Agentic AI

More than 4 in 5 IT leaders believe the proliferation of AI agents will yield more complexity than value due to integration challenges and silos, according to the 2026 Salesforce Connectivity Benchmark Report. The company partnered with Vanson Bourne and Deloitte Digital, surveying 1,050 IT leaders for the report. 

Enterprises use an average of 12 agents, a number that’s expected to increase to 20 by 2027. Yet agentic AI’s long-term effectiveness hinges on data integration, according to 96% of IT leaders. The average organization manages 957 applications, but only 27% of those applications are connected due to data fragmentation, meaning agents lack access to the data they need, the report found. 

Nearly all enterprises experience data barriers for AI use cases, with 64% of IT leaders expressing concerns about their ability to meet AI implementation goals. 

As enterprises continue to adopt AI agents, IT leaders face the growing challenge of orchestrating and enabling a multiagent ecosystem. 

The Salesforce report found that agents working in silos can lead to disjointed workflows, redundant automations and a higher risk of shadow AI — the unauthorized use of AI tools. 

Integrating siloed apps and data remains a primary hurdle for 35% of survey respondents. IT leaders are assessing APIs as one way to connect AI agents, according to the report. One-third of IT teams are already using APIs to speed integration across systems. 

Agents need to be treated as part of a connected ecosystem that’s solving a customer or an internal-facing issue, said Kurt Anderson, managing director and API transformation leader at Deloitte Consulting LLP. 

“To be effective, you have to reimagine that integration strategy,” Anderson said during a press briefing. “It means building that API-driven architecture and using the API layer as that connective tissue for the enterprise. This foundation enables agents to have a secure path to data in the context it needs to deliver value.” 

Alcon, an ophthalmology company, is using MuleSoft Agent Fabric to govern its AI agents. MuleSoft is an integration platform Salesforce acquired in 2018. 

“Building agents across will not only improve our products but it allows us to move faster to the market,” Sreenivasa Patibandla, director of system integrations and APIs at Alcon, said during the press briefing. “We are building the platform to create the agents, to govern them properly.” 

More broadly, the industry is still searching for a common language to orchestrate agent interactions across vendors as adoption grows, Andrew Comstock, SVP and GM of MuleSoft at Salesforce, said during the press briefing. 

“Enterprises aren’t just sitting on one agent,” Comstock said. “This creates the multiagent enterprise, where agents from different vendors and ecosystems — Agentforce, Amazon Bedrock, Google’s Vertex AI and more — must coexist and collaborate.” 

AI vendors have worked to build open standards for agentic AI that will allow tools to communicate with each other across vendor platforms. 

In December 2025, the Agentic AI Foundation was created and co-founded by Anthropic, Block and OpenAI with support from Google, AWS, Microsoft, Cloudflare and Bloomberg, to offer a neutral foundation for AI agent standards development.

Half of software engineering teams say agentic AI is a top investment priority for them this year, and 84% say it will be a leading investment in the next three years, according to a SoftServe report. The report, conducted by MIT Technology Review, surveyed 300 CIOs, CTOS and other tech leaders in December 2025 and January 2026. 

Despite efforts to deploy the technology, agent integration with existing systems and the cost of computing resources represent barriers to AI adoption, the survey found. Just 12% of teams report widespread use of agentic AI in their organizations, the report found. 

AI experts said in the report that once fully integrated, agentic AI will become ubiquitous in areas like code generation, testing, refactoring and deployment of projects. In about three years, the report said, early adopter fields such as software expect agentic engineering to participate in overall lifecycle management.

Software development teams were early adopters of agentic AI, but their use has recently scaled up across multiple industries as tools have matured. 

Most organizations expect to pour more spending into workflows in the next several years, according to the SoftServe report.

But the cost of integrating these tools is a significant factor for tech leaders, according to the State of Analytics Engineering report released today by Dbt Labs, which surveyed 363 data practitioners and leaders in December 2025 and January 2026.

More than half of respondents said they’re increasing their warehouse and compute spend to keep up with agentic demand, compared with just 36% of leaders upping their team budgets. 

Although technical integrations have gotten easier for many teams, trust in quality of data remains a concern, Dbt’s report found. Clean ownership, validated outputs and documented data models are the foundation for data teams, said Dan Poppy, senior manager of content at Dbt Labs in an email. 

“Data leaders who frame governance as infrastructure are the ones who will be able to say yes to AI faster and more reliably than everyone else,” Poppy said. 

