50% to 70% productivity gains, but no AI test runs without a human at Emerson’s NI 

Baskar Ceri does not oversell what generative AI can do for his engineers. As Managing Director for India at Emerson’s Test and Measurement business, formerly National Instruments, Ceri has spent the past year watching the company’s AI advisor, Nigel AI, move from an internal pilot to a product used by test engineers across semiconductor, automotive and aerospace accounts. In conversation, he returns more than once to a single caveat. Nothing Nigel produces goes into production without an engineer signing off on it first.

Context, not just chat

Nigel AI was rolled out internally around May 2025, built on large language models and layered with what Ceri calls three or four decades of National Instruments’ own domain knowledge in test and measurement. His pitch for why that matters is a comparison with generic AI chatbots. “Ask a general purpose model to help design a test, and it might take eight or nine questions before it lands on something usable,” he says, adding further, “What we have put into Nigel AI is the context, all the knowledge we have built internally. By the second prompt it will give you an appropriate answer, this is the spec you are testing and this is what you need to validate.”

The underlying model is still an off the shelf large language model, currently from OpenAI, though Ceri suggested that could change. “Right now it is an LLM. But I think in the agentic world we are going into, it is very open, you will be able to pick whichever model you want to use.” At the company’s NIConnect event earlier this year, NI demonstrated a version of Nigel AI generating Python code and workflows independent of its own LabVIEW software, a signal that the tool’s ambitions extend beyond its own platform.

From data silos to a single thread

A recurring theme in Ceri’s answers is the historic separation between design, validation and manufacturing teams, each typically running on different data platforms and rarely comparing notes. He argues that generative AI only becomes useful at scale once that data is stitched together, which is where NI’s SystemLink software, used to track test assets across factories, comes in.

He offers one example from an automotive parts maker producing horns across roughly 20 manufacturing plants. Identifying which production batches had drifted outside tolerance used to be a manual exercise spanning months. “Using Nigel AI on this, it was able to find within minutes that this batch, this production run, was completely out of range, either lower or higher,” he says. Asked whether the company tracks formal before and after metrics, Ceri is candid that the evidence so far is largely anecdotal. “We have specific customer examples. This is not a broad KPI. We have seen productivity gains of 50% to 70% in some targeted areas.”

An engineer must always sign off

Questioned about the risks of autonomous AI agents, including a recent, widely reported incident of an AI agent acting outside its intended sandbox at another company, Ceri is firm that Emerson’s approach keeps a person in the workflow at every stage. “We call this engineer in the loop. An engineer needs to accept it, validate it, verify it, and then run it, even when an agent has done the work.” On data governance, he reveals Nigel AI is built primarily on internal engineering knowledge rather than customer data. Separately, he points to the European Union’s Cyber Resilience Act, which takes effect in December, as an area where he believes the company’s hardware traceability is ahead of much of the industry.

Notably, when asked whether Nigel AI is self-learning, Ceri answers simply “yes”, a framing that sits somewhat awkwardly against NI’s own public documentation, which states customer data is not used to retrain the underlying model.

Productivity, not necessarily lower cost

Pressed on whether AI would bring costs down, in the way cloud computing was once pitched, Ceri pushes back gently on the framing. He argues the saving shows up at the level of a whole project rather than a unit of output, largely through fewer engineering hours needed at the front end of a design cycle. He is also candid about the company’s own missteps in code generation. Customers began asking for AI powered coding tools as the broader “vibe coding” trend took hold, he said, and NI’s early release fell short. “We were way off. But with the new version we have now, you can actually generate code based on your problem, picking up from previous examples.” He argues that as agentic tools mature, engineers will be able to check tens of thousands of test conditions instead of thousands, pushing manufacturing quality benchmarks from around 99% toward what he described as 99.9999%.

Five percent employable, and counting

Ceri did not dispute that India faces a shortage of AI ready engineering talent. “That has been a problem not just with AI. Even before, we produced a lot of engineers, and only about 5% were considered employable.” He points to partnerships with IITs, including a lab set up to widen access, and a tie-up with Tata Technologies that reached 627 engineering colleges last year, as ways the company is trying to narrow that gap through hackathons and system design contests offering students access to LabVIEW.

One brand, several silos inside

On Emerson’s ownership, Ceri describes NI as operating as a largely independent business group within Emerson, with its own go to market strategy separate from Emerson’s other divisions, such as its process and energy focused business based out of Pune. He estimates customer overlap between the two at around 5%  globally. Any intellectual property Nigel AI and its successors generate, he confirms, stays with Emerson’s Test and Measurement group. The next milestone is agentic workflows for customers, expected around the middle of 2027, extending Nigel AI from code suggestions to spec-to-manufacturing planning, still, he insists, with an engineer approving every step.