Backstory: I've built digital products for almost 30 years, and I've seen the same story evolve over and over. We built something nobody wanted or asked for. The only "proof" we had, was someone's gut feeling. Finding product-market-fit on a gut feeling is expensive and long-lasting. The constant refactoring and new alignments often meant adding some duct tape, chewing gum, pins and needles keeping it all together. Now I see the same thing happening using AI. At speed. Everybody building what no one asked for at 10x the speed, ending up with the same shaky structures. Because we let the agents continue our own guessing discipline, in an amplified state.
Where this is different: Most other product-focused plugins I've seen focus on the output and the delivery. Mycelium starts further upstream so that you are sure the product you think about building is actually worth building. The focus throughout the loops is to deliver value through outcome, not output. The agent needs to learn what to build and test its riskiest assumption before it gets to write any code.
Real life stories:\n(1) A mathematician at NTNU gave it a spin on Vibe+Mistral, albeit I developed it for Claude Code first and foremost. He created a MiniLisp interpreter with it, and published it on GitHub (https://github.com/dagfinndybvig/minilisp). For him it was more of an educational project, for me a proof of concept. \n(2) Mycelium stopped me from creating a file viewer that nobody but me ever wanted. The facts are available as a receipt in the repo (https://github.com/haabe/mycelium/blob/main/docs/receipts/ca...)
The plugin is still living in its early days. I'm developing it dogfooding it on itself. Any feedback you might have is most welcome, good and bad.
I did post about Mycelium on HN months ago, but I forgot to use the "Show HN" tag in the title. Mycelium has changed a lot since then. The original post is at https://news.ycombinator.com/item?id=47723889
If you're curious, and want to take it for a ride, you can do so by installing it through the plugin's own marketplace. You will need a coding agent, such as Claude Code, and access to a working LLM model.\n/plugin marketplace add haabe/mycelium\n/plugin install mycelium@haabe-mycelium\n/mycelium:start # one command: setup + 10-minute discovery"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["mycelium"],"value":"Show HN: Mycelium \u2013 AI agent plugin guiding you from purpose to market"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["mycelium"],"value":"https://github.com/haabe/mycelium"}},"_tags":["story","author_haabe","story_48785299","show_hn"],"author":"haabe","children":[48944985],"created_at":"2026-07-04T13:35:19Z","created_at_i":1783172119,"num_comments":0,"objectID":"48785299","points":1,"story_id":48785299,"story_text":"What this is: A plugin for your coding agent that guides you through the birth of a product, from purpose to market.
Backstory: I've built digital products for almost 30 years, and I've seen the same story evolve over and over. We built something nobody wanted or asked for. The only "proof" we had, was someone's gut feeling. Finding product-market-fit on a gut feeling is expensive and long-lasting. The constant refactoring and new alignments often meant adding some duct tape, chewing gum, pins and needles keeping it all together. Now I see the same thing happening using AI. At speed. Everybody building what no one asked for at 10x the speed, ending up with the same shaky structures. Because we let the agents continue our own guessing discipline, in an amplified state.
Where this is different: Most other product-focused plugins I've seen focus on the output and the delivery. Mycelium starts further upstream so that you are sure the product you think about building is actually worth building. The focus throughout the loops is to deliver value through outcome, not output. The agent needs to learn what to build and test its riskiest assumption before it gets to write any code.
Real life stories:\n(1) A mathematician at NTNU gave it a spin on Vibe+Mistral, albeit I developed it for Claude Code first and foremost. He created a MiniLisp interpreter with it, and published it on GitHub (https://github.com/dagfinndybvig/minilisp). For him it was more of an educational project, for me a proof of concept. \n(2) Mycelium stopped me from creating a file viewer that nobody but me ever wanted. The facts are available as a receipt in the repo (https://github.com/haabe/mycelium/blob/main/docs/receipts/ca...)
The plugin is still living in its early days. I'm developing it dogfooding it on itself. Any feedback you might have is most welcome, good and bad.