Tech leaders expect productivity improvements from agentic AI will take time, with most respondents in the MIT and SoftServe report saying they expect a slight or moderate improvement in the next two years. Tech leaders foresee AI agents speeding up their teams’ delivery of software projects, and they’re hoping AI agents can manage the product development and software development lifecycles over time. 

But AI failures will present different challenges than tech teams have faced before, Poppy said. 

“When a traditional system breaks, you know it,” Poppy said. “If an executive asks AI for net new ARR trailing the last 30 days and the AI system fails, it can produce a plausible, confident wrong answer that reaches a board deck or a customer before anyone catches it.”

Sponsored content

By Google Cloud

For over a decade, the enterprise technology narrative was defined by a single, seductive phrase: "There’s an app for that." It was a mantra that promised efficiency but ultimately delivered a fragmented reality. We traded manual processes for suites of applications, each requiring its own learning curve, and often their own logins and datasets.

But the era of the isolated application is ending. We are entering the age of the "Agentic Enterprise."

We believe the conversation has shifted from "there’s an app for that" to "there’s an agent for that." Yet, as we stand on the precipice of this shift, CIOs and IT leaders face significant challenges: the potential "agent-washing" of the market and the possibility of a flood of passive chatbots and rigid workflows rebranded as "agents."

If we are to realize the true value of AI, we must look beyond the hype. Real productivity isn't found in a smarter chatbot; it is found in a governed ecosystem of specialized, high-caliber agents that can execute complex, end-to-end tasks within the daily flow of work.

The Fallacy of the Bolt-On Agent
The greatest risk to the agentic transition is the "bolt-on" mentality. We need to avoid adding yet another disconnected platform. This creates friction, complicates security, and leaves data stranded in new silos.

To avoid this, the enterprise needs a unified environment where third-party agents and internal custom-built agents live side-by-side. The goal is a cohesive ecosystem that allows an organization to scale its AI capabilities infinitely without fracturing its underlying architecture. When agents from across the partner ecosystem—built for specific industries and functions—natively run on a unified platform like the Gemini Enterprise Agent Platform, the "silo problem" finally begins to dissolve.

Trust as the Primary Currency
For the CIO, the promise of autonomy is often overshadowed by the specter of risk. How do you grant an agent the power to act without sacrificing quality, control, and oversight?

The answer lies in moving away from "black box" AI toward a model of rigorous validation and cryptographic security. Every agent entering the enterprise environment must earn its keep. At Google Cloud, we believe this requires a strict evaluation of not just basic functionality, but output accuracy and autonomous execution standards.

Furthermore, we must solve the identity crisis of AI. In an agentic enterprise, every agent should possess a cryptographically secure identity. This creates a clear audit trail and ensures that while agents are autonomous, actions are not opaque. In addition, by utilizing tools like Agent Gateway and Model Armor, we can also help maintain that your enterprise data remains exactly that—your enterprise data—and is never used to train external models.

Governance: Keeping IT in the Driver's Seat
The most successful digital transformations are those that balance user autonomy with IT governance. We cannot expect employees to be productive if they have to jump through hoops to access the tools they need. Conversely, we cannot allow "shadow AI" to run rampant.

The solution is a "pull" rather than "push" model of deployment. By creating a centralized and vetted agent gallery, organizations can allow employees to browse and discover specialized tools—from functional agents to industry-specific agents and agents that span business processes—all from within their existing workflows. A “request agent” button can trigger a streamlined procurement flow for IT administrators and procurement teams. This keeps administrators in the driver's seat, allowing them to approve or deny deployments with granular control.

The Builder’s Advantage: Scaling at the Speed of AI
For technology leaders building these agents, the challenge is no longer just about engineering—it is also about distribution and getting these solutions directly into the daily flow of work for enterprise users and knowledge workers. We are facilitating a shift that turns everyday user discovery into a seamless procurement and deployment flow. By standardizing contracts and removing the traditional hurdles of enterprise sales, we are accelerating purchasing cycles by as much as 50%. This allows agent builders to focus on building rather than navigating red tape, creating a faster path to revenue and a quicker impact for the end customer.

A Shared Future
The move to an agentic enterprise is not a journey any company can take alone. No single provider can build every specialized tool required by every industry. The real value of this revolution will be realized through the power of a strong partner ecosystem.

We are building a future where agents don't just help us find answers, they redefine how every industry and line of business functions from the ground up. By focusing on high-quality standards, built-in safeguards, and unified orchestration, we are building the foundation for a more intelligent and secure agentic enterprise.

