Are firms wasting their money on AI agents?
Leslie D'Monte Most firms are still trying out AI agents, hailed by Big Tech as game changers. Analysts warn that many of these may be scrapped within two years as they hide high costs, uncertain returns, and weak risk controls. Can firms unlock value through human-AI agent teamwork? Google and Microsoft recently called AI agents the next big shift, unveiling Project Mariner, Gemini tools, Azure AI Foundry and NLWeb. Gift this article Why is Big Tech bullish on AI agents?
Unlike basic chatbots, AI agents are autonomous systems that can plan, reason and complete complex multi-step tasks with minimal input—such as coding, data analysis and generating reports. Developers use them across IT, customer support, and enterprise workflows. Google and Microsoft recently called AI agents the next big shift, unveiling Project Mariner, Gemini tools, Azure AI Foundry and NLWeb. Salesforce, Amazon, IBM, and Meta are also building these platforms to automate workflows and enhance productivity. Nasscom has said 46% of firms are experimenting with AI agents, mainly in IT. Also read | Global poverty: How to deal with funding cuts What should firms be wary of?
Many so-called agentic use cases today can be done with simpler tools, says Gartner, which predicts over 40% of such projects will be scrapped by 2027 due to high costs, vague value or weak risk controls. It adds that of thousands of vendors, only about 130 are seen as credible; and many engage in 'agent washing"—rebranding chatbots, robotic process automation (RPA), or assistants as agents without real autonomy. Most current systems also lack the maturity to deliver complex outcomes or sustained return on investment (ROI). Nasscom echoes this, noting 62% of enterprises are only doing in-house agent testing.
Also read | How India's $12 bn R&D push could reshape its tech future How big is the market for AI agents?
The global AI agents market, valued at $5.4 billion in 2024, is pegged to touch $50.3 billion by 2030, per Grand View Research. North America led with 40.1% revenue share, while Asia-Pacific is the fastest-growing region. Machine learning, single-agent systems, and ready-to-deploy agents dominated through tech, system type, and deployment model, respectively. Also read | Shades of grey: Inside the world of pre-IPO trading How can companies unlock real value?
AI agent adoption is hindered by privacy concerns, regulation gaps, and limited focus on talent readiness. Integration with legacy enterprise software remains complex and costly. Regardless, Gartner predicts 15% of daily work decisions will be made autonomously by agentic AI by 2028, up from 0% in 2024. It also urges agentic AI-use only when the returns are clear. Nasscom sees strong potential in real-time decision-making and agility. Key focus areas include data governance and AI risk protocols. Also read | China's economy beats the gloom. Can it do more? Can humans make AI agents more effective?
Though AI agents are designed to be autonomous, Nasscom's recent study of over 100 global enterprises reveals how businesses are transitioning from early-stage GenAI applications towards more goal-oriented, human-plus, AI agentic systems. It believes that to scale responsibly, enterprises must prioritize human-AI collaboration, trust and data readiness. Nasscom adds that most (77%) firms adopt AI agents with a 'human-in-the-loop" approach, reflecting the need for oversight and contextual judgment. Also read | Residential sales: Where have all the buyers gone? Topics You May Be Interested In Catch all the Business News, Market News, Breaking News Events and Latest News Updates on Live Mint. Download The Mint News App to get Daily Market Updates.
