AI Agents: From Legacy Boosters to Future Business Leaders – But Not Yet
Nearly half of all companies are already experimenting with AI agents, but the promise of truly autonomous AI running the show remains years away. That’s the consensus from IT leaders at Microsoft Ignite, who are finding initial success automating existing workflows, not replacing them entirely.
The Current State: Agentic Assistance, Not Agentic Control
The current wave of AI agent adoption isn’t about radical reinvention; it’s about optimization. Companies like EY, Pfizer, and Lumen are leveraging these tools – often powered by large language models (LLMs) – to tackle tasks like knowledge management, content creation, and research. A recent McKinsey study confirms this trend, highlighting the widespread use of AI in these areas.
EY, for example, has already deployed 41,000 agents across its 30 million documented internal processes. “Moving those processes faster through agentic assistance like Copilot are kind of the low-hanging fruit,” explains John Whittaker, Director of AI Platform and Products at EY. The focus is on augmenting human capabilities, not eliminating them. EY’s “tax assistant” agent, trained on 21 million documents and continually updated with the 100 daily tax changes, exemplifies this approach – providing rapid access to critical information.
Phased Rollouts and the Confidence Gap
Pfizer is taking a measured, phased approach. Rather than immediately overhauling processes, they’re building confidence through iterative deployments. Starting with call center agents, Pfizer is expanding its use of AI as it demonstrates tangible efficiency gains. “Being able to start with a couple of them and make them more efficient, then gives us the opportunity to do it again and again,” says Tim Holt, VP of Colleague and Consumer Technology and Engineering at Pfizer. This cautious strategy reflects a broader industry sentiment: a need to validate AI’s performance before entrusting it with core business functions.
The key, according to Pfizer, is to first understand how AI works within existing frameworks before attempting to redesign those frameworks around AI. This is a critical distinction. The ultimate goal, Holt suggests, is to “blow it up and reimagine it,” but only after a solid foundation of understanding is established.
Beyond Human-to-Agent: The Path to Orchestration
Lumen is framing the evolution of AI agents in terms of gaming levels. Sean Alexander, SVP at Lumen, describes a progression: Level one is human-to-agent interaction, level two involves humans coordinating with multiple agents, and level three – the ultimate goal – is “full orchestration happening between the different agents.”
This vision of interconnected, autonomous agents is driving Lumen’s long-term strategy. They’re actively planning for the next 36 months, aligning AI agent capabilities with broader business objectives. Every new employee in Lumen’s connected ecosystem group receives a Copilot license, accelerating onboarding and knowledge transfer – reducing the time to full productivity from six months to just three.
The Importance of Fine-Tuning and Domain Expertise
Whittaker at EY emphasized the critical role of fine-tuning LLMs. “A regular large language model deployment… can be very good, but nowhere near the quality of what you get out of a fine-tuned model.” This highlights the need for specialized training data and domain expertise to unlock the full potential of AI agents. Generic AI tools are useful, but truly impactful applications require customization.
Looking Ahead: The Future of Agentic Systems
While fully autonomous AI-driven businesses are still on the horizon, the trajectory is clear. The initial focus on automating existing processes is paving the way for more ambitious applications. As companies gain confidence and develop the necessary infrastructure, we can expect to see AI agents taking on increasingly complex tasks, ultimately leading to a fundamental shift in how work is done.
The early innings of agentic technology are about building trust and demonstrating value. The next phase will be about reimagining processes, unlocking new levels of efficiency, and ultimately, allowing AI to take the reins – but only when we’re confident it can steer us in the right direction. What are your predictions for the evolution of AI agents in your industry? Share your thoughts in the comments below!