I did post about Mycelium on HN months ago, but I forgot to use the "Show HN" tag in the title. Mycelium has changed a lot since then. The original post is at https://news.ycombinator.com/item?id=47723889
If you're curious, and want to take it for a ride, you can do so by installing it through the plugin's own marketplace. You will need a coding agent, such as Claude Code, and access to a working LLM model.\n/plugin marketplace add haabe/mycelium\n/plugin install mycelium@haabe-mycelium\n/mycelium:start # one command: setup + 10-minute discovery","title":"Show HN: Mycelium \u2013 AI agent plugin guiding you from purpose to market","updated_at":"2026-07-17T09:08:26Z","url":"https://github.com/haabe/mycelium"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"lazybean"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["mycelium","claude"],"value":"Last night I couldn't sleep and started questioning current tech and its future. Is there no more sustainable way?
Do we really want giant concrete buildings sinking megawatts of electricity and millions of gallons of water? For what? Displacing jobs and disrupting society?
I believe AI can be good. But not at that cost.
So what if it was running on mushrooms? (Not hallucinating \u2014 actually running on mycelium.)
This is a research compilation. The shiitake memristor paper is real (PLOS ONE, Oct 2025). DARPA is funding mycelium computing chips. Adamatzky's lab has been publishing in Nature Scientific Reports since 2021. A mushroom field produces food, creates jobs, sequesters carbon \u2014 and might compute.
I don't know if this is the answer. But I think it's a question worth asking.
It turns out, we're not that far (but not that close either). Claude burnt a small forest worth of compute to help me compile what could be done about it."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"What if AI centers were producing food and jobs?"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://lazybean.github.io/a-shroom/"}},"_tags":["comment","author_lazybean","story_47599455"],"author":"lazybean","comment_text":"Last night I couldn't sleep and started questioning current tech and its future. Is there no more sustainable way?
Do we really want giant concrete buildings sinking megawatts of electricity and millions of gallons of water? For what? Displacing jobs and disrupting society?
I believe AI can be good. But not at that cost.
So what if it was running on mushrooms? (Not hallucinating \u2014 actually running on mycelium.)
This is a research compilation. The shiitake memristor paper is real (PLOS ONE, Oct 2025). DARPA is funding mycelium computing chips. Adamatzky's lab has been publishing in Nature Scientific Reports since 2021. A mushroom field produces food, creates jobs, sequesters carbon \u2014 and might compute.
I don't know if this is the answer. But I think it's a question worth asking.
It turns out, we're not that far (but not that close either). Claude burnt a small forest worth of compute to help me compile what could be done about it.","created_at":"2026-04-01T11:33:05Z","created_at_i":1775043185,"objectID":"47599456","parent_id":47599455,"story_id":47599455,"story_title":"What if AI centers were producing food and jobs?","story_url":"https://lazybean.github.io/a-shroom/","updated_at":"2026-04-01T16:28:00Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"flowerthoughts"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["mycelium","claude"],"value":"I've used the analogy of a circular saw before ("it's not really sawing, because you can't feel the wood...")
It's easy to just slab on a Skil saw, cut through the beam and it'll be somewhat straight. But when every manual stroke counts, there's enough time on a human time scale to correct every little mistake. It's definitely possible to become skilled at using the circular saw, but it takes effort that it feels like you don't need at first.
This is similar. LLMs are so powerful for writing code that it's easy to become complacent and forget your role as the engineer using the tool: guaranteeing correctness, security, safety and performance of the end result. When you're not invested in every if-statement, forgetting to check edge cases is really easy to do. And as much as I like Claude writing test cases for me, I also have to ensure the coverage is decent, that the implicit assumptions made about external library code is correct, etc. It takes a lot of effort to do it right. I don't know why Mycelium thinks they invented interfaces for module boundaries, but I'm pretty sure they are still as suceptible to that "0" not behaving as you'd expect, or the empty string being interpreted as "missing." Or the CSG algorithm working, except if your hole edges are incident with some boundary edges.
Edit: spelling"},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["claude"],"value":"Claude built a system in 3 rounds, latent bugs from round 1 exploded in round 3"},"story_url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["mycelium"],"value":"https://github.com/mycelium-clj/mycelium/blob/main/benchmarks/SCALING.md"}},"_tags":["comment","author_flowerthoughts","story_47287350"],"author":"flowerthoughts","children":[47306863,47310449,47329844],"comment_text":"I've used the analogy of a circular saw before ("it's not really sawing, because you can't feel the wood...")
