01 / NOWIN THE GLASS
One developer.
Many agents.
Git-backed specifications, Kanban tasks, independent clones, parallel local workers, verification and six CLI adapters.
Linux + Bubblewrap today
A planning + execution layer for agentic software development
You had an idea.
Your coding agent picked one of 17 possible meanings.
Kool.ad/e turns fuzzy product intent into a living spec, approved tasks, isolated agent work, and verified code. The important bit is everything in the middle.
Linux x86_64 today · Rust · Git-native · Agent-powered · 0g sugar
Next batch: shared planning repos, Docker-backed execution and remote workers. See the recipe ↓
kool.ad/e
human still in charge
NO CHAT
ARCHAEOLOGY
ARCHAEOLOGY
GIT
REMEMBERS
REMEMBERS
actual software →
AI can write code.
Great. Who told it what to build?
Because “just add saved searches, should be straightforward” is not actually a specification. Somehow.
THE PROBLEM WE KEEP PRETENDING IS SOLVED
Coding got faster.
So did getting the wrong thing built.
Coding agents don't need perfect requirements to start. That's the problem. They'll happily turn ambiguity into implementation before you've noticed the ambiguity was there.
YOU “Add saved searches.”
ALSO YOU “It should be pretty straightforward.”
ONE PR LATER “Wait... why did the agent do that?”
KOOL.AD/E LIVES HERE
The part everyone keeps skipping.
Not another chat box. A durable process for turning intent into work without losing the plot.
01Idea“What if...”
→
02ClarifyFind the sharp edges
→
03Living specGit-backed truth
→
04ApproveHuman says go
→
05BuildIsolated workers
→
06VerifyNo vibes-based QA
Idea → clarification → living spec → approved plan → task stories → implementation → verification. Planning that eventually has the good manners to end.
MEET KOOL.AD/E MAN.AGER
Your slightly obsessive technical project manager.
No relation.
Kool.ad/e Man.ager sits between you, your repository, and the coding agents doing the work. It pulls ambiguity forward, maintains the spec, turns approved plans into ordered work, watches implementation, and asks for human judgment only when it actually needs it.
Less “AI pair programmer.” More “project manager who has read the entire repo and would really like everyone to stop improvising the acceptance criteria.”
Prefers independent task repositories to entering rooms through the wall.
THE BOARD IS THE INTERFACE
One giant chat is not a project plan.
Questions, assumptions, decisions, blockers, implementation tasks, and reviews live as first-class items. Each gets its own focused conversation and context.
Your database migration no longer needs to know what the authentication task was arguing about three days ago.
Questions live where the work lives. Revolutionary stuff.
A CLOSER LOOK
The context. The controls. Right where you need them.
Screenshots use an example workspace. Select an image to see it at full size.
THE NEXT BATCH IS ALREADY ON THE BOARD
Started on one machine.
Planning for a whole crew.
Today, Kool.ad/e helps one developer plan and coordinate isolated agents. The work underway takes it toward shared planning, reproducible Docker execution, trusted remote workers, and agents that can actually test what they build.
The labels mean something. Built is built. Queued is queued. We're not adding imaginary features to the ingredients list.
One developer.
02 / IN FLIGHTMIXING NOW
One plan.
Multiple people.
A separate planning Git repository per workspace, per-record storage, shared sync, task claims and branch-safe coordination.
03 / QUEUEDTHE BIG REBUILD
One policy.
Any trusted worker.
Docker-only execution, consistent CLI capabilities, crash-proof discovery and authenticated remote Linux workers.
04 / FUTUREON THE MENU
Agents that
check the work.
Interactive visual and functional QA, Linux desktop and web drivers, then macOS and Windows host support.
Plus alerts and smarter context
Reality check: Docker, remote workers, multi-operator cutover and autonomous runtime QA are not shipping features yet. Read the full roadmap ↗
Kool.ad/e Facts
Serving size: one dangerously vague feature request
Git-backed living specs100%
Human-gated plans100%
Isolated task repositories100%
Current sandboxBubblewrap
Docker + remote workersQueued
Verification before publishRequired
Telepathic requirements gathering0%
Sugar0g
* Percent Daily Value based on one developer who would rather be writing software than project-managing software.
WHAT'S IN THE PACKET?
Guardrails, not magic.
Kool.ad/e is deliberately boring about the parts that should be boring. Project knowledge lives in Markdown and Git. Implementation happens in independent task repositories. Verification commands are explicit. Publication is a separate decision.
01
Git keeps the history.
The specification keeps the truth.
02
The agent saying “it works” is not a test strategy.
Failed verification blocks publication.
03
No boundary, no build.
Today's Bubblewrap sandbox fails closed. Docker is the planned replacement, not a shipped feature.
CURRENTLY POURING
Useful. Opinionated. Still being invented.
Today
Linux x86_64, Git-backed planning, independent task clones, six CLI adapters.
In flight
Separate planning repositories, multi-user sync, task claims and safer coordination.
Queued
Docker-only sandboxing, CLI parity, reliable scanning and remote Linux workers.
On the menu
Interactive runtime QA, more platforms, better alerts and leaner context.
ENOUGH PLANNING ABOUT THE PLANNING TOOL
Drink it and get S*** done!
Planning software for when “agent, go” is faster than “wait, that’s not what I meant.”
EXTREMELY IMPORTANT LEGAL BEVERAGE INFORMATION
Kool.ad/e is an independent software project. It is not affiliated with, sponsored by, endorsed by, or otherwise connected to any beverage company, powdered drink mix, animated pitcher, wall-destroying mascot, childhood sugar rush, or suspiciously red liquid you may remember from another era.
Any resemblance is completely coincidental. Obviously.