Blog & Insights
Discover expert perspectives on expertise, knowledge management, technology and professional growth
AI in Engineering
Contradiction detection: why hybrid approaches outperform AI alone
AI can identify candidate contradictions in requirements, but language understanding alone produces too many false positives for engineering quality gates. This is why hybrid approaches , combining semantic search, deterministic validation, and structured evidence is the only path to precision at scale.
Most AI-Powered Engineering Tools Share One Hidden Dependency
Foundation model capability is a fast-moving market, which is good for buyers except for the part of a product built directly on top of it. For tools that analyze engineering specifications, that part matters more than it looks.
Data Format Selection for Multi-Agent LLM Systems: An Empirical Analysis of Token Efficiency
We bet on TOON to cut token costs across our agent network, then watched it collapse on deeply nested data. JSON held up where TOON and CSV broke down, but no single format won everywhere. The real lesson: match the format to the data, not the other way around.
Business
The Requirement Collision: Why SDV Complexity Demands a New Kind of Detective
Weak foundations create invisible cost that only becomes obvious when it is too late to avoid. Software does not infer intent. It executes instructions exactly as written, even when those instructions are incomplete or contradictory. And despite dramatic advances in validation capability, modern vehicle programmes still discover critical issues late. Deterministic analysis allows ambiguity, duplication, and conflict to be surfaced early, while intent can still be clarified and changes are still cheap.
When Simulation Runs Faster Than Understanding
From Clay to Code (Part 3) Simulation is no longer a support activity. It is a primary validation engine. Crash, ADAS, powertrain, calibration, emissions, and software verification are now explored virtually at a scale that replaces large portions of physical development. Yet despite this progress, familiar problems persist. Programmes still slip. Integration remains painful. Quality risks are often discovered later than anyone would like.
Software is obedient, not psychic
The Delay Starts Before Software. (Part 2 - From Clay to Code) The cost of vague requirements is invisible until the worst possible moment: integration, validation, and certification. That is when the bill for "moving fast" arrives. AI has made this easier to explain. Vague prompts lead to "hallucinations" or garbage output. Structured prompts with constraints lead to excellence.
From Clay to Code
Bringing Craftsmanship into Digital Automotive Experiences with Software. For decades, the heart of automotive design has lived in clay. Today, the digital experience is the car. Software defined vehicles, centralised compute, AI and over the air updates now shape how drivers connect, navigate, charge, and how they feel the brand over time. That digital layer deserves the same studio mindset, not a rushed afterthought.