Blog

Practitioner notes. No hype, no vendor pitches — just what we've built and learned.

20 May 2026

The Limits of Pure Hierarchical Agent Orchestration

Why evidence-based safety assessment needs controlled cross-checking between specialist agents.

28 April 2026

Developing a Non-Deterministic AI Tool According to a Safety Standard

A practical ISO 26262 argument for qualifying non-deterministic AI tools: Clause 11 as the tool qualification frame, a selected Part 6 and Part 2 development process, Clause 12-style evidence where it helps, and TR 5469 as AI-specific support.

25 February 2026

Which AI to Use for ISO 26262 Documentation — And What It Actually Costs

Choosing AI infrastructure for automotive safety documentation is different from normal AI procurement. This article covers deployment options (standard API, enterprise tier, private cloud, self-hosted), model comparison for safety work, real cost estimates, and what we actually run.

25 February 2026

From RAG to Agents: Improving AI-Assisted Safety Documentation Workflows

We promised a follow-up on agentic workflows. Here it is. After two years of RAG for ISO 26262 documentation, we rebuilt the system around multi-agent orchestration. In our internal evaluation set, completeness before human review improved significantly, and engineer review shifted from cleanup toward quality judgment. This is what we built, what improved, and what still does not work.

31 July 2024

We Built an AI-Assisted Workflow for Drafting ISO 26262 Safety Case Material. Here's What We Learned.

We built a working AI-assisted workflow for drafting functional safety requirements, technical safety requirements, and safety plan material for expert review. Not a generic compliance claim, not autonomous approval — a prototype tested against real documentation. Here's an honest account of what works, what surprised us, and where human judgment is still non-negotiable.

19 July 2024

Your ISO 26262 Assessor Doesn't Trust Your AI. Here's Exactly Why — And How to Fix It.

Most automotive safety teams are making the same mistake: they bolt AI onto their existing process and hope the assessor plays along. They don't. After 15+ years in functional safety assessments, I've seen exactly why AI-generated safety cases get flagged — and it's not the AI's fault. It's how teams present it. This post breaks down the 5 real objections assessors raise, and what you actually need to do to get them on board. Spoiler: transparency beats perfection every time.