AI agents — digital colleagues that can reason, act autonomously and learn from context — have captivated the business world over the past year.
The hype suggests a future where companies are solely driven by code. But as enterprises scramble to keep pace, a more pressing question emerges: are they building on solid ground, or on quicksand?
IBM’s latest research into UK and Irish consumer attitudes toward AI assistants suggests the appetite is there, with nearly three-quarters of respondents ready to embrace AI assistants to help make decisions.
“Readiness really varies by industry,” Sebastian Weir, executive partner for AI, analytics and automation at IBM Consulting UK, explains. “Financial services have a strong foundation in model risk management, so they’re better prepared to assess and control AI risk.”
“By contrast, consumer goods companies often aren’t ready for the level of disruption AI agents bring to the customer experience.”
Risk, trust and readiness
That disruption, Weir argues, goes beyond technology. “AI agents are completely reshaping the front door of business.”
“Some industries can be contemplative about how they implement change,” he says. “Others have to move because the change is happening to them.”
For financial institutions — already steeped in mature frameworks of governance and compliance — that shift is relatively smooth. Retail banks, for example, began experimenting with natural language interfaces more than a decade ago.
“Many started using [IBM Watson technology] as far back as 2013 or 2014,” Weir notes. “That experience gives them a head start.”
Weir contrasts this with sectors such as consumer retail or manufacturing, where AI remains largely confined to predictive analytics or marketing automation. “There’s a clear divide between organizations taking appropriate steps and those struggling to keep pace.”
Part of that struggle comes down to perception and privacy — issues that can derail adoption faster than a bad product launch.
He cites a recent example of a prompt injection embedded in a calendar invite, which allowed an AI assistant to control connected home devices. “There was no way to foresee that until it happened. That’s why continuous vulnerability assessment is critical. The technology moves too fast for one-off solutions.”
Still, enterprise controls have matured: “Using large language models securely in an enterprise context is well understood now, which wasn’t the case a few years back,” he says. And IBM’s methodology is secure by design: “We embed security from the start, anticipating risks as the threat landscape evolves.”
The four pillars of responsible AI
IBM’s approach rests on four familiar corporate virtues: transparency, trust, responsibility and ethics.
These four pillars are embedded across the company’s product development and consulting work, and they have tangible implications for return on investment.
“When we launched our large language model, Granite, some questioned why it was smaller than frontier models like GPT,” he says. “The answer is transparency. We know exactly what data goes into it, we can validate it, and we can indemnify it. That’s our strength. If you don’t know what’s in your model, you can’t quantify or control the risk.”
In other words, smaller and explainable can sometimes beat larger and opaque — particularly for regulated sectors.
“Enterprise AI is about augmenting human capability with transparency and control,” Weir says. “As long as that’s maintained, businesses can determine whether the associated risk is acceptable.”
From assistants to agents
AI assistants are evolving from reactive tools into proactive agents able to interpret context, make recommendations and take limited actions independently.
“Assistants will become pervasive,” Weir predicts. “We’ll see them used for everything from trip planning to research.”
But the bigger transformation will come from agents that operate continuously and autonomously. “Agents will take over repetitive tasks entirely. The biggest operational efficiencies will come from assistants and agents working together to augment what humans do.”
Augmentation rather than replacement is key. “A big misconception is that AI agents are one-to-one replacements for people,” Weir says. “In reality, one person might interact with thousands of agents performing specific, contained activities.”
Another misunderstanding is around data security. “People often blame AI for data breaches when the cause is human error,” Weir adds. “We hold AI to a higher standard than we hold ourselves.”
Oracle’s marketplace model
Oracle is betting enterprises can’t afford to wait. The company’s new AI marketplace, unveiled at Oracle’s AI World event in October 2025, puts prebuilt agents directly into customers’ hands — today, not someday.
“Organizations are under pressure to deploy AI to stay competitive, but many struggle to find the right solution,” says Guy Armstrong, senior vice president of applications for Oracle UK and Ireland.
Rather than asking companies to build from scratch, the marketplace connects customers with trusted partners offering prebuilt AI agents for specific business functions, all within Oracle’s enterprise-grade cloud.
“It’s designed to reduce friction, cost and complexity,” Armstrong explains. “Partners can contribute their innovations, and customers benefit from secure, accelerated AI adoption.”
Armstrong describes AI agents as a leap beyond traditional automation. “Automation executes predefined, rule-based tasks,” he says. “AI agents, by contrast, act as intelligent collaborators that understand business context, interpret data in real time and make proactive recommendations.”
The company has already embedded hundreds of such agents across its Fusion Applications suite.
“We see AI agents becoming an integral part of daily business operations,” Armstrong says. “They learn from enterprise data, understand context, and act autonomously. That empowers employees to focus on higher-value work.”
Where to start
Like IBM, Oracle advises clients to begin where the balance of risk and reward is clearest.
“AI agents can make a meaningful impact across all areas of the enterprise,” Armstrong notes, but he says Oracle sees the fastest ROI in critical business functions — finance, HR, supply chain and customer service.
He points to a customer experience agent embedded in Oracle Fusion Cloud CX, which analyzes sentiment in service requests to predict which cases are likely to escalate.
“It can pull up relevant information instantly, route tickets, manage follow-ups and resolve straightforward issues without human input,” Armstrong explains. “That frees customer service teams to focus on complex or high-impact interactions.”
In finance, agents can perform real-time variance analysis to improve forecast accuracy. In sales, they qualify leads and suggest next-best actions based on live data. In HR, they can map skills and create personalized development paths.
“This means more personalized and proactive engagement, faster resolutions and ultimately happier customers,” Armstrong says.
Building the foundations
Despite the hype and associated pressure to deploy quickly, Armstrong cautions that AI agents are only as effective as the environment around them.
“The best value comes when they’re targeted at high-impact challenges,” he says. “Some businesses overlook the basics — clean data, defined processes, employee readiness. Without those, even the smartest agents will struggle to deliver meaningful results.”
IBM’s Weir voices the same warning from another angle: “There’s no shortcut to maturity. AI literacy, security by design and clarity about what success looks like are fundamental. Once those are in place, the technology becomes an amplifier of human potential.”
Both leaders agree that responsible deployment and strategic clarity matter far more than scale for its own sake.
“We’re already seeing startups operating 90% through AI agents,” Weir notes. “Whether they become unicorns or remain niche is hard to predict, but the model offers real flexibility — especially for those building from scratch. Larger, established organizations face more challenges making that shift.”
A trusted colleague, not a replacement
Both IBM and Oracle envision a world where agents quietly underpin day-to-day operations — from forecasting cash flow to planning logistics or supporting employees’ learning paths.
Armstrong expects industry-specific agents to become “trusted advisors that understand context, predict outcomes and guide decisions.”
Weir agrees that agents and assistants will increasingly merge into an ambient layer of productivity. “The biggest efficiencies will come when assistants and agents work together to augment what humans do,” he says.
The opportunity is substantial, but so is the need for caution: “If you don’t know what’s in your model, you can’t control the risk,” Weir reiterates.
“As adoption grows, literacy becomes essential. Not everyone needs to be a data scientist, but everyone should understand the consequences of interacting with AI responsibly.”
