Are AI agents skipping the trough? Early evidence from life sciences

A familiar expectation in technology goes like this: breakthrough innovation sparks euphoric predictions, investors pile in, reality disappoints, valuations crash and only then does the technology mature quietly into actual utility. This is the Gartner Hype Cycle, a framework created in 1995 that became an industry institution after one analyst used it to correctly predict the dot-com crash in 1999. That single prescient call gave the model enormous credibility. It suggested that technology adoption follows a knowable, repeating pattern.

Not every technology neatly fits that arc. Cloud computing and mobile scaled into foundational infrastructure without a trough-style collapse. As cloud growth began to lose some of its momentum, John Dinsdale, Chief Analyst, Synergy Research Group noted in 2019 that the growth clip was still “impressive.” He added: “The decline in growth rate should not be viewed as a weakening market but as an inevitable consequence of a market that has now reached massive scale.”

GenAI, and by extension AI agents, could potentially also dodge a trough phase. “I don’t think it’s going to follow the traditional hype curve because I don’t think we’re going to have this crash of exuberance,” said David Rosner, Deloitte Digital’s Life Sciences leader. “Everyone we see and interact with is very rational and practical about where it’s going. The technology just needs to evolve.” He added: “Everyone’s expectations are: we get it, it’s not there yet, it’s not going to be there tomorrow. It might be there a year from now, and every day it just gets incrementally better.”

AI and agents are going to impact everything, eventually

Rosner went further: “AI and agentic is like the internet. It’s here to stay. It’s going to truly impact everything. That’s not hyperbole.” Unlike past hype cycles, everyone has measured expectations and capital keeps flowing anyway. Yet demonstrating ROI remains challenging. Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027 due to escalating costs, unclear business value, or inadequate risk controls. In finance functions, only 7% of CFOs report high ROI from AI. And McKinsey’s 2025 State of AI finds most organizations have yet to see enterprise-wide, bottom-line impact from gen AI, with a complementary survey showing only 1% of executives call their rollouts “mature.”

Deloitte survey data supports this measured optimism. Seventy-five percent of leading healthcare organizations are either experimenting with or planning to scale GenAI, and 82% have implemented or plan to implement governance frameworks. Leaders cite concrete benefits in the survey: 92% see efficiency improvements and 65% point to better decision-making. Meanwhile, 67% of enterprises increased GenAI investment in 2024 even as they confronted data quality, scaling and risk management hurdles. Deloitte predicts that 25% of GenAI-using enterprises will deploy AI agents this year, climbing to 50% by 2027.

Still, Rosner highlighted the emerging nature of the technology: “there’s some tension between the ambition and the reality of the technology right now.” The seven-figure, eight-figure impact applications aren’t fully deployed yet, he said, “but I’d also say it’s probably just a matter of time.”

From theory to practice

Some near-term applications are already delivering measurable value. “Patient services is an example,” Rosner said. “That is an example where agentic AI can really take cost out and increase productivity today.”

Patient services represents just one category of near-term productivity gains. “When you start getting into clinical trials and the whole development process, there are tons of productivity opportunities, and some of those are very real and very near-term,” Rosner said.

On October 6, Deloitte Digital announced its Agentic Patient Services suite built on Salesforce’s Agentforce platform: a multi-agent ecosystem designed to automate patient onboarding, adherence monitoring and care management. The agents aim to reduce manual touchpoints by 50-70% by autonomously handling tasks like enrolling patients, resolving duplicate records, and monitoring medication adherence.

According to Deloitte Center for Health Solutions research, 66% of commercial leaders in life sciences say significant change is needed in their patient services function, with technology and data capabilities presenting the greatest improvement opportunity. Pete Lyons, Deloitte’s Life Sciences Consulting leader, noted that “working with AI-powered agents frees up resources that clients can use to not only create a broader range of offerings and services for patients but can lead to improved conversion rates and better patient outcomes.”

“Offloading the monotonous administrative work from the patients, caregivers, providers and payers, and handing it off to AI allows people to provide true, empathetic engagement, and put the human back in health care,” said Frank Defesche, GM and SVP of Life Sciences at Salesforce.

The compute arms race

The patient services field represents just one category of near‑term productivity gains. “When you start getting into clinical trials and the whole development process, there are tons of productivity opportunities, and some of those are very real and very near‑term,” Rosner said. Agent‑friendly clinical trial workflows include tasks such as auto‑drafting protocols and study‑start‑up documents, personalized retention outreach and automated assembly/tagging for regulatory submissions.

Agent-friendly clinical trial workflows include tasks such as auto-drafting protocols and study-start-up documents, personalized retention outreach, and automated assembly and tagging for regulatory submissions. While specific productivity gains vary by workflow, the pattern mirrors patient services: manual, repeatable processes being handled autonomously while human experts focus on complex decisions and patient interactions.

How it’s getting implemented

Despite the momentum, Rosner stresses that “there’s some tension between the ambition and the reality of the technology right now.” The seven- or eight-figure impact applications aren’t fully deployed yet, but he adds that “it’s probably just a matter of time.” Practically, enterprises are building heterogeneous systems. “Not one platform is going to solve everything,” Rosner said. They’re pairing multi-agent orchestration with validator agents and governance so outputs can be trusted in high-stakes contexts. Deloitte’s guidance mirrors that architecture: mix models and tools, instrument quality checks, and scale only after governance, testing and data readiness are in place.

The same pragmatism explains why big pharma is investing in infrastructure that will compound those gains over time. Eli Lilly’s plan with Nvidia to build what they describe as the most powerful pharma-owned AI supercomputer signals a shift in competitive dynamics. Rosner sees it as a potential inflection point: “Is this now an example where yes, it’s still important, but the competitive advantage is: do you have enough compute, and access to compute, and ability to take advantage of the compute that will allow you to leapfrog and do things that others can’t? That could be a shift from the best human scientists in the world to just good enough science.”

In other words, life sciences may be entering an era where access to compute infrastructure becomes as critical as scientific talent—a fundamental change in how competitive advantage is built and maintained.