Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsMycelium is a software workflow harness designed to put product discovery ahead of AI-assisted implementation. Its creator, Håvard Bartnes, ran it on the project itself and found that only four of 912 logged decisions preceded the first source file; just one of those four cited outside evidence, and no ideas were recorded as killed before code. Those are the tool’s counts of his own process—not evidence that Mycelium improves software outcomes.
What Mycelium is—and what it is not
Mycelium is an open-source workflow for software builders using AI coding agents. It is not a physical device or a code editor. Bartnes describes it as “a harness in front of the coding agent”: the aim is to make the builder and agent answer questions about the problem, audience, assumptions, and evidence before implementation begins.
As an Amazon Associate I earn from qualifying purchases.
The project README presents it for solo builders and small teams working on software, online courses, AI tools, and services. Its central questions are practical: What problem exists? Who has it? Which assumption is riskiest? What small step could test that assumption? The premise is that an agent can produce code quickly, but speed is not useful if the team has not established what is worth building.
Sources: Bartnes’s account, published September 18, 2026; the Mycelium README.
How the discovery-first workflow is meant to work
Bartnes frames the path to implementation as purpose, strategy, opportunity, specification, and then code. The project documentation describes decisions stored in plain YAML and versioned in git. Claims are meant to carry a confidence level and a source, so an assumption is distinguishable from something supported by evidence.
- Clarify purpose: State the problem and the people who experience it rather than starting with a feature idea.
- Set strategy: Make the intended direction and the choices behind it explicit.
- Examine opportunities: Identify assumptions, especially the riskiest ones, and look for a small way to test them.
- Write a specification: Turn the chosen opportunity into a more concrete account of what should be built.
- Begin implementation: Bring in the coding agent after the preceding questions have answers.
This is a gate before delivery, not a claim that every project can be fully understood in advance. The README says work can move back a step when implementation exposes a bad assumption. That makes the intended process iterative: evidence from building can prompt a revision to the opportunity or specification.
What the self-audit counted
After 126 sessions using Mycelium on his own project, Bartnes ran a script over its logs and canvas files. He reported that the first source file was created on May 2, 2026, while only four of 912 logged decisions came before it. One of those four cited outside evidence, and the log contained no ideas recorded as killed before code existed.
| Self-audit measure | Bartnes’s reported result |
|---|---|
| Sessions | 126 |
| Logged decisions | 912 |
| Decisions before the first source file | 4 |
| Early decisions citing outside evidence | 1 of 4 |
| Ideas recorded as dropped before code | 0 |
| First source file | May 2, 2026 |
These figures describe one creator’s project and depend on how he defined and logged activity. A “decision” was a dated log entry with alternatives; “outside evidence” meant a source beyond Bartnes himself; and a “kill” was an idea recorded and then dropped before code existed. As he cautions, the script can count entries and dates, but cannot determine whether a decision was good. The counts therefore show a gap between the process’s intended discipline and how this project actually unfolded; they do not measure product quality or demonstrate that the harness changes outcomes.
Who may benefit—and who may find it cumbersome
Mycelium’s likely fit depends less on whether a project uses AI than on the risk and uncertainty surrounding the product decisions.
- It may be useful when the user need is uncertain, building the wrong thing would be costly, and discovery decisions are otherwise scattered or implicit.
- It may add unnecessary ceremony when the project is low-risk, the need is already clear, or a team has a reliable discovery practice that the harness would duplicate.
- It is a weaker fit for concurrent, cross-role work where multiple functions need to edit shared decisions at the same time. The README identifies that organizational use case as outside the current design.
A useful decision test is to ask: How uncertain is the user need? What would a wrong build cost? How many roles must edit the same decisions concurrently? Does the team already have a discovery habit it trusts? If uncertainty and the cost of error are high, and decisions need more structure, a discovery harness may be worth trying. If discovery already works well or coordination needs exceed the tool’s design, its process could be more burden than benefit.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Setup and compatibility
Bartnes says Mycelium runs in Claude Code, opencode, and Mistral Vibe. The repository documents a Claude Code setup, so readers considering another agent should check the current README for its instructions rather than assume identical installation steps or ongoing compatibility. The repository and agent integrations may change over time.
The project is open source; its GitHub repository is the relevant destination for current setup details and project files: github.com/haabe/mycelium.
What the evidence does—and does not—establish
The available account is written by Mycelium’s creator, and the README is first-party project documentation. Neither is an independent evaluation of whether the tool helps users choose better ideas, ship more successful products, or save time. Bartnes also reported 124 pre-registered tests and a correction log with 300 entries in his September 18, 2026 account; those are project-process figures, not evidence of product effectiveness.
The strongest supported conclusion is narrower: Mycelium formalizes a discovery-first workflow and can count certain kinds of logged activity. Its self-audit exposed how little of Bartnes’s decision logging came before implementation, but it cannot establish whether using the harness produced better decisions. Readers should judge it as a process tool to assess against their own discovery needs, not as a proven outcome-improvement system.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
