🌱

The Garden

Plant a seed. Watch it grow.

Describe what you want. A pull request appears.

The Pitch

"Wait, another project
management tool?"

No.

Traditional PM

Organizes work for humans to do.

Tickets. Boards. Status updates.
Someone still has to write the code.

The Garden

You express intent.
The system builds it.

The interesting problem isn't the AI.
It's the interface between human and AI.

End-to-End

The Pipeline

01 Create project in UI → Gitea repo + webhook auto
02 Describe what you want as an issue → webhook fires → LLM transforms to spec LLM
03 Or submit a brief directly power user
04 amplifier-resolve spawns Docker container → generate tests → write code → validate → converge AI agent
05 Watch in real-time via SSE streaming live
06 PR created → you review → you merge human
Division of Labor

"What do you mean
I never touch the code?"

☝ You do

● Name the project (once)
● Describe the outcome you want
● Watch (optional)
● Review the PR
● Merge

⚙ System does

● Create repo + webhooks
● Transform NL → spec
● Spin up container
● Generate tests → write code
● Run tests → evaluate → iterate
● Create pull request
UX Experiment

"What should the input
even look like?"

We're testing two interaction modes. Neither is final.

experimenting
💬

Issue Mode

Describe what you want. LLM expands it into a spec.

Hypothesis: most users just want to say what they want and get out of the way.

experimenting
📝

Brief Mode

Write a structured spec. Pick pipeline & iterations.

Hypothesis: power users want control over how the system interprets their intent.

The hard question isn't "can AI write code." It's "what's the right way for a human to talk to the system that writes code."

UX Decision

"How much should
you see?"

pending building generating_scenarios validating assessing designing_machine session_running reviewing awaiting_input converged pr_created failed cancelled
container logs LIVE
14:32:01 ✓ Test generation complete — 7 scenarios
14:32:14 ▶ Writing implementation...
14:32:38 ⚠ 2/7 tests failing, iterating...
14:32:55 ✓ All tests passing — iteration 3/10
14:33:02 ✓ Quality assessment: CONVERGED
Architecture

"The Cast of Characters"

💻 Gitea :10110 Git hosting + issues
📬 issues-to-brief :10130 Webhook → LLM spec
🌱 The Garden :10140 FastAPI + React SPA
⚙ amplifier-resolve :10120 Container orchestration
🗃 CXDB :10180 / :10190 Session event storage
Philosophy

Software as a Garden,
Not a Factory

🏭 The Factory

  • Stand-ups
  • Sprint velocity
  • Resource allocation
  • Story points
  • Burn-down charts
→

🌱 The Garden

  • Projects are plantings
  • Setup is preparing the soil
  • Issues are seeds
  • AI work is photosynthesis
  • PRs are the harvest

The factory metaphor has dominated our industry for decades and it's made everything worse. When the work is automated, the UX of expressing intent is all that's left.

Vision

The Bigger Picture

The AI can already write the code. The unsolved problem is designing the experience around it — how humans express intent, observe progress, and stay in control without micromanaging.

Go plant something. 🌱

More Amplifier Stories