NVIDIA's CEO says future companies will be built on harness engineering. Mine has been for six months — here's the half he left out
On July 8th, Jensen Huang sat down with LangChain's founder and said this: Today most companies are built on business processes. In the future, most companies will be built on harnesses. The harness is the layer you wrap around the model: the domain knowledge you inject, the workflows you've refined, the guardrails, the memory, the evals, the runtime. His argument is that models keep getting…
On July 8th, Jensen Huang, NVIDIA's CEO, spoke with the founder of LangChain, a company specializing in AI-infused software development. During the conversation, Huang said that most companies today are built on business processes, but in the future, they will be built on harnesses. The harness comprises domain knowledge, workflows, guardrails, memory, evaluations, and runtime.
Huang argued that while models continue to improve, it is the outer shell that defines a company, as it contains elements unique to that specific entity.
Jensen Huang's remarks resonated with the author, who had been working on a governance layer for his coffee e-commerce system for six months. The author agreed that the harness is crucial and decided to verify Huang's claims by checking if his system contained all the components mentioned. The author used a simple test to compare Huang's list to what he had actually implemented. The results showed that the author had fulfilled seven out of nine of Huang's components.
However, the author realized that simply having these components does not necessarily make it a harness. The author pointed out that the author's system had close to a hundred skills, but more than half of them had never been called. The skills were present on the disk, but they did not reach the model when needed. The author learned that having the component exists and the component getting used when it should are two different things, and only the second one is a harness.
Another critical insight the author gained was that guardrails block paths, not just failure modes. The author experienced a bug in their PostgreSQL database, which led to the realization that the guardrails were only preventing certain paths from being taken but not the underlying failure mode. The author then learned the importance of verifying the load-bearing thing itself instead of relying on descriptions or specifications.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