Businesses are launching into AI investments and it's having a profound effect on every aspect of the organization including how work gets done. The organizational shifts are wide-ranging and deep — affecting what kind of talent companies need and how job roles are defined.

Experts believe businesses will spend ample time and resources in 2026 revamping processes in response to AI adoption as they look to upskill workers and rethink operations. IT is not immune, but CIOs can play a larger role in guiding their companies and steering collaboration to usher change.

In some areas such as recruitment and customer service, the shift has been in play for a while. Software development, for example, saw a generative AI boom last year with the rollout of AI coding tools. 

“We have one big technology customer we work with that has shifted 25% of their developers to prompt engineers,” said Kye Mitchell, head of Experis U.S., referring to employees who specialize in crafting inputs for generative AI models. “We're definitely seeing the changes. I don't think folks are 100% sure what that’s going to look like, but it is evolving.” 

Alongside the addition of new job categories and changes in some job descriptions, companies are doubling down on existing roles that touch AI. Mitchell said the firm has tracked an uptick in demand for the roles supporting AI deployment, including data scientists, data architects and AI engineers.

AI adoption’s effect on the tech workforce is complex, but evolving, according to a CompTIA analysis published in November 2025. Almost two-thirds of businesses acknowledged AI as cover for unpopular staffing decisions, including hiring freezes or layoffs. But among companies that used AI to replace positions, 48% said they reallocated workers to other positions.

“AI’s relationship to human effort via automating and augmenting workflows will mean many job roles will change; new job roles will emerge, and some job roles will fade,” the IT trade group said in its report. 

For IT job categories, change in roles and responsibilities has been uneven thus far, according to Forrester VP and Principal Analyst J. P. Gownder.

“There's been a lot of automation and AI that’s been used in things like the helpdesk,” Gownder said. “Some years back, you would often have somebody who was running around doing password resets. That’s going to look different from an information security analyst who is going to have to use AI tools … but it is more of an augmentation tool.”

As companies adjust role expectations, CIOs can help steer organizations through change by focusing on the upskilling potential of in-house workers both within and outside of IT.

“We need to find ways to use AI in a way that will help us to be more effective in solving problems,” said Gownder.

Culture shifts and training

As AI changes the definition and scope of IT roles, CIOs can navigate the transition by putting the spotlight on underlying processes — training their employees along the way to help them adjust. 

“You're going to have to foster skills that operate quicker, that embrace innovation and that are really focused on taking that reimagination approach with this technology, so that you're fundamentally doing different things,” said Audi Rowe, EY Americas AI experience, strategy and transformation consulting leader.

Tech executives can start by standing up a cross-functional team of leaders dedicated to rethinking processes and training. That's the approach software company Informatica took to guide its own AI journey, CIO Graeme Thompson told CIO Dive.

The cross-functional team helped put together a training curriculum that featured external resources, including Coursera and LinkedIn Learning, to set up common ground understanding around AI definitions.

“The mistake a lot of us made was we thought that if we give Gen AI to employees, they’ll become more productive, and somehow that will create a business benefit,” Thompson said. “That’s not working. We have to redesign how the process gets done.”

Agentic AI is adding another wrinkle — introducing new roles and even new kinds of colleagues. The addition of “digital coworkers” means agent management will become a critical skill to hone across the enterprise, Dorit Zilbershot, group VP of AI innovation at ServiceNow, previously told CIO Dive.

“Workers at every level will be responsible for guiding, supervising and optimizing these digital coworkers,” Zilbershot said.

The influx of agents in the enterprise is already shifting employee expectations. A KPMG report found nearly two-thirds of organizations have altered their approach to entry-level hiring in response to agentic AI adoption. The shift underscored rising demand for competencies in data, automation and responsible AI, according to the firm.

Focusing on in-house skills can also help meet that demand, according to Rowe. “As you lean into an agentic way of working, there also has to be an upskilling of the workforce to understand how to shift into different roles, how to shift into roles that are higher value or more focused on innovation, but also roles that are focused on oversight, judgment calls, on making sure that you're developing and operating in a responsible way,” said Rowe.

Although the task of revamping technology roles is complex, CIOs should find ways to leverage the potential upside, according to Thompson. 

Informatica, for example, used an agentic AI tool internally to monitor cloud instances across several providers, optimizing usage and ultimately shaving $1 million in cloud spend last year, he said.

