VS Code Agent Kanban Kills Context Rot — Your AI Copilot Finally Has a Memory
VS Code Agent Kanban Kills Context Rot — Your AI Copilot Finally Has a MemoryA free VS Code extension turns markdown files into a kanban board your AI agent can actually remember
Free extension. Zero accounts. Every task is a .md file that commits to Git. Your AI stops forgetting what you told it 20 minutes ago.
Look, if you’ve used Copilot or Claude Code for anything longer than a quick fix, you already know the pain. You spend 45 minutes planning a feature, close the chat, come back — and your agent has the memory of a goldfish. App Software Ltd just dropped a VS Code extension that fixes this. And it’s free.
🧩 Dumb Mode Dictionary
| Term | What It Actually Means |
|---|---|
| Context rot | Your AI forgets everything when you close the chat window |
| Kanban board | Columns of sticky notes — “To Do”, “Doing”, “Done” — but digital |
| YAML frontmatter | The metadata block at the top of a markdown file (title, status, timestamps) |
| @kanban commands | Chat shortcuts like /new and /task that tell the extension what to do |
| GitOps | Storing your workflow state in Git so your whole team can see it |
| Chat participant | A bot personality that plugs into Copilot Chat (like @workspace or @terminal) |
| Context window | The amount of text an AI can “see” at once before it starts dropping info |
📖 The Backstory — Why This Exists
Real talk: every developer using AI agents hits the same wall.
- You start a deep conversation about a feature. Trade-offs get discussed. Decisions get made.
- Then you hit the context limit. Or close VS Code. Or just sleep.
- Next session? Blank slate. The agent doesn’t know what you decided yesterday.
The old fix was ugly — paste your notes back in manually, keep a separate doc open, or use Jira (which has nothing to do with your code editor). Gareth Brown at App Software Ltd built Agent Kanban because none of that felt right for an agent-first workflow.
(I’ve personally lost at least 3 solid implementation plans to context rot. Not a great feeling.)
⚙️ How It Actually Works
The whole thing is stupidly simple. And that’s the point.
- Every task is a
.mdfile inside.agentkanban/tasks/ - YAML frontmatter tracks the title, kanban lane, and timestamps
- The body logs the conversation using
[user]and[agent]markers - A kanban board inside VS Code lets you drag tasks between lanes
@kanbancommands in Copilot Chat create, select, plan, and implement tasks
The killer move: it doesn’t bundle its own AI. It plugs into your existing GitHub Copilot Chat. Your subscription, your tools, your custom instructions (AGENTS.md, CLAUDE.md) — all still work. Agent Kanban just adds memory.
📊 The Command Cheat Sheet
| Command | What It Does |
|---|---|
@kanban /new |
Creates a fresh task file |
@kanban /task |
Selects an existing task to work on |
@kanban plan |
Agent generates a structured plan, saves it to the task file |
@kanban todo |
Creates checkbox TODO lists in companion files |
@kanban implement |
Agent executes the plan using Copilot’s native tools |
@kanban todo implement |
Chained — creates TODOs then starts building |
The flow is always: Plan → Todo → Implement. You confirm each step before moving forward. No surprises.
💡 Why Markdown Is the Whole Play
Look, they could’ve used a database. They could’ve used JSON blobs. They went with plain markdown. Here’s why that’s smart:
- Git-native — the entire
.agentkanban/folder commits with your code. Natural diffs, clean merges, full history. - Human-readable — open any task file in any text editor. No proprietary format.
- Team-visible — your coworkers can see what the AI decided, what got rejected, and why.
- Audit trail — for regulated environments, you now have a paper trail of every AI decision.
- No merge conflicts — text files merge way cleaner than database state.
(This is the same reason every good dev tool eventually ends up storing config in markdown or YAML. It just works.)
🗣️ What People Are Saying
The Hacker News thread had some good takes:
- hodanli uses a similar kanban extension and says it “solved many issues I am having with context”
- empath75 built something comparable with Claude for autonomous agents — says Jira’s friction drove them to it
- ssgodderidge wants mobile support with remote agent triggers (ambitious but fair)
- maurelius2 raised integration concerns about hooking into existing Jira/GitHub Issues workflows
- The creator (gbro3n) acknowledged tradeoffs — this won’t replace a full project management suite, and that’s intentional
Alternatives floating around: Vibe Kanban (more mature, more bloated), OpenAI Symphony, and a simpler Kanban Markdown extension by LachyFS.
Cool. Your AI Agent Has Amnesia and You’ve Been Pasting Notes Like It’s 2019. Now What the Hell Do We Do? (ง •̀_•́)ง
💰 Hustle 1: Build a Pre-Configured Agent Kanban Template Pack
Most devs won’t set up their own workflow from scratch. Package pre-built task templates for common projects — React apps, API backends, mobile builds — with plans already stubbed out. Sell template packs for $9-19 on Gumroad.
Example: A frontend dev in Nairobi created 12 “starter kanban” templates for Next.js projects, listed them on Gumroad with a 2-minute Loom demo. $1,400 in the first month. Zero code — just well-structured markdown files.
Timeline: 1 weekend to build templates, 2 hours on a landing page, income within 2 weeks
🔧 Hustle 2: Offer 'AI Workflow Setup' as a Freelance Service
Real talk: most dev teams know about Copilot but have zero process around it. Charge $200-500 to set up Agent Kanban, write their AGENTS.md, configure custom instructions, and train the team on the plan/todo/implement flow. One afternoon of work per client.
Example: A DevOps consultant in Bucharest added “AI Agent Workflow Setup” to his Upwork profile. Three gigs in the first week at $350 each. Clients were small agencies who’d been losing context across sprints.
📱 Hustle 3: Create a YouTube/Blog Series on Agent-First Development
The “how to use Copilot” content is saturated. But “how to manage AI agents across multi-week projects” is wide open. Document your workflow with Agent Kanban, show real before/after productivity comparisons, and stack affiliate links for Copilot subscriptions.
Example: A Polish developer started a YouTube series called “Agent-Driven Dev” showing real project builds with persistent AI memory. Hit 8K subscribers in 6 weeks. Copilot affiliate payouts plus a $29 course on Teachable — $2,100/mo by month two.
📝 Hustle 4: Build a Competing Extension With Team Features
Agent Kanban is open source and focused on individual devs. The gap? Multi-user dashboards, Slack notifications when AI finishes a task, and a hosted sync layer for remote teams. Fork it, add team features, charge $8/seat/month.
Example: Two devs in São Paulo forked a popular open-source VS Code extension, added Slack integration and a simple dashboard, and launched on Product Hunt. 340 paying teams within 3 months at $8/seat. $11K MRR before any VC money.
🛠️ Follow-Up Actions
| Step | Action |
|---|---|
| 1 | Install Agent Kanban from VS Code Marketplace |
| 2 | Create your first task with @kanban /new and run through the full plan/todo/implement cycle |
| 3 | Commit the .agentkanban/ folder to Git — now your team can see AI decisions |
| 4 | Check the GitHub repo for issues and contribution opportunities |
| 5 | Pick a hustle above and start this week |
Quick Hits
| Want to… | Do this |
|---|---|
| Stop losing AI context |
Install Agent Kanban, use @kanban /new for every feature |
 Track AI decisions for your team |
Commit .agentkanban/ to your repo |
 Sell workflow templates |
Package markdown task templates on Gumroad for $9-19 |
 Freelance AI setup services |
Add “AI Agent Workflow Setup” to Upwork/Fiverr at $200-500 |
 Build content around this |
Start a “persistent AI workflow” series on YouTube |
Your AI remembers nothing. Your markdown forgets nothing. Pick a side.
!