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
The Hindu
8 minutes ago
- The Hindu
How artificial intelligence is tackling mathematical problem-solving
The International Mathematical Olympiad (IMO) is arguably the leading mathematical problem-solving competition. Every year, high school students from around the world attempt six problems over the span of three hours. Students whose scores cross a threshold, roughly corresponding to solving five of the six problems, obtain Gold medals, with Silver and Bronze medals for those crossing other thresholds. The problems do not require advanced mathematical knowledge, but instead test for mathematical creativity. They are always new, and it is ensured that no similar problems are online or in the literature. The AI gold medallist IMO 2025 had some unusual participants. Even before the Olympiad closed, OpenAI, the maker of ChatGPT, announced that an experimental reasoning model of theirs had answered the Olympiad at the Gold medal level, following the same time limits as the human participants. Remarkably, this was not a model specifically trained or designed for the IMO, but a general-purpose reasoning model with reasoning powers good enough for an IMO Gold. The OpenAI announcement raised some issues. Many felt that announcing an AI result while the IMO had not concluded overshadowed the achievements of the human participants. Also, the Gold medal score was graded and given by former IMO medalists hired by OpenAI, and some disputed whether the grading was correct. However, a couple of days later, another announcement came. Google-DeepMind attempted the IMO officially, with an advanced version of Gemini Deep Think. Three days after the Olympiad, with the permission of the IMO organisers, they announced that they had obtained a score at the level of a Gold medal. The IMO president Prof. Gregor Dolinar stated, 'We can confirm that Google DeepMind has reached the much-desired milestone, earning 35 out of a possible 42 points — a gold medal score. Their solutions were astonishing in many respects. IMO graders found them to be clear, precise and most of them easy to follow.' Stages of development Even as it became a popular sensation, ChatGPT was infamous both for hallucinations (making up facts) and for simple arithmetic mistakes. Both these would make solving even modest mathematical problems mostly impossible. The first advance that greatly reduced these errors, which came a few months after the launch of ChatGPT, was the use of so-called agents. Specifically, models were now able to use web searches to gather accurate information, and Python interpreters to run programs to perform calculations and check reasoning using numerical experiments. These made the models dramatically more accurate, and good enough to solve moderately hard mathematical problems. However, as a single error in a mathematical solution makes the solution invalid, these were not yet accurate enough to reach IMO (or research) level. Greater accuracy can be obtained by pairing language models with formal proof systems such as the Lean prover — a computer software that can understand and check proofs. Indeed, for IMO 2024 such a system from Google-DeepMind called AlphaProof obtained a silver medal score (but it ran for two days). Finally, a breakthrough came with the so-called reasoning models, such as o3 from OpenAI and Google-DeepMind's Gemini-2.5-pro. These models are perhaps better described as internal monologue models. Before answering a complex question, they generate a monologue considering approaches, carrying them out, revisiting their proposed solutions, sometimes dithering and starting all over again, before finally giving a solution with which they are satisfied. It were such models, with some additional advances, that got Olympiad Gold medal scores. Analogical reasoning and combining ingredients from different sources gives language models some originality, but probably not enough for hard and novel problems. However, verification either through the internal consistency of reasoning models or, better still, checking by the Lean prover, allows training by trying a large number of things and seeing what works, in the same way that AI systems became chess champions starting with just the rules. Such reinforcement learning has allowed recent models to go beyond training data by creating their own synthetic data. The implications Olympiad problems, for both humans and AIs, are not ends in themselves but tests of mathematical problem-solving ability. There are other aspects of research besides problem-solving. Growing anecdotal experiences suggest that AI systems have excellent capabilities in many of these too, such as suggesting approaches and related problems. However, the crucial difference between problem-solving and research/development is scale. Research involves working for months or years without errors creeping in, and without wandering off in fruitless directions. As mentioned earlier, coupling models with the Lean prover can prevent errors. Indications are that it is only a matter of time before this is successful. In the meantime, these models can act as powerful collaborators with human researchers, greatly accelerating research and development in all areas involving mathematics. The era of the super-scientist is here. Siddhartha Gadgil is a professor in the Department of Mathematics, IISc