Tasking AI with repetitive tasks once relegated to junior staffers means IT can focus on big picture projects, according to Thompson. 

"Plus," Thompson said, "it gives them another $1 million to innovate."

Speed is a key factor in enterprise AI adoption efforts. Some businesses rushed into generative AI, only to find a plethora of implementation barriers. Others, like Mary Kay, paused to assess the situation. 

“When ChatGPT came out, that just changed the game overnight — it was the only thing that anyone wanted to talk about,” Mary Kay CIO James Whatley told CIO Dive. “We didn’t ignore it, but we didn’t jump all in.” 

The beauty company credits its slower start in generative AI with clearing the path for adopting AI agents down the road. 

“We knew once we had that foundation, that level of understanding, when the time came with the right tool and the right use case, we were going to be able to go faster,” Whatley said. “The worst thing you can do with AI is be naive about it.”

Mary Kay’s blended technology portfolio of generative and agentic AI is helping to build the digital toolset it wants to provide for customers, beauty consultants and staff, according to Whatley. 

AI agents are already live in one market with plans for more areas and use cases in the queue. The company, like others exploring agentic AI, is leaning on trusted partners to usher in the new capabilities. 

More than 2 in 5 companies were actively deploying AI agents in Q3 2025, a sizable jump from the mere 11% of enterprises doing so in Q1 of that year, according to a KPMG report. Business leaders have high hopes for how the technology can boost operations, but caution is critical. The majority of leaders are trying to hedge risks by accessing agentic AI only through proven providers, keeping humans in the loop and limiting access to sensitive data. 

“We have these partnerships with key vendors… and I’m leaning on them and taking advantage of their AI investment to benefit my customers,” Whatley said, pointing to Salesforce’s Agentforce offerings and Microsoft’s Copilot. 

While off-the-shelf solutions give Mary Kay adoption speed, the company is also beginning to develop customized agents for its IT service desk and other areas. 

“The agentic AI world is real and there’s true ROI, so it is going to be a part of our future,” Whatley said. “The use cases will only grow.”

Laying the foundation

Mary Kay’s early success with AI agents was only possible because it worked out the kinks before adopting generative AI. 

“We’ve been focusing on transformation,” Whatley said. “We’ve been able to close down five regional data centers over the last couple of years, moved and modernized our tech stack from custom solutions into a modern platform and deployed that around the world.”

The custom code that the company wanted to retain was also migrated to different cloud providers, helping to clean up years of technical debt during a transition to a cloud-first model. 

The projects changed workflows significantly and required a degree of upskilling among IT pros to handle the modernized tech stack. The rest of the business had to become familiar with the tools as well. 

“There were many layers of change that we had to go through, and we had to do this at every single market as we deployed around the world,” said Whatley, who transitioned to the executive team as CIO in January 2023 after initially joining Mary Kay in 1998. 

When the generative AI conversations started, Mary Kay quickly created and communicated an AI policy to communicate the dangers, especially as it relates to sensitive data. Shortly after, an AI committee was created to oversee implementation, ensure regulatory compliance and manage risks. 

“From three years ago to today, there’s a ton of tools out there,” Whatley said. “They’re not all great. We have to ensure that whatever we invest in, we get the ROI on.”

AI governance wasn’t an afterthought; it was the foundation of all the projects to come — and Mary Kay’s AI initiatives are better off for it, according to Whatley. 

“Committees and governance have a bad perception out there that it just slows things down or it's no gain, and that’s definitely not the case here,” Whatley said. “It has helped us consolidate the usage of approved tools as opposed to being scattered across groups and countries and that’s been very advantageous for us.”

The focus on governance has also helped project prioritization efforts. Before a proof-of-concept is even approved, the committee requires tools to prove their value in a sandbox environment. 

“A lot of times, it doesn’t pan out — what’s being sold is not what’s delivered,” Whatley said. “We’ve been able to be cost-conscious from that perspective and really been able to make progress without losing a lot of ground as well.”

AI use cases

Amid adoption efforts, leaders are looking closely at the intersection between AI and productivity.

Mary Kay has reviewed manual, repeatable tasks and assessed where AI could step in. Code generation, video translation and image creation are a few of the earliest use cases that have proven valuable. 

“Our goal is to get everyone understanding AI enough to help them in their specific role,” Whatley said. “In order to get there, you need more grassroots-type approaches to share the power and the risk of it.”

Mary Kay hosted a monthly series for teams to showcase potential use cases and foster discussion. One event in November of 2025 drew more than 150 people, Whatley said. 

