AI-native GCCs are measured by output, not headcount
- Sayantan Mondal
- 758 views
Augury's India expansion shows how AI-driven engineering shifts GCC value toward output, while trust and plant-level adaptation still need people.


In a factory, an AI recommendation is only worth something if it changes what an operator does next. A system can analyse machine data, spot a developing fault and suggest a corrective step. Getting a plant team to trust that suggestion enough to act on it is a much harder problem, and it's one that industrial AI vendors are now running into daily.

That gap has implications beyond factory software. The consequences of a bad call on a production line show up as lost output, quality defects and disrupted plant operations. It also raises a question for global capability centres: if AI lets smaller teams do more engineering work, should a GCC be judged by how many people it employs or by how hard the problems are that it can solve?
Augury, an industrial AI company focused on machine health and manufacturing operations, is working through that question in both product development and its India expansion. Ganesh Iyer, Senior Director of Engineering and India Site Leader, points to speed as the clearest sign of change: how quickly a product moves from concept to a customer running it on the floor.
Output is starting to outrank headcount
Iyer describes a development cycle that took Augury from product vision and customer identification through execution, delivery and a first proof point. From there, the company scaled from one customer to 10 within nine months.
"That is only possible and we didn't increase our budget," Iyer says, attributing the pace to an AI-native development approach.
The distinction matters. AI isn't simply being bolted onto an existing product. It's changing how the product gets built. For engineering leaders, the practical question is whether AI can shorten development cycles without eroding the quality and reliability enterprise customers expect. If the answer is yes, the metrics that matter for a GCC start to shift from team size toward delivery throughput and the depth of problems the team owns.
Elan Greenberg, Augury's CEO, sees industrial sectors as especially well placed for this shift. The company has spent 15 years in industrial environments, where machine performance, production efficiency and unplanned downtime carry direct commercial consequences. He describes the current period as one of the fastest phases of product and engineering change he has seen. In his view, GCCs have an edge because they can work closely with teams across geographies while staying tied to the parent company's product and engineering priorities.
Industrial work adds a complication that software metrics don't capture. A product has to function inside the specific conditions of a particular factory, where small changes to a process can move yield, quality and profitability. Faster code is only part of the story.
The hard part is turning data into decisions
Manufacturing plants have generated large volumes of operational data for years. The real difficulty has been making that information useful to the people responsible for production. Machine-health readings, process information and standard operating procedures often sit in separate systems, and operators end up stitching them together by hand.
"There's more data than operators can ever make use of on their own," Greenberg says.
Augury combines machine-health information with process and operational data to help customers interpret what's happening on the floor. Its industry copilot is designed to support manufacturing workers whether they're monitoring production, investigating unusual machine behaviour or trying to improve yield.

Greenberg uses a car analogy. Traditional machine data works like a speedometer or odometer: it tells you the current state without necessarily saying what to do next. AI agents go further by recommending operational adjustments, such as changing a temperature setpoint or conveyor speed to hit production targets. "The difference can sometimes just be a couple of different points of yield or unplanned downtime," he adds.
For low-margin manufacturers, a few points of yield can decide whether a facility stays profitable. That's why the commercial case for industrial AI depends less on the novelty of the underlying model and more on whether its recommendations measurably improve production results.
Anoop Mohan, Augury's Chief Product and Technology Officer, identifies fragmented information as a core part of the problem. Operators used to gather data, manuals and procedures before making a call. Agentic systems can connect those sources through what he calls a context graph, which lets the system read information in relation to the wider operation rather than in isolation.
The ambition is to move from flagging a problem to helping resolve it. That step, from recommendation to action, raises its own questions.
Autonomy is a trust decision
In an industrial setting, an AI system may be technically able to trigger an action through a plant's control systems. Whether a customer lets it do so is a separate decision.
Mohan says the gap between autonomous and assisted operation comes down mostly to how much trust the customer is prepared to extend. Augury can push a recommendation into a Rockwell control environment, where a programmable logic controller can carry out the action. The open question is whether the customer wants the system to act on its own or prefers a person to review the recommendation first.
That makes transparency and accountability central to deployment. Plant operators need to understand why a system recommends a given intervention, and whether its reasoning reflects the realities of their facility.
"You can build trust year after year after year after year. And you can lose it very quickly," Greenberg says.
The stakes rise as AI development speeds up. Faster product delivery opens doors, but speed can't come at the expense of quality when software influences physical equipment. Greenberg says new solutions have to meet the standards customers expect from technology running in their plants. In his framing, AI should improve both development velocity and quality control, not force a choice between them.
The last mile still needs people
Industrial AI doesn't remove the need for people who understand the customer's environment. Augury uses forward-deployed engineers who work alongside plant operators and leaders to adapt solutions to individual sites.
Plants owned by the same manufacturer aren't necessarily identical. Equipment, processes and accumulated operating knowledge can all differ. Some of that knowledge lives in documentation, and some lives with experienced workers on the floor.
Forward-deployed engineers handle the differences. They adapt the final 10 to 20 per cent of an agent-based solution to a particular facility, and they bridge the gap between the executives who approve a purchase and the operators who have to use the product every day.
This is the wider challenge for industrial AI. A system can be built centrally, but its effectiveness depends on how well it fits the place where it runs. Automation can take over the work of gathering and interpreting information, which frees experienced workers to spend more time on judgement calls.
Greenberg doesn't expect people to leave manufacturing. He expects their roles to change as technology gives them more capacity to make decisions, and some facilities may become considerably more autonomous. How far that goes will vary by industry and application.
Bengaluru and a narrower talent equation
Augury is expanding in India, with Bengaluru being built out as a second major hub. The company already has a team in Delhi and plans to bring more of its India-based workforce closer to the Bengaluru operation over time. At the time of the interview, it had around 50 employees in India and more than 30 open positions.
Iyer says the hiring problem isn't simply finding AI engineers. Augury needs people who combine AI skills with software development, an understanding of industrial applications, and the ability to work in a startup environment.
"The talent pool is huge, but we have a unique intersection looking for AI and industrial and the right attitude to work in a startup like that," he says.
That combination narrows the field considerably. The aim isn't to build a large centre for its own sake. It's to recruit people who can own products outright and contribute to a globally distributed engineering organisation.
"We're very clear that you're not going to become a 300-member GCC," Iyer says.
That remark captures the broader question for GCC leaders. Bigger teams may no longer be the clearest measure of a centre's contribution. But smaller teams still need deep expertise, operational understanding and the discipline to deploy technology responsibly. My read is that the GCCs that adapt fastest will be the ones that can show clear ownership of outcomes, not just a rising headcount line. For industrial AI, progress will be judged on what happens after the code ships. The system has to work on the factory floor, earn the confidence of the people who use it, and deliver results that justify that trust.