India's manufacturing sector has been stuck for two decades. NITI Aayog's August 2026 report on the sector found it accounts for 17.5% of gross value added, a share that has stayed between roughly 16% and 18% for 20 years. India's share of global manufacturing value added rose from about 1.5% in 1995 to 3.2% in 2023, while China's rose from about 5% to nearly 32%.
Output is growing. Manufacturing production rose 7.3% in July 2026 and 9% in August, according to industrial production figures cited by Policy Circle. The question is whether large companies can raise productivity without waiting for infrastructure and policy to change. Many are betting on AI agents, software that can plan and carry out multi-step tasks, to help.
What the gap looks like
NITI Aayog's 2025 comparison, as reproduced by a manufacturing sourcing guide, put India's output per worker at $10.7 an hour, against $12.4 in Vietnam and $19.8 in China. It also ranked India 38th on logistics in 2023, behind China at 19th.
The August report lists import dependence, fragmented supply chains, infrastructure and logistics gaps, limited domestic value addition, technology constraints and skill shortages as the main hurdles across sectors. Most of those problems lie outside a single company's walls.
How far agentic AI has spread
Indian enterprises are experimenting widely. A Nasscom study published on October 15, 2026 surveyed 200 India-headquartered enterprises and found that 95% had implemented agentic AI in some form, whether as a pilot or in production. About 97% had dedicated budgets for it.
Scale is a different matter. Nasscom said truly autonomous, enterprise-grade agents in production remain the exception, and that most enterprises require 100% human approval before any agentic action. Trust in the technology, it said, is the real constraint on scaling.
EY India's 2026 research found that 24% of Indian organizations have agentic AI actively deployed, according to the firm's survey of 200 enterprises. Nasscom's separate report on AI-first enterprises found that 38% were scaling at least a quarter of their agentic proofs of concept.
Both surveys rely on companies describing their own efforts.
A large-company example: Tata Steel
Tata Steel is the most detailed public case. On April 22, 2026, it announced with Google Cloud that it had deployed more than 300 specialized AI agents in nine months across operations, finance, human resources and the shop floor.
The company described agents that evaluate equipment health and propose maintenance plans, process invoices, classify GST credits, analyze contracts and route customer complaints, including by reading defects from images. It also built a low-code platform, Zen AI, that lets employees create agents without formal technical training.
The announcement gave no figures for savings, downtime or output. A 2025 Deloitte study cited by Inc42 reported a sharp drop in unplanned downtime at Tata Steel from using generative AI with autonomous agents for maintenance, but the figure was not quantified in the report as quoted.
Where the case for agents is strongest
The documented uses cluster in repeatable tasks inside one company: maintenance planning, quality inspection, invoice processing, complaint routing and reporting. Those are areas where a large firm controls its own data and processes.
A commentary on Nasscom's community site says legacy systems are often not ready for agents, and recommends human review for high-value or regulatory-sensitive decisions. That makes early gains more likely in back-office and maintenance work than in decisions that change production plans.
Where the case is weaker
Agents do little about the problems NITI Aayog lists, such as logistics and import dependence. That is this article's analysis, not a finding from the sources.
Costs and failure rates are open questions. Gartner predicted in June 2025 that over 40% of agentic AI projects will be canceled by the end of 2027 because of escalating costs, unclear business value or inadequate risk controls. In August 2026, it also forecast that AI inference costs per agentic workflow will rise more than fivefold through 2028, even as the price of each unit of processing falls. Both are forecasts, not measured outcomes.
Jobs are another question. A Policy Circle commentary argues that manufacturing growth has yet to deliver proportionate employment, and the sources reviewed do not say how agents would change staffing.
What to watch
The tests are measurable. Companies will need to publish results for downtime, yield, cost per unit or cycle time, not counts of agents. Surveys should show how many enterprises move from human approval of every action to approval by exception. And the productivity numbers, such as output per worker and manufacturing's share of GDP, will show whether any gains reach beyond individual firms.
For now, the evidence supports a narrow conclusion. AI agents are being deployed widely in large Indian companies and can plausibly improve defined operating tasks. Whether they can close the country's industrial productivity gap is not yet shown.
Nexuswild welcomes factual corrections. Email contact@nexuswild.com with evidence and the article URL.