The company is also exploring customer-facing use cases, such as its AI Foundation Finder. The tool, which launched in August of 2025, uses AI to identify 151 facial feature points, analyze the user’s skin tone and match it to the corresponding Mary Kay foundation. 

“There was a tremendous amount of testing,” Whatley said, pointing to both internal teams and third-party reviews. “The model was tweaked many times until we felt like it was right.”

Companies are often more cautious about AI use cases that are customer-facing. Accuracy needs to be higher, and the tolerance for hallucinations is lower. 

The tool has been used more than 185,000 times, according to the company. Teams are working on future iterations that can simplify the shopping process, allowing customers to add the foundation to their cart and purchase after being recommended.

“We’re continuing to tweak the model, and that’s the way this works,” Whatley said. “It’s only going to get better from this point forward.”

Marriott International is executing a multiyear digital and technology transformation plan that aims to retool the company, replace systems and develop a cloud-native innovation fabric to accelerate modernization. 

The company’s technology teams are simultaneously helping to scale high-priority generative AI use cases and exploring where AI agents could bring benefits down the road, according to Naveen Manga, global chief information officer at Marriott. 

“We’re in deep discovery with agents,” Manga told CIO Dive. “We have an agentic mesh capability that we’re building in our horizontal AI architecture, and we’re looking at use cases.”

The agentic layer will allow Marriott to create capabilities once and reuse them in multiple places. It’s core to the company’s AI playbook and part of its model-agnostic chassis, Manga said. 

Automating high-cost processes and improving experiences are two key priorities. The goal is to move associates away from manual, repetitive tasks to focus on better serving guests. 

“We’re looking at those capabilities to fix outdated processes, but we’re definitely not in any pilot with agentic,” Manga said. “We’re always experimenting with emerging technologies with both ambition and caution.”

Enterprises have high hopes of streamlining processes and improving workflows with AI agents, but there are plenty of unknowns accompanying the nascent technology. Fast movers are expected to face roadblocks, with Gartner predicting more than 40% of enterprise AI projects will be scrapped by the end of 2027. 

“We want this tech to be powered by trust,” Manga said. “We want to deploy technology that is responsible, ethical and lawful.”

Security focus

Most enterprises are exploring opportunities but not ready for widescale implementation as tech foundations get upgraded. 

Better cybersecurity and technology is a priority for Marriott, which owns 30-plus hotel and timeshare brands, including MGM Collection, Westin and The Ritz-Carlton. 

“Marriott’s cybersecurity strategy spans governance, risk management, operations and compliance, and using the right AI tools and technology are a key part of that strategy,” Manga said. “We continuously monitor the landscape as AI is a fast-evolving space. We remain committed to creating a culture of education and the responsible, ethical, and lawful use of AI.”

The sustained focus comes after the Federal Trade Commission settled a yearslong investigation into a series of Marriott breaches in October, requiring the Maryland-based company to make several data privacy improvements. 

“As part of the resolutions with the FTC and the State Attorneys General, Marriott will continue implementing enhancements to its data privacy and information security programs, many of which are already in place or in progress,” Marriott said in an October statement.

The company is beginning to see its efforts around replatforming and cloud come to fruition, following several years of higher-than-historical investment. 

“As of this week, we’ve begun the beta launches of our technology,” Manga said in late August of 2025, referring to the new central reservation system, property management system and loyalty system. About six hotels have beta access, and the phased launch will continue over the next 18 months. 

“From the onset, we’ve told our 800,000 associates that are going to use our technology that we’re not going to build it for you; we’re going to build it with you,” Manga said. “We created end-user councils, and we brought them along the journey.”

Manga has helmed the initiative, joining the multinational hospitality giant in 2021 as global chief technology officer before assuming his current role a few months ago. 

“In the new role of CIO, I’m also responsible for the technology department, the broader technology budget and our related aspirations for the company,” Manga said. “We have not backfilled the CTO role, nor do we have plans to. The head of technology is the CIO role.”

‘Ruthless prioritization’

In the era of AI everywhere, CIOs have had to polish off their project prioritization techniques. 

Marriott devised a four-part framework to guide its AI efforts, focusing on trust, accountability, prioritization and human-centered innovation.

“We set out by creating our AI principles and governance policies right out of the bat,” Manga said. “We use the governance that we’ve established in our company as the backbone to ensure that every AI use case is safe and reliable.”

