Skip to main content

AI

July 8, 2026

 

AI agents are ready to act. Most systems aren’t ready to trust them

Here are the five organizational challenges stopping AI from moving beyond pilots into everyday use.

google logo
Kaushik Gopal profile photo

Kaushik Gopal

Executive Vice President, Business & Market Insights, Mastercard

Most organizations don’t have an AI innovation problem. They have a scaling problem. Across industries, companies have spent the past few years experimenting with AI pilots, yet many are still struggling to move from promising proofs of concept to real-world deployment.

The obstacle is rarely a lack of ambition — private global investment in AI grew more than thirteenfold between 2014 and 2024. More often, leaders haven’t built the conditions needed to trust AI, in particular agentic AI, at scale: culture, security, accountability, governance and data readiness.

Trust isn’t built upon a single control or policy. It depends on organizations creating confidence across multiple dimensions at the same time. Companies may have sophisticated AI models but weak governance, strong security controls but limited workforce readiness, or high-quality data but unclear accountability. Those making the most progress are not necessarily deploying more AI: They’re advancing across all these dimensions simultaneously.

Organizations can build the trust needed to move from insights to action only if they can answer all five of these questions. 

 

Do people know how to work with AI?

One of the most persistent signals we’re seeing throughout the industry is a disconnect between leadership enthusiasm for AI and employees’ understanding of how it should be used. That gap can make adoption difficult. Asking those same teams to embrace AI requires more than giving them access to new tools. They need to understand how AI should be used, where it can add value and how it will change the way their work gets done.

This begins with clear communication. While executive attitudes toward AI have changed quickly, that sentiment doesn’t always trickle down to their team members. According to Thomson Reuters Institute’s 2026 AI in Professional Services Report, nearly 40% of professionals say they’ve received conflicting directions about AI usage from leadership and clients. 

Leaders can close this gap by moving beyond broad AI mandates and giving employees practical guidance on when to use AI, where it should be applied in their workflows, and where human judgment remains essential. When employees understand how AI supports their work and what is expected of them, organizations can move from experimentation to adoption.

 

Should this agent be allowed to do this?

As AI agents move closer to decision-making and execution, the question shifts from whether they can generate insights to whether they can be trusted to act.

That’s why security, privacy and responsible use need to be built into the operating model from the start. For businesses, that entails establishing guardrails to protect sensitive information and create accountability before agents are trusted to access data or take on more complex tasks.

At its core, this is an identity and access challenge. Before an AI agent can retrieve information, initiate a payment or interact with a customer, organizations need confidence in what the agent is, what systems it can access and what permissions it has been granted. If your company doesn’t trust it, how can you expect your customers to? 

 

Who owns the outcome?

As AI agents take on more autonomous responsibilities, organizations face a different obstacle: determining who is accountable when an agent makes a decision, executes a task or produces an unexpected outcome.

While many companies are interested in deploying agents, only 32% say they’ve actually factored agents into workforce planning, according to a 2025 Cisco report on AI readiness. Without clear ownership structures, businesses risk creating uncertainty around decision-making, oversight and accountability.

Leaders need to establish clear ownership models for AI-enabled work, outlining which decisions can be delegated to agents, where human judgment remains essential and how issues will be escalated when systems behave unexpectedly. The organizations that scale agentic AI successfully will be those that take the extra step to redesign existing workflows so that humans and AI can operate together with clearer roles, responsibilities and expectations. 

 

How do we know it’s ready to act?

Once organizations have established accountability, a new question emerges: How do you know when an AI agent is ready for the real world?

Businesses need ways to test performance, evaluate outcomes and understand how agents behave before they are entrusted with customer interactions, operational processes or business-critical decisions. The focus shouldn’t be on creating an agent, but rather on taking the right steps to validate that the agent can perform reliably, consistently and safely across a wide range of scenarios.

Before agents are trusted to act and deployed at scale, organizations increasingly want environments where they can evaluate performance, identify failure points and refine controls.  That is one of the reasons Mastercard created Agent Suite, which helps organizations move from pilots to trusted deployment, and why we recently expanded it with Proto, a sandbox environment to explore, test and validate agentic AI use cases before introducing them into production workflows. Mastercard also introduced three new AI agents designed to support key enterprise functions: merchant, onboarding and dispute management. Together, these capabilities can help organizations identify where AI can drive business value while scaling within secure and governed frameworks.

 

Does it have the information it needs?

Even organizations that address adoption, security and governance often encounter another obstacle: AI agents can act only on the information they can understand — then the right data can be used safelyl, with clear permissions, when agents are acting in real time. 

But many organizations still operate with fragmented data spread across systems, inconsistent formats and disconnected workflows. While humans can often work around those gaps, AI agents struggle when information is incomplete, conflicting or difficult to access. This may mean reducing silos, cleaning up inconsistent data and fixing patchwork processes that may limit AI’s ability to work as intended.

 

Trusted AI action may become one of the most important competitive differentiators in business. Not because organizations can build AI, but because they can rely on it.

Kaushik Gopal
Kaushik Gopal

 

In an agentic environment, data quality directly shapes decision quality. If agents can’t reliably understand the information in front of them, they can’t reliably act on it either. To accelerate that transition, Mastercard sharpened its focus within Start Path, its award-winning startup engagement program, by engaging companies focused on business intelligence and data-driven capabilities.

Over the past decade, Start Path has welcomed over 500 companies in more than 60 countries, helping them scale by combining hands-on support with access to Mastercard’s global network and customers. This increased focus on startups that can help customers translate data into actionable intelligence can inform can inform better decision-making, drive performance and support real-time action — in turn building the confidence needed to move AI from isolation into the everyday.

In the coming years, trusted AI action may become one of the most important competitive differentiators in business. Not because organizations can build AI, but because they can rely on it.

Meet Start Path’s Business & Market Insight startups

   

Luca AI

Luca AI is an AI-powered digital team member that brings together sales, marketing, finance and operations in a single conversational analytics interface. Through this unified view, SMEs can access real-time insights across their business, helping them identify opportunities, improve efficiency and make day-to-day business decisions with greater confidence. Mastercard has collaborated with Luca AI to make these capabilities available to European business cardholders through the Mastercard Business Bonus and is exploring how its technology can further support agent-driven activities for SME issuers.

Luca AI logo

    

Yabx

Yabx uses AI-driven intelligence with low-code, no-code lending infrastructure that leverages alternate data to deliver scalable, risk-managed digital lending in emerging markets, in partnership with banks, fintechs, payment companies, merchant aggregators and other ecosystem players. Mastercard is working with issuers in the Middle East and Africa to incorporate alternative credit data such as mobile wallet and telecommunications information into credit risk assessments, enabling more inclusive card issuance and improved access to financial services.

yabx logo

   

BizMachine

BizMachine provides a B2B sales intelligence platform that helps companies identify and acquire new business customers. By combining BizMachine’s and Mastercard’s proprietary insights, the partnership enables European-based acquirers and other payment ecosystem players to identify high-potential merchants.

BizMachine logo

Guestologie

Guestologie gives restaurants the intelligence to deliver precision hospitality in real time, because lasting loyalty is created in the dining room, one personalized experience at a time. 

Guestologie logo