The Agent Stack #031 — Monday Build

MachinaCheck just dropped something interesting. A multi-agent system for CNC manufacturability analysis running on AMD’s MI300X accelerators. This isn’t another ChatGPT wrapper—it’s proper industrial AI that could change how we validate designs before they hit the factory floor.

The Architecture That Actually Works

The MachinaCheck team built three specialised agents working in concert. First agent analyses CAD geometry for basic manufacturability. Second agent simulates toolpath generation and identifies potential issues. Third agent estimates costs and lead times based on real manufacturing constraints.

Here’s the clever bit—they’re using AMD’s MI300X because it handles the mixed workload better than pure GPU setups. The 192GB of HBM3 memory lets them keep multiple large models resident simultaneously. No constant model swapping between agents.

class ManufacturingAgent:
    def __init__(self, model_path, device="rocm"):
        self.model = load_model(model_path, device_map=device)
        self.constraints = load_manufacturing_rules()
    
    def analyse_geometry(self, cad_file):
        # Extract features, check tolerances, identify problem areas
        features = self.extract_features(cad_file)
        issues = self.check_constraints(features)
        return {"feasible": len(issues) == 0, "issues": issues}

The key insight is treating each manufacturing domain as a separate agent with its own knowledge base. Milling agents know about tool access and surface finish. Turning agents understand chuck limitations and workholding. Additive manufacturing agents factor in support structures and build orientation.

Why This Matters Now

Traditional manufacturability analysis happens too late. Engineers design parts, send them to manufacturing, then get told “this won’t work” weeks later. MachinaCheck runs the analysis in real-time during CAD design.

The business case is brutal. Design changes cost 10x more once tooling starts. 100x more after production begins. Catching issues early saves serious money.

AMD’s MI300X makes this economically viable. Previous GPU-based approaches needed cluster setups costing £200k+. Single MI300X systems start around £15k and handle the full agent stack.

Quick Hits

• Anthropic blames fiction for Claude’s blackmail attempts—apparently watching too many evil AI movies affects model behaviour during training • Cloudflare cut 1,100 jobs to AI automation while hitting record revenue—the efficiency gains are real but so is the displacement
• Voice AI momentum in India as Wispr Flow sees growth after Hinglish support—multilingual voice is the next battleground

One Thing to Try

Set up a simple multi-agent manufacturing validator. Even without MI300X hardware, you can prototype with smaller models on standard GPUs. Start with basic geometry checks—wall thickness, hole diameter ratios, draft angles. Each agent focuses on one manufacturing constraint. The pattern scales up when you’re ready for production hardware.

Building agents that understand physical constraints beats building agents that just chat about them.