News18
an hour ago
- News18
Google Pixel 10 Series Could Launch With eSIM Support In Some Markets: Know More
Last Updated: Google Pixel 10 series is launching soon and it seems the company will have different variants for its markets. Google could follow Apple's path by ditching the physical SIM slot for eSIM only support with its new Pixel 10 series. We saw the iPhone 14 launch without traditional SIM support and only work through eSIM and it seems Google could have a similar plan for the Pixel 10 models, at least in some countries. eSIM has surely proven to be more effective and safe than regular SIM usage but the market is still very limited for its support, which explains why Apple and Google have kept its compatibility to select regions. SIM-Free Pixel 10 Launching Soon? The details about this rumour have come via Evan Blass, also known as EvLeaks. He shared a post on X where it says, the Pixel 10, 10 Pro and the 10 Pro XL will not have regular SIM tray, instead support eSIM connectivity. It is intriguing to see Blass share the details, and also points out that the Pixel 10 Pro Fold model will have both the SIM options, which has seemingly been covered via multiple design leaks of the new Pixel foldable. He is quick to mention that Google might keep the eSIM models limited to the US market, which is what Apple has done with its iPhones in the last few years in its home market. We'll have to wait for over a week to see what Google actually plans to offer and if it does make two different versions of its Pixel 10 models this year. Google is likely to persist with regular Pixel 10 variants in markets like India, Europe and other countries for a few more years. Google Pixel 10 series is launching in New York on August 20 and we expect the company to bring four new Pixel models including the Pixel 10 Pro Fold version. Google could also have a few products lining up, and some more updates around the new Android 16 version and Gemini AI updates for these devices. view comments First Published: August 11, 2025, 08:14 IST Disclaimer: Comments reflect users' views, not News18's. Please keep discussions respectful and constructive. Abusive, defamatory, or illegal comments will be removed. News18 may disable any comment at its discretion. By posting, you agree to our Terms of Use and Privacy Policy.
Indian Express
an hour ago
- Indian Express
When UP legislators gathered for an AI workshop: Volley of queries and a mantra — ‘Google cycle hai, toh AI motorcycle'
A special workshop for UP legislators on Artificial Intelligence (AI) on Sunday led to a volley of queries, ranging from concerns to curiosity. 'How would AI help us to get votes?'; 'Would AI help us to know which road construction would fetch us more votes in our constituency?'; 'Would AI tell us which party is getting widespread support in the state?'; 'AI is like a thug. How can we trust the data provided by it?'; and 'How would you ensure that it does not defame us?' The 90-minute AI awareness workshop was organised by the Secretariat since plans are afoot to introduce AI tools to assist in the smooth functioning of the Assembly, right from attendance to recording statements and accessing archives. The Assembly session begins on Monday. Harshit Mishra, a Phd from IIT-Bombay and CEO of Clim Agro Analytics, who led the session along with other AI experts, explained the importance of AI and its uses and challenges to the legislators. But as the jam-packed session progressed, the queries posed by the legislators threw it off track. 'Vote maange kaise jayenge? AI se mil sakti hai? Jara technically samjha dijiye, kyunki humara labo labab NASA ka engineer to banna nahi hai. Humko vote hi lena hai (How will one seek votes through AI? Can we meet AI? Please explain the technology to us. Because we are not going to become NASA engineers. We just want to get votes,' an MLA said. Realising that the discussion was going off track, Speaker Satish Mahana stepped in along with Parliamentary Affairs Minister Suresh Khanna. To simplify AI, Khanna told legislators: 'Agar Google cycle hai, toh yeh (AI) humein motorcycle mein baitha dega (If Google is a bicycle, then this AI is like riding a motorcycle).' Explaining to the legislators that AI was merely a tool to assist them and that it would not tell their political future, Khanna said: 'AI gyan hai… jo bhi advancement chahta hai lena chahe to le (AI is knowledge…Those who want to advance can take it).' Like every other thing, Khanna told the legislators, AI too has its positives and negatives. Some of the legislators surprised experts when they shared that they use a tool such as ChatGPT for framing questions and making pointers for speeches, among other things. As some of the MLAs saw the AI-linked new initiative as a way to keep a tab on their attendance in the Assembly, Speaker Mahana assured, 'We have placed cameras in different parts of the Assembly. We can get to know which MLA has given a speech, on which topic, on which days in a year, or during their five-year term. It is not just information and attendance collection. As some of the MLAs said that AI would analyse their arrival and departure patterns … It would evolve with time to help them.'