The company has a few buckets of use cases: create, find, analyze and automate. Each unlocks different value for the company and are in service of guests, associates and owners or franchisees. 

“We pilot first, then we adapt, then we refine before we scale our solutions to the global portfolio,” Manga said. “It’s about ruthless prioritization.”

In early 2025, Marriott conducted an initial pilot of Microsoft’s 365 Copilot with about 100 associates across different disciplines, Manga said. Now, the company is rolling it out to thousands of its workers for meeting summaries and transcriptions, among other use cases. 

Marriott is also closely monitoring an AI-powered trip planning tool that a number of Bonvoy members have access to. The company also plans to expand its access to hundreds of associates. 

The hospitality company has boosted SEO traffic to its website by using generative AI to generate content as well. Other use cases include an AI coach for contact center agents. 

“In the past experiments, we were building these enterprise use cases and vertically integrating them into the technology stack. What we've learned is that for us to really get a faster speed to market — because we're looking at 10 use cases for 2025 — we don’t want to build each of these projects independently,” Manga said, referring to the in-progress, model-agnostic technology architecture. 

The governance framework guiding these efforts is about three years in the making — and Manga expects it to not stay the same forever. 

“It’s going to change and continuously iterate,” Manga said.

Agentic AI adoption is accelerating across industries, but governance frameworks are struggling to keep pace, according to Deloitte’s State of AI in the Enterprise report. The company surveyed more than 3,000 business and IT leaders across 24 countries.

Nearly 3 in 4 companies plan to deploy agentic AI within two years. Despite ambitious plans for deployment, only 1 in 5 report having a mature model for governing autonomous agents, a gap that can increase the risk of vulnerabilities, according to Deloitte.

To improve accountability as technology plans unfold, businesses should “anchor agentic AI programs in measurable impact from day one,” Jim Rowan, U.S. head of AI at Deloitte told CIO Dive in an email.

With AI adoption on the rise, agentic applications are emerging as a key growth catalyst. The tech is shifting AI from a decision-support tool to one capable of autonomous action.

While the capabilities race ahead, the necessary guardrails have been slow to follow, raising the alarm on accountability, risk management and regulatory compliance, according to Deloitte.

To scale AI safely and effectively, organizations must restructure from the bottom up and embed governance early in transformation efforts. 

“Speed and governance should not be viewed as opposing forces, as there can be alignment when you establish the right foundational capabilities upfront,” Rowan said. “We’ve seen clear patterns indicating that organizations without strong governance have difficulty scaling beyond the pilot stage. Focusing on this early allows for innovation and growth at pace with the technology itself, instead of facing stalled deployments and operational risks.”

CIOs facing pressure to deploy agentic AI quickly should prioritize governance elements such as a cross-functional team that brings IT, legal, compliance and business leaders together, according to Rowan.

Executives should also establish clear autonomy boundaries, create real-time monitoring for anomaly detection and add comprehensive audit trails to “capture the full chain of agent actions for compliance and improvement,” Rowan said. 

“The fastest-moving CIOs recognize that governance upfront eliminates the far costlier cycle of deployment, remediation and rollback,” Rowan added. “These elements of a governance strategy enable speed by building trust and reducing crisis management overhead.”

Demand for strong governance is only expected to continue rising as agentic, physical and sovereign AI see ramping deployment across industries. 

A previous report from IBM found governance gaps can exacerbate the financial and reputational repercussions of breaches, while a 2025 EY survey found more than 3 in 5 businesses experience AI risk-related losses of more than $1 million.

Bank executives see ample opportunity in agentic AI, but getting the most out of the technology means rethinking how it’s incorporated into business processes. 

AI agents can “truly transform operations,” said Ted Paris, SVP, head of analytics, intelligence and AI at TD Bank U.S. However, Paris said executives across the industry are paying attention to reports about AI project failure rates, which has pushed him to carefully consider how to incorporate AI so it becomes successfully implemented in the company's environment. 

“I reflect on that a lot,” Paris said. “It’s not an issue of getting the model wrong. It’s more about how well you’re integrating these things into your business processes so that you get adoption and usage.” 

AI represents a “massive opportunity” for TD Bank, Raymond Chun, group president and CEO of its parent company TD Bank Group, said in the company’s Q4 2025 earnings call. The bank implemented approximately 75 AI use cases generating $170 million in value in 2025, he said. It expects AI to drive $200 million in incremental value in 2026. 

