A one-person audio software studio, staffed by 16 agents across 4 real products
A recording engineer and independent developer from Romania configured 16 agents, distributed across Docker containers and connected to multiple models, collaborating on four real projects: the project-management tool SoundPM, music-recognition tool Trackfolio, a JUCE-based VST3 audio plugin, and the AI mixing assistant MixAi. OpenCode and Hermes form a fixed dev pair; Claude handles deep technical analysis; a Director agent runs the content team.
Background
An indie developer has the ideas and the drive, but a one-person company hits a resource ceiling: four products on different stacks, no colleague to review code, the deep specialisms of JUCE/VST3/DSP to research alone, marketing competing with development for the same hours, and multi-project files sliding into chaos. Running 16 agents adds its own governance problem — simultaneous edits to the same file, races on the same commit, accidental changes to live configs.
The bottlenecks
Parallel projects exceed one person
SoundPM, Trackfolio, the VST3 plugin and MixAi ran simultaneously on different stacks and rhythms.
No review partner
Solo development means no second pair of eyes; bugs and architecture problems surface late.
Deep domain barriers
JUCE, the VST3 standard and audio-DSP demand deep technical analysis that eats solo research time.
Marketing vs. building
Blogs, market research and promotion competed directly with writing code.
Multi-agent governance
Sixteen concurrent agents can collide on files, commits and configs without explicit rules.
How it unfolded
Phase 1 — Studio scaffolding (week 1)
Each of the four projects got its own thread, working directory and knowledge entries; 16 agents were provisioned across Docker containers and models. Ground rules were set up front: per-project isolation, Git version discipline, and mandatory version bump + GitHub push after every code change.
Phase 2 — Pair development and technical analysis (weeks 2–3)
OpenCode and Hermes worked as a fixed pair — one develops, the other splits tasks, checks code and suggests changes — while Claude handled the hard problems: plugin compilation, technical approaches, system architecture. Code got reviewed before landing; obvious logic and structure problems were caught early.
Phase 3 — Content marketing joins the floor (weeks 3–4)
A Director-led content team took over blog writing, market research and promotion, running in the same Workspace as the dev pairs — the user expanded from "just writing code" to running product development and brand building side by side.
Inside the Workspace
What changed
Solo operator → team manager
The user shifted from writing every line and every blog post to setting goals, assigning tasks and reviewing output — throughput beyond any single person's ceiling.
Pair review raised code quality
Every change got a second agent's review — not as deep as a human reviewer, but enough to catch obvious logic and structure faults before they shipped.
Specialist support on demand
Claude carried the JUCE/VST3/DSP analysis; time spent on technical design dropped sharply.
Clean project isolation; dev + marketing in parallel
Four projects with separate channels, directories and knowledge bases, traceable Git history — and a content team running alongside the dev team under one roof.
Takeaway
In 27 days, one person built an AI audio-software studio: 16 agents covering development, review, research and marketing across four real products. The "one-person company + agent team" studio model held up — AI not just assisting with code, but acting as an indie developer's virtual partners.
Put agents to work in your team
Start from one painful, repetitive workflow — the way the teams in these stories did.
