Today marks the start of the AMD Developer Hackathon (May 4, 2026).
Instead of building something from scratch, we’re treating this as a controlled stress test of a structured approach to agent systems.
URL: https://lablab.ai/ai-hackathons/amd-developer
Why this matters
Hackathons are noisy.
Most teams:
- start from zero
- overbuild
- never reach a stable system
We’re doing the opposite:
Bring a structured approach → adapt → measure → iterate
The goal is not just to “build something cool”.
It’s to validate:
- how fast we can adapt to unknown constraints
- how efficiently we can coordinate multiple agents
- how quickly we can reach a working system under pressure
Our Focus: Systems that Execute (not just respond)
We’re participating in Track 1: AI Agents & Agentic Workflows, but the goal is not another “agent demo”.
We’re focusing on:
- agents that coordinate and execute tasks
- real workflows with state, decisions, and side-effects
- systems that can move beyond simple prompt chaining
Architecture Direction
1. Structured multi-agent coordination
- clear roles
- shared context
- controlled execution flow
No unnecessary complexity — everything must justify itself quickly.
2. Hybrid execution (local + cloud)
- Electron → controlled local execution
- AMD Developer Cloud (ROCm) → high-performance compute
Goal:
combine local control with scalable processing
3. Iteration over perfection
define → implement → connect → test → repeatSpeed of iteration matters more than initial correctness.
What we want to learn
This is the real objective:
- Where this approach breaks
- Where it slows us down
- Where it creates leverage
If it works: → it becomes a strong foundation going forward
If it fails: → we refine the approach, not just the implementation
I’ll share results after the hackathon — including what worked, what didn’t, and what we’re changing next.
👉 Part 2 (deep dive): WIP — coming soon