“We are prioritizing our AI investments with use cases focused across categories such as customer acquisition, customer insights and risk management,” Chun said during the call.

As the bank scales AI, leaders have worked to involve the full ecosystem and understand how employees and customers will be interacting with the tools, Paris said. While the bank is implementing an AI strategy that includes traditional and generative AI, it’s also assessing, testing and building its agentic capabilities. 

TD Bank added a generative AI virtual assistant in its U.S. contact center. The Knowledge Management System handles questions from agents and provides consistent responses in order to reduce client hold time on calls and resolve issues faster while lowering the amount of times problems are escalated to supervisors, TD Bank told CIO Dive in an email. 

When bringing in AI agents, Paris said it’s critical to involve partners, including risk, legal, IT and change management, to help ensure “our colleagues are ready for this.” 

“You have to start from the beginning in thinking through: How are we going to adopt this? What’s it going to look like in its full production environment, where people are going to have to engage with it?” he said. 

Crafting an AI agent strategy

Financial services firms can employ different AI agent strategies, Paris said. One involves finding tasks that can be easily automated and inserting agents into those processes. Banks “can get a lot of benefit out of that,” Paris said. 

But another way forward is to rethink business processes to take an “agent-centric approach,” Paris said. 

“That means: Let me now pretend like I had these agents from the beginning as opposed to simply plugging them into a process that was designed for a different type of work environment,” Paris said. “Some of these work processes were created years and years ago.” 

Tools for enterprise AI agent design and deployment have matured to a point where financial services firms will take greater advantage of the technology this year, according to Accenture’s Top Banking Trends for 2026. 

Indeed, there are multiple types of AI agents banks can consider that serve different purposes, Paris said, such as administrative agents to help automate routine tasks, task-oriented agents to execute actions within workflows or orchestrator agents that help dictate what’s happening with other agents. 

“You need them all, with anything you’re doing,” Paris said. 

Morgan Stanley’s wealth management business, for example, is building super agents to serve as an orchestration layer, Jed Finn, head of wealth management at Morgan Stanley, said during the UBS Financial Services Conference in 2026. 

Meanwhile, Citi updated its AI platform Citi Stylus Workspaces last year to include agentic AI capabilities, enabling employees to condense multi-stage workflows into a single, automated process and gather insights from large datasets. 

Regardless of a bank’s approach to incorporating AI agents, vendors are zeroing in the technology’s potential in financial services. 

Software provider Creatio recently released six AI agents for banking that focus on specific revenue generation and operational efficiency tasks. Salesforce unveiled Agentforce for Financial Services in 2025, while AWS launched mortgage approval agents powered by Amazon Bedrock, a generative AI platform. 

Looking ahead, Paris said he wants to lean in on creating end-to-end projects. 

“This is not like the traditional world, in which you build a model and you can just deploy it,” Paris said. “There’s far more involved in the ecosystem and, to get the most out of it, you really want to make sure you’re completely rewiring it, not just automating individual bits and pockets.”

AI agents are proliferating as enterprises automate more of their operations. But ensuring agents can accomplish their tasks properly and securely remains a problem without an obvious solution.

Agents built to run in different environments typically don’t work well together. For instance, one organization might have a commercial off-the-shelf agent and another might have built a bespoke agent. They can’t communicate without some code to stitch them together.

“How do you get those things to work together?” asked Arnal Dayaratna, research VP of software development at IDC. “No one really knows, so part of this involves the invention of a shared data layer that allows agents to collaborate.”

For now, experts say, IT leaders must keep their AI infrastructure and architectures modular and flexible, in order to pivot when the market changes — and the market is always changing.

When communication breaks down

AI agents built today rely on special protocols, or communication methods, that imbue them with the structured vocabulary needed to break complex goals into manageable sub-tasks and hand them off. Naturally, a few protocols have emerged.

The most prominent is Model Context Protocol. Built by Anthropic, MCP determines how agents access resources, tools and information, such as APIs or databases. MCP is primarily optimized for agents within a shared trust boundary or a single application stack, such as Claude Desktop. 

Google’s Agent2Agent focuses on how agents communicate with each other, coordinate tasks and collaborate on workflows. A2A agents advertise themselves using a structured metadata format called agent cards, a profile of sorts that enables agents to discover each other.

On the fringes is Agent Network Protocol, an open-source offering that supports open web agent discovery.

The kicker: Each protocol was built on fundamentally different assumptions about where agents live and who they trust. And they weren’t designed to talk to each other natively, according to Microsoft research. This means that organizations must build APIs to bridge the protocols.