It's easy to just slab on a Skil saw, cut through the beam and it'll be somewhat straight. But when every manual stroke counts, there's enough time on a human time scale to correct every little mistake. It's definitely possible to become skilled at using the circular saw, but it takes effort that it feels like you don't need at first.
This is similar. LLMs are so powerful for writing code that it's easy to become complacent and forget your role as the engineer using the tool: guaranteeing correctness, security, safety and performance of the end result. When you're not invested in every if-statement, forgetting to check edge cases is really easy to do. And as much as I like Claude writing test cases for me, I also have to ensure the coverage is decent, that the implicit assumptions made about external library code is correct, etc. It takes a lot of effort to do it right. I don't know why Mycelium thinks they invented interfaces for module boundaries, but I'm pretty sure they are still as suceptible to that "0" not behaving as you'd expect, or the empty string being interpreted as "missing." Or the CSG algorithm working, except if your hole edges are incident with some boundary edges.
Edit: spelling","created_at":"2026-03-09T09:05:30Z","created_at_i":1773047130,"objectID":"47306509","parent_id":47287350,"story_id":47287350,"story_title":"Claude built a system in 3 rounds, latent bugs from round 1 exploded in round 3","story_url":"https://github.com/mycelium-clj/mycelium/blob/main/benchmarks/SCALING.md","updated_at":"2026-04-21T16:56:34Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"yogthos"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["mycelium","claude"],"value":"I'd argue that another key aspect is to break programs up into small independent units that can be verified in isolation, and to compose them into larger programs with contracts between them. I've had a pretty good experience using Claude with a framework where I express the program as a state graph, and each node is treated like a microservice that gets some input and produces some output. Then the workflow engine verifies that the output matches the declared schema and then decides which step to execute next. https://github.com/yogthos/mycelium
As the state travels across the graph, I keep a trace of the steps which were executed, which means that when an error happens, the agent has a lot more information than it normally would, it can see what decision points the code passed through already, it can cross references that with the declared workflow, and quickly find where it screwed up.
The idea of workflow engines has been around for a long time, but they feel too awkward to use when you're writing code by hand. Writing conditional logic directly in the code keeps you in your flow, and having to jump out and declare it in config somewhere feels awkward. Coding agents completely change the dynamic though because they don't have that problem. If the LLM is writing the code, then I can just focus on ensuring the code meets the contract, while the agent can deal with the implementation details."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Deterministic Programming with LLMs"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://www.mcherm.com/deterministic-programming-with-llms.html"}},"_tags":["comment","author_yogthos","story_47158834"],"author":"yogthos","comment_text":"I'd argue that another key aspect is to break programs up into small independent units that can be verified in isolation, and to compose them into larger programs with contracts between them. I've had a pretty good experience using Claude with a framework where I express the program as a state graph, and each node is treated like a microservice that gets some input and produces some output. Then the workflow engine verifies that the output matches the declared schema and then decides which step to execute next. https://github.com/yogthos/mycelium
As the state travels across the graph, I keep a trace of the steps which were executed, which means that when an error happens, the agent has a lot more information than it normally would, it can see what decision points the code passed through already, it can cross references that with the declared workflow, and quickly find where it screwed up.
The idea of workflow engines has been around for a long time, but they feel too awkward to use when you're writing code by hand. Writing conditional logic directly in the code keeps you in your flow, and having to jump out and declare it in config somewhere feels awkward. Coding agents completely change the dynamic though because they don't have that problem. If the LLM is writing the code, then I can just focus on ensuring the code meets the contract, while the agent can deal with the implementation details.","created_at":"2026-03-01T03:25:29Z","created_at_i":1772335529,"objectID":"47203359","parent_id":47158834,"story_id":47158834,"story_title":"Deterministic Programming with LLMs","story_url":"https://www.mcherm.com/deterministic-programming-with-llms.html","updated_at":"2026-03-05T23:38:41Z"}],"hitsPerPage":8,"nbHits":7,"nbPages":1,"page":0,"params":"query=mycelium+claude&hitsPerPage=8&advancedSyntax=true&analyticsTags=backend","processingTimeMS":9,"processingTimingsMS":{"_request":{"roundTrip":15},"fetch":{"query":7,"total":8},"total":9},"query":"mycelium claude","serverTimeMS":10}