AI agents promise a productivity revolution - here’s what leaders must get right

In 1987, Nobel Prize-winning economist Robert Solow famously quipped:

 You can see the computer age everywhere but in the productivity statistics.

It took the better part of a decade — and the internet — before PCs and productivity software truly transformed the way we worked. Today, we’re at a similar inflection point. Generative AI has captured the imagination of the business world, but the productivity surge many expected hasn’t yet materialized. The promise is there, but the leap from potential to measurable impact is complex. One of the most promising accelerators? AI-powered agents.

From hype to hands-on help

Unlike traditional automation tools, AI agents are built on Large Language Models (LLMs) and embedded directly into business applications. They don’t just follow scripts, they reason, adapt, and act. With conversational inputs instead of code, agents can pull data from ERP, HR, supply chain, sales, and customer service systems, then execute tasks or answer questions in natural language.

AI agents could automate various business tasks ranging from simple assignments, like drafting job descriptions, to more complex processes, such as triaging service requests, processing returns, or guiding credit underwriting.

Early adopters of AI agents are seeing real, tangible benefits. For example, Ford is using AI agents to convert 2D design sketches into 3D models. Sonos has agents that recall past customer interactions to help troubleshoot devices, even ones they didn’t manufacture. And in healthcare, AtlantiCare uses agents to help doctors navigate electronic medical records more efficiently.

Build or buy?

For IT leaders, one of the first decisions is whether to build custom agents from scratch or customize prebuilt ones. In most cases, customizing vendor-provided templates offers faster time to value and reduces complexity. Low- and no-code tools, which are already familiar to many business users, are making agent customization much more accessible without deep AI engineering skills.

That said, building from scratch offers maximum flexibility for specialized processes, though it requires more technical expertise and robust data preparation. Many enterprise vendors are investing heavily in agent design studios. For example, Oracle AI Agent Studio integrates with Fusion Cloud Applications, offering preloaded libraries of ready-to-build templates and the ability to create entirely new agents.

Data - the make or break factor

An AI agent is only as good as the data it can access and understand. Clean, well-structured, and 'self-describing' data is essential for agents to perform accurately. Unfortunately, most enterprises still have data scattered across silos in inconsistent formats.

Retrieval-augmented generation (RAG) can help by feeding proprietary business data into LLMs at runtime, but if that data is messy, incomplete, or poorly labeled, the results will be sub-par. If it’s not obvious to a bystander what the data means, then agents won’t help you.

The takeaway? Before deploying AI agents, invest in data quality, metadata management, and process integration.

From investment to ROI

Generative AI is a general-purpose technology, like electricity or the microprocessor, that can touch every industry and function. But history tells us it takes time for such breakthroughs to pay off.

McKinsey research shows that companies focusing on just a few well-chosen AI pilot projects see twice the ROI of those spreading efforts across too many. Business leaders looking to find success with AI should start narrow, prove value, and then scale.

The standards and governance gap

For AI agents to scale, the industry needs interoperability standards. Anthropic’s Model Context Protocol (MCP), Oracle’s Agent Intermediate Representation, and Google’s Agent2Agent (A2A) aim to simplify how agents communicate and access data and are already being used to connect data with LLM services.

Equally important is governance. Regulators in the UK, Singapore, and Hong Kong have made it clear that boards are responsible for overseeing AI risks. In the US, corporate case law is evolving in ways that could hold directors accountable for AI oversight. Yet today, only 17% of companies report board-level AI governance. Ensuring your organization has standard protocols around AI governance and oversight from the get-go is critical to future-proofing your enterprise AI strategy.

The path forward

AI agents won’t flip a switch and deliver instant productivity gains. But with careful investment in data readiness, process optimization, and sound governance, they could unlock a new wave of operational efficiency.

At Oracle, we’re building AI agents into our Fusion Applications to tackle everything from finance reconciliations and supply chain inspections to customer upsell recommendations. With Oracle AI Agent Studio, business users — not just developers — can create, test, and deploy agents that work across departments and data sources.

If history is any guide, the productivity payoff from AI agents will come, but only for those who lay the groundwork today. The PC and internet revolutions taught us that technology alone isn’t enough. It’s how we apply it, integrate it, and scale it that creates lasting impact.

The businesses that get this right won’t just see productivity gains, they’ll redefine what’s possible.