Building a bridge as the market evolves

The bridge-building is happening. However, it might not keep pace with new innovations and evolutions in agentic technology. Few agents today share an access layer that helps maintain the identities and consistency large enterprises require to honor governance.

Moreover, the agentic protocols of today won’t necessarily be viable tomorrow.

“All of these will get overturned and subsumed, and it’s going to happen super fast,” said Exabeam Chief AI Officer Steve Wilson. For example, Wilson said MCP is falling out of fashion due to its complex, server-based experience, while developers want a simple text file that enables them to build a skill.

Most IT leaders tend to err on the side of caution when it comes to high-stakes technical issues. Yet IT leaders must experiment with AI applications using disparate protocols, albeit leaving themselves the flexibility to course correct quickly.

“Given how fast things are evolving, if you’re not experimenting and pushing to see where things are at, you can get left behind quickly,” said Johnson & Johnson CIO Jim Swanson. 

To that end, the speed of the evolving AI market led J&J to apply modular architectures and reimagining business workflows that span multiple platforms and datasets, Swanson said. 

“There are significant opportunities across your business processes but you have to marry it with the quality of the data and maturity of the full technology stack inclusive of the AI component,” Swanson said. 

The safest course of action is for CIOs to insert orchestration, policy and transactional control between agents and systems, said Workato CIO Carter Busse. This will afford IT leaders the power to control “how AI acts across the business,” Busse said.

The narrative arc around agentic AI interoperability or incompatibility might seem familiar to longtime IT pros. One fitting analogy is the network protocol wars from 30 years ago, with IBM, Novell and others vying to standardize solutions before the industry settled on TCP/IP as the internet exploded.

Even so, consolidation among agentic standards will happen more quickly than the decade it took for TCP/IP to become dominant.

“All this [agentic AI] stuff is just in its infancy,” Dayaratna said.

Anthropic is pushing for open AI development with the donation of the Model Context Protocol, a universal standard that enables AI applications to connect to external systems, to the nonprofit Linux Foundation. The newly formed Agentic AI Foundation will govern the project, Anthropic said. 

Co-founded by Anthropic, Block and OpenAI – with support from Google, AWS, Microsoft, Cloudflare and Bloomberg – the AAIF aims to ensure transparent development of agentic AI, the release said. Anthropic’s MCP donation joins OpenAI’s Agents.md, an open format to guide AI agent development, and AI agent Goose by Block, as founding projects for the AAIF.

“The more often tech companies implement and improve upon this standard, the faster we work towards that plateau of productivity, and that’s what we need for increased enterprise adoption,” Gartner Analyst Jim Scheibmeir said in an email to CIO Dive.

Tech companies are capitalizing on the growing wave of enterprise AI agent adoption by establishing standards for the technology’s development that allow greater interoperability between different systems. By 2026, 40% of enterprise applications will feature task-specific AI agents, Gartner predicted.  

Retailers, financial services firms, beauty products, food and beverage makers and others are considering how to incorporate AI agents to improve customer service, support employees and improve business operations. Walmart, for example, deployed “super agents” to support customer shopping experiences, assist suppliers and centralize information. 

With enterprises finding and acting on use cases, vendors are taking steps to ensure the rapidly growing number of agents can talk to each other. 

Anthropic first released MCP in November 2024, and it has since grown in popularity with Microsoft Copilot, ChatGPT and Gemini adopting the standard. Salesforce announced in June 2025 that Agentforce 3 also supports MCP as interoperability plays a crucial role in the company’s overall AI strategy. 

The AAIF will provide a neutral foundation for a shared and open ecosystem of standards, innovation and tools for AI agent development, according to the Linux Foundation, which fosters and supports open source ecosystems. 

“Within just one year, MCP, AGENTS.md and goose have become essential tools for developers building this new class of agentic technologies,” Jim Zemlin, executive director of the Linux Foundation, said in a press release. “Bringing these projects together under the AAIF ensures they can grow with the transparency and stability that only open governance provides.” 

But MCP implementation faces some hurdles, including security concerns and frequent updates for early adopters, Scheibmeir said. 

Still, MCP adoption is anticipated to grow among enterprises in 2026. Its donation to the Linux Foundation will be a “net positive for the protocol’s future,” Scheibmeir added. 

“The more enterprises experience [MCP] through vendor-created solutions, the more enterprises will implement within their own software engineering efforts,” Scheibmeir said.