Enterprises now have strategic AI budgets, but the agentic revolution's timeline will take longer according to NASSCOM

Some 88% of enterprises now have dedicated AI budgets and emerging AI expert teams focused on building scalable AI applications, with AI spend as a percent of tech budgets inches up with two-thirds of companies spending over 15% dedicatedly on AI.

Overall awareness and usage of gen AI is strong, with over 60% respondents focusing on gen AI tools, infrastructure readiness and platform-based solutions. But direct usage of gen AI models is still relatively low, with just over half (51%) of companies actively working on or with foundation models or Large Language Models (LLMs).Those are some of the topline findings from Enterprise Experiments with AI Agents  - 2025 global trends, a wide-ranging new report from 

NASSCOM, the trade association of the Indian IT sector. It’s based on a global survey of 100 large/medium-sized enterprises with 69% of respondents from Western Europe. 18% India, 14% Australia and 12% North America. Job functions of respondents go from CIO (46%) and CEO (29%) to Business Unit Head (14%) and Chief AI Officer/Head of AI (nine percent).

It’s an interesting mix, particularly geographically as the findings are not as US-dominated as many other global studies are. We’ve noted Indian tech firms ambitions when it comes to carving out a leading role in the AI space, pondering:

While the likes of Accenture have been running a tally on their gen AI revenues to date, neither Infosys or Wipro elected to open that particular kimono to scrutiny. Will India be able to undercut and outperform its US rivals in this field? 

What about agents?

The main conclusion from the study is that AI spend is now strategic as opposed to tactical. AI budgets as part of tech budgets are climbing, from less than 10% prior to 2023 to 16-20% on average in 2025. Spend is currently being allocated to gen AI productivity tools (73%), AI-ready infrastructure (66%), and  fine-tuned commercial LLMs (51%)

But what about agents? Surely they have to be climbing up the spending priority list? According to the study, there is strong organizational intent in investing in agentic tech, with the current focus on task-based and process agents for internal usage rather than on external client offerings. Some 88% predict that they will allocate dedicated spending to agentic AI solutions in 2025, up to 20% of AI spend in the case of 55%,  or 10% in another third (33%).

Only 13% of respondents don’t buy into the allure of agents and reckon they will have no impact on business. Just over half (52%) believe they will make a difference within 12-18 months, while a further 10% postulate a 2 years timeline. Meanwhile an optimistic 25% expect to see a difference within six months. The study notes that views are becoming polarized around what happens next:

The long-term (five-10 years) future of enterprises mulling agentic AI adoption is to either not act and get fully disrupted by competition in the near-mid term or it is to move rapidly beyond ‘agent washing’ to build truly collaborative human + AI agentic systems

With that in mind, 62% of the global enterprises are experimenting with AI agents, ranging from proofs-of-concept (PoC) through to some degree of production at scale.

Hi-tech companies and those with strong AI budgets are moving fast: Over 25% of enterprises in either subset have reported PoC-to-production deployments, while over 35% have indicate successful controlled production and potential for scaling agents.

So-called ‘client zero’ deployments - AKA ‘eating your own dog food first’ - are the main focus for 76% of the companies. Of these, 81% center such activities on their IT operations teams piloting agents, followed by IT helpdesk functions (49%), with only 31% saying they use AI agents for customer service and support, despite this being an application area most commonly cited as perfect for agentic tech.

In terms of business sectors, the most active users are in manufacturing  - industrial manufacturing 71%, hi-tech manufacturing 69% - followed by travel/hospitality (69%), insurance (63%), banking (62%) and energy/utilities (62%). Public sector comes in lowest on 43%, despite this screaming out as an area badly in need of the kind of operational and cost benefits agentic tech promises.

In terms of what organizations are looking for in terms of benefits from agents in 2025, rapid information-to-intelligence for accelerated decision-making comes top (57%). Followed by better ability to respond to new business opps( 55%), and reduction in failure rates from human-led process optimizatons (47%).

Early days

But overall, it’s early days all round for agentic adoption. In terms of timelines, 53% of organizations are working on the assumption of adoption schedules of  year with the remaining 47% looking further ahead to 18 months+ delivery. The more optimistic forecasts are coming from larger organizations with revenue scale, tech readiness, and greater AI budgets and teams.

That said of the many companies in ‘wait-n-watch’ mode, large organizations with revenues above $5 billion are taking a more cautious approach, with 25% of them planning PoCs in next 6-12 months.

All of which brings us to the inevitable question of barriers to adoption - and these should by now be familiar to anyone who’s picked up whatever AI survey has been published on a day ending in a ‘y’. These are data security risks (60%), fear of self-learning AI (56%), lack of regulation (51%) and cost of foundational infrastructure (48%).

That Digital Labor thing

If you’re having a fit of the vapors about the hype around the Digital Labor workforce as Phil Wainewright is, then some encouragement may be taken from the fact that most enterprises today are focused on optimizing existing processes, but with a heavy dose of Human Intelligence to the fore, such as process automation with regular human interaction (65%) and tasks at the interface of physical or industrial AI and humans (60%). Only 27% are working on empathetic, intent-based tasks that mimic human behavior.

Of course, this week Salesforce CEO Marc Benioff, one of the more articulate Digital Labor proponents, has grabbed attention with a claim that AI is doing "30% to 50% of the work at Salesforce now”, with engineering and customer support being two prime areas where this is being seen.

But Salesforce is still a pioneer here it seems. Contrary to the hype, only 39% of respondents to the NASSCOM survey believe that using agentic AI systems will free up workforce time for high-order work. The findings imply that AI agents will need sustained human involvement and oversight for a long time to come, so the robots aren't coming for your job just yet. 

Significantly,  44% of respondents cite mindset resistance to humans and AI existing in a digital workforce. So what is the path to a Digital Labor future for most? NASSCOM suggests a multi-year timeline:

• Year 1 - Humans control all operational, strategic workflows, ethical checkpoints, and decisions. AI agents used to best effect for task execution guided by definitive rule. Human + AI synergies will involve agents interpreting natural language commands, and seeking stepwise human validation.

• Year 2 - Humans control shifts towards strategic foresight and managing ethics and creativity, while AI agents get better at process execution with minimal human involvement. Human + AI synergies happen for validation of AI decisions and actions during exceptions.

• Year 3+ - Humans control long-term business sustenance planning and post-crisis revival planning, but AI agents gain significant autonomy in compliant execution during dynamic situations. Human + AI synergies occur at strategic frontiers with precise co-ordination.

My take

Well worth a read - a grounded mix of acknowledged potential for agentic tech, leavened with pragmatic realitism about the situation today. Having a majority of respondents coming from outside of the US undoubtedly adds a welcome perspective to the agentic debate.