Is AI delivering on its promise of transformation?

The past decade, particularly the last year and a half, has seen a revolution in the artificial intelligence (AI) space with Generative AI (GenAI) unfolding new frontiers in the way organisations create, communicate, and innovate. Currently, agentic AI, though still evolving, is transforming the way humans and machines co-exist. It is one of the fastest-developing technologies in history, advancing rapidly and opening new avenues. While there has been mixed sentiments about intelligent agents, they have started to show promising results in business processes across industries such as pharmaceuticals & life sciences, auto, manufacturing, consumer packaged goods/fast-moving consumer goods/retail, BFSI, technology, media & telecommunications, as well as the government sectors.

Delivering on use cases

GenAI and agentic AI are helping achieve major efficiency gains in areas such as code and testing automation, code conversion across different programming languages, reverse engineering from code and the creation of design documents and synthetic test data at scale smoothly. Other use cases are context-based complex data and document search and summarisation, hyper automation of processes using digital agents, customer sentiment analysis and more.

Within a short span of time, digital agents are beginning to deliver highly effective use cases and execute complex business processes. While orchestrating tasks in collaboration with supervisor agents, they are already showcasing that humans can come into the loop only when needed. These agents are transforming front office (e.g. sales and marketing, customer services), middle office (e.g. finance, pricing, procurement) and operations (e.g. supply chain, IT, HR, manufacturing). In the pharmaceutical industry, for instance, agents are being used to automate the Food and Drug Administration (FDA) regulatory drug compliance report writing, earlier done manually by drug researchers. It is now being done by supervisor agents, who oversee and orchestrate other agents each to search the intranet, write text, draw diagrams, etc.

Within the FMCG sector, agents are now helping marketing teams convert plain English queries from customer databases into SQL queries. This reduces the dependency on IT teams to process customer insights.

In another instance, a GenAI-based code optimisation solution helped the offshore teams of a leading company with coding and design documentation, including language translation, thus ensuring offshore operations without support from onsite.

Agentic AI is now situated where GenAI was about two years ago. In these two years, we have witnessed GenAI projects go into production at an enterprise level. Given the speed at which agentic AI is being explored, it shows promise to scale effectively in the future. Having said that, it is highly critical to set a limit on how “human-like” we want the agents to become. The insertion of human in the loop needs to be appropriately decided and implemented to make the most out of agentic AI use cases.

Delivering return on investment

With the right mix of expert AI technologists and techno-functional folks who are skilled at translating between businesses and AI technologies, organisations can begin to identify and deliver the right business use cases with the expected impact.

The success of GenAI and agentic AI implementation hinges on how well it delivers against the organisation’s ultimate business outcomes and strategic goals. Enabling the right use cases at the right time with the right team helps organisations achieve ROI in terms of process improvements, cost savings, faster turnaround times, enhanced utilisation, increased sales and revenue, improved profitability and more.

Organisations must approach GenAI and agentic AI as long-term strategic initiatives to reap sustained success. This demands upfront investment in planning, cross-functional collaboration, change management, and, most importantly, in the right design and setting up the right infrastructure. Hardware and chip original equipment manufacturers (OEMs) are working to make the graphics processing units (GPUs) cheaper, while also working with data centre providers and cloud service providers to provide rent-a-GPU service. OEMs are also working to make LLMs deployable on-premises cheaply.

Moreover, effective design and software development techniques, such as efficient chunking and embedding, can make these AI solutions more cost-efficient, and efficient API calls can reduce cost. Effective platform Ops and LLMOps can be used to keep operational expenses on the lower side.

Addressing security and ethical concerns for long-term success

As GenAI and agentic AI systems transition from experimentation to large-scale deployment, ensuring safety and security becomes paramount. Best practices in model training, such as using high-quality, representative data and rigorous human review, are essential to mitigate bias and uphold ethical standards.

Techniques like retrieval-augmented generation (RAG) play a critical role in minimising hallucinations and enhancing response accuracy. This is especially vital for agentic AI systems, where a single misstep in a multi-step process can cascade into significant errors across the decision-making chain.

By proactively addressing the root causes of inaccuracies, implementing robust mitigation strategies, and aligning AI systems with societal and ethical norms, we can responsibly harness AI’s transformative potential in a secure and ethical fashion.

(The author is Partner, Deloitte India. Views are personal.)