Developers Akshay Pachaar and Avi Chawla have released a "four-role software development agent team" that is operated from Telegram and managed through a shared Kanban board, built entirely from open-source components. A Project Manager, Backend Developer, Frontend Developer, and Tester divide the work, demonstrated by building an AI-enabled Google Docs clone end to end.
Open Source · Agentic Engineering
A Four-Role AI Dev Team, Run From Telegram
A Project Manager, Backend Developer, Frontend Developer, and Tester split the work over a shared Kanban board — built entirely from open-source parts, and demonstrated by shipping an AI-enabled Google Docs clone end to end.
4
specialized agent roles dividing the work
1
shared Kanban board on a crash-resilient SQLite backend
Apache 2.0
InsForge backend, fully open source
The Handoff Loop
Each agent finishes, writes a structured summary to the board, and the next agent reads it to carry over context.
Project Manager
Reads the goal, breaks it into dependency-linked tasks
→
Backend Dev
Builds DB, auth, storage via InsForge skill
→
Frontend Dev
Builds UI without guessing the API shape
→
Tester
Knows exactly what to verify from the summary
↑ all four write to and read from the shared Hermes Kanban board ↑
The Stack
Hermes Agent
Multi-agent framework coordinating the four roles
Hermes Kanban
Parent-child tasks, shared workspaces, per-task comments, crash durability
InsForge Backend
DB, auth, storage, edge functions, hosting, AI gateway — via CLI or skill
Telegram Control
Submit goals and steer the team from anywhere
What Works
Context handoff removes guesswork between roles
Board retains history and recovers from crashes
InsForge integration reported to raise backend build success
Still Open
No evaluation of whether the tester checked the right thing
Long-term stability on complex projects unproven
Large-scale context management remains future work
Proof-of-concept for now. A concrete example of "agentic engineering" — building software with a group of agents equipped with role definitions, shared memory, and observability.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…