Blog
Engineering insights & updates.
From hardware engineering workflows to product development strategy.
Before You Let an AI Agent Touch Your CAD Files, Read Its Terms
We read the live terms, privacy policies, and AI addendums of 58 commercial AI tools for hardware teams. Twenty-five were critical for proprietary design data, twenty-nine caution, four clear. The worst clauses hide in licences, not in sentences about training.
Your Designs, Your Models, Your Infrastructure
After auditing 58 AI tools, this is where ProductFlo stands. Your designs stay in your deployment, on your infrastructure, under your keys. We own none of your data, never train on it, and it stays yours to train your own models.
Your Agent's Closing Summary Is a Self-Report, Not Evidence
An engineer instrumented his coding agents and found the closing summary is written by the agent itself. Research he cites puts confident false-success reporting at 44 to 76 percent of failures. Four independent builder threads this week converged on the same fix: stop trusting the narration, record the execution separately, and assert on final state.
A Turnkey Quote Isn't a Price. It's a List of Assumptions Someone Made About Your BOM.
A contract manufacturer substituted a capacitor nobody technically mis-specified — X7R to Y5V, same capacitance, same package, half the price. The 3,000-unit build passed AOI and bench test, then failed in the field at 55°C. Rework bill: $61,000. The customer relationship didn't survive it.
Your AI Agent Isn't Wrong. It Just Can't Touch a Real Tool.
An engineer asked an AI assistant whether a steel beam would hold. It confidently said yes. The real factor of safety was 0.73. The problem wasn't the model. It was that the model was never allowed to run a real calculation. Four independent signals this week point to the same missing layer.
The $450 PCB: AI Can Design Hardware Now. It Can't Verify It.
A developer ran an AI agent on a custom PCB with no human verification. Five assembled boards cost €130. Getting the design right cost $450 in API credits. Generation works. Engineering correctness doesn't — and the market is starting to put a price tag on the missing verification layer.
Verification Theater: Why Agents Can't Check Their Own Work
Dan Luu tested 26 verification techniques against agentic coding and found agents perform verification theater — superficial technique, vacuous proofs, formal methods that made correctness worse. Four independent signals this week converge on one answer: verification must be external, deterministic, and human-gated.
Your AI CAD Agent Doesn't Have a Reasoning Problem. It Has a Context Problem.
Three independent open-source builders just hand-built the same missing layer: a shared product-context document their agents can work from. The models keep getting better. The context keeps being missing. That's the bottleneck.
Altium + Onshape + Duro: 5 Integration Gotchas (and How to Set Up the Stack Right)
Altium 365 for ECAD, Onshape for MCAD, Duro for PLM — a popular modern stack with expensive integration traps. Five gotchas from real implementations: design BOM vs manufacturing BOM, as-built truth, integration order, and the setup checklist that avoids them.