’re still working with one AI agent.\n\nBehind the scenes, however, that agent is consulting multiple AI models before making important decisions.\n\nImagine you’re making an important architecture decision.\n\nWould you rather ask one senior engineer?\n\nOr have three senior engineers independently review the problem first?\n\nOne might immediately spot a security issue.\n\nAnother might question whether the architecture will scale.\n\nThe third might find a much simpler implementation.\n\nNone of them necessarily has the complete answer.\n\nBut a decision-maker who can see all three perspectives has more information before choosing a direction.\n\nLLMs work in a similar way.\n\nDifferent models have different strengths, weaknesses, biases, and reasoning patterns.\n\nThe goal of Mixture of Agents isn’t simply to add more models.\n\nIt’s to give the final model access to diverse perspectives before it commits to a solution.\n\nThe basic workflow looks like this:\n\n```\n                  YOUR PROMPT\n                       │\n          ┌────────────┼────────────┐\n          ▼            ▼            ▼\n      AI Model     AI Model     AI Model\n      Advisor      Advisor      Advisor\n          │            │            │\n          └────────────┼────────────┘\n                       ▼\n               Aggregator Model\n                       │\n                 Makes Decision\n                       │\n                       ▼\n             Tool Calls + Final Answer\n```\n\nThe important distinction is this:\n\nReference models advise. The aggregator acts.\n\nThe advisors don’t modify your project or execute tools.\n\nThey analyze the problem independently and provide additional perspectives that the aggregator can use when deciding what to do.\n\nFor my experiment, I configured:\n\nI then gave the agent a practical task:\n\nBuild a Kanban board for a solo YouTube creator.\n\nWhen I submitted the prompt using `/moa`\n\n, Hermes sent the same problem to both reference models in parallel.\n\nEach analyzed the task independently.\n\nOnce their responses were ready, the aggregator received:\n\nMy original prompt + DeepSeek’s perspective + MiniMax’s perspective\n\nIt could then synthesize those ideas and decide how to approach the actual implementation.\n\nThe result?\n\nAfter roughly 14 minutes, the agent had built a functional single-page Kanban application with:\n\nAll from a single initial prompt.\n\nThe interesting part wasn’t just the finished application.\n\nIt was being able to see how multiple models contributed perspectives before the acting model committed to an approach.\n\nOne of the most interesting parts of Hermes Agent’s implementation is the visibility into what the reference models are doing.\n\nWhile the task is running, you can see each reference model independently analyzing the problem.\n\nThis makes it possible to observe:\n\nIt’s a useful reminder that different LLMs don’t always approach problems in exactly the same way.\n\nAnd that’s where the potential value of Mixture of Agents comes from.\n\nYou don’t necessarily need three models to agree.\n\nSometimes the disagreement is the useful part.\n\nOne setting worth understanding is:\n\n`reference_max_tokens`\n\nThis controls how much output each reference model can generate for the aggregator.\n\nGiving every advisor thousands of tokens isn’t necessarily useful.\n\nThe aggregator often needs the key insights, not another complete solution.\n\nLower limits can therefore mean:\n\nHermes recommends 600 tokens as a practical default for concise reference responses, though the ideal value will depend on your task.\n\nAnother feature I found useful is the ability to create multiple Mixture of Agents presets.\n\nThat means you don’t need one universal AI team.\n\nYou could have:\n\nModels selected specifically for software engineering.\n\nModels with different strengths in analysis and long-context reasoning.\n\nModels that provide complementary perspectives on system design.\n\nA completely different combination optimized for analytical tasks.\n\nInstead of constantly asking:\n\n“What’s the best AI model?”\n\nThe more interesting question becomes:\n\n“What’s the best combination of models for this particular problem?”\n\nI wouldn’t enable Mixture of Agents for every prompt.\n\nIf you’re asking:\n\nWhat’s 15 × 27?\n\nThree AI models aren’t going to make the answer three times better.\n\nThe additional reasoning becomes valuable when there are multiple valid approaches to a problem.\n\nSome good candidates include:\n\nThese are problems where a second or third perspective can expose something the first model overlooked.\n\nMixture of Agents isn’t a free intelligence upgrade.\n\nYou’re making additional model calls.\n\nWith two reference models and one aggregator, you’re paying for multiple perspectives instead of relying entirely on one model.\n\nHermes preserves prompt caching, which helps, but the additional reference calls still introduce extra cost.\n\nThe aggregator needs the reference outputs before it can use them.\n\nThat means complex tasks can take longer than simply sending everything directly to one model.\n\nAdding models blindly isn’t the goal.\n\nIf all your reference models have similar strengths and approach problems similarly, the additional perspectives may not add much.\n\nThe real value comes from diversity.\n\nA strong combination might include models that excel at different things rather than simply choosing the three highest-scoring models on a leaderboard.\n\nThe most interesting takeaway for me wasn’t the Kanban application itself.\n\nIt was the change in how we can think about AI models.\n\nMost of the AI industry conversation revolves around competition:\n\nWhich model is #1?\n\nWhich model has the highest benchmark score?\n\nWhich model should replace the one I’m currently using?\n\nBut maybe that’s only part of the story.\n\nThe next step in AI systems might not simply be:\n\nOne smarter model.\n\nIt could also be:\n\nMultiple specialized models working together.\n\nInstead of choosing between Claude, DeepSeek, MiniMax, GPT, or whatever comes next, AI agents can potentially use different models as a team—taking advantage of their individual strengths before making a final decision.\n\nThe question then changes from:\n\n“Which AI model is the smartest?”\n\nto:\n\n“Which combination of AI models makes the smartest decisions?”\n\nAnd I think that’s a much more interesting problem.\n\nHere’s the experiment I’d love to see:\n\nYou get to build your own AI team.\n\nYou can choose two advisors and one final decision-maker.\n\nWhich three models are you picking?\n\nAnd more importantly:\n\nWhy that combination?\n\nDrop your AI team in the comments. 👇", "url": "https://wpnews.pro/news/i-combined-3-ai-models-using-hermes-agents-mixture-of-agents-heres-what-happened", "canonical_source": "https://dev.to/vivek_shetye/i-combined-3-ai-models-using-hermes-agents-mixture-of-agents-heres-what-happened-14c1", "published_at": "2026-07-17 19:45:56+00:00", "updated_at": "2026-07-17 20:29:52.439549+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-agents", "ai-tools", "developer-tools"], "entities": ["Hermes Agent", "GPT", "Claude", "DeepSeek", "MiniMax"], "alternates": {"html": "https://wpnews.pro/news/i-combined-3-ai-models-using-hermes-agents-mixture-of-agents-heres-what-happened", "markdown": "https://wpnews.pro/news/i-combined-3-ai-models-using-hermes-agents-mixture-of-agents-heres-what-happened.md", "text": "https://wpnews.pro/news/i-combined-3-ai-models-using-hermes-agents-mixture-of-agents-heres-what-happened.txt", "jsonld": "https://wpnews.pro/news/i-combined-3-ai-models-using-hermes-agents-mixture-of-agents-heres-what-happened.jsonld"}}