Agentic AI Is Moving into Core Operations. Governance Will Decide Who Captures the Value.

On this page

Self-driving cars put autonomous decision-making on the boardroom agenda a decade ago. Agentic AI now brings that same capability into the enterprise, where the decisions concern credit, claims, supply chains, and customers. The technology has left the lab. The open question for leadership is how much authority to hand over, to which processes, and under what controls.

What Agentic AI Means for the Enterprise

Agentic AI systems combine large language models with planning, memory, and tool use. They pursue a defined goal, choose their next action, execute it across enterprise systems, and adjust when conditions change. Earlier chatbots answered questions. Agents complete work.

The shift shows up in the economics. BCG reports that agents already account for 17% of total AI value in 2025, and it expects that share to reach 29% by 2028 (BCG, September 2025). BCG also finds that the leaders, which it calls future-built firms, allocate about 15% of their AI budgets to agents. The gap between leaders and laggards keeps widening, and agents sit at the center of it.

For a concise primer on how autonomous systems differ from earlier automation, Tredence’s explainer on agentic AI covers the core concepts.

Where Enterprises Stand Today

Adoption rhetoric runs ahead of production reality. Four data points frame the picture.

  • Experimentation is widespread; scale is rare. In McKinsey’s 2025 State of AI survey, 62% of respondents say their organizations are at least experimenting with agents, and 23% are scaling agents somewhere in the enterprise. No more than 10% of respondents report scaling agents in any single business function (McKinsey, 2025).
  • Full deployment remains limited. Capgemini’s 2025 survey of 1,500 executives at organizations with over $1 billion in revenue found that 2% have fully scaled agentic AI deployment, and 14% have begun implementation (Capgemini, July 2025).
  • Trust is the binding constraint. In a Harvard Business Review Analytic Services survey of 603 business and technology leaders, only 6% fully trust agents to run core business processes autonomously. Only 12% say their risk and governance controls are fully in place (Fortune, December 2025).
  • Failure rates will rise before they fall. Gartner predicts that more than 40% of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls (The Star, June 2025).

The pattern is clear. Organizations that move agents into production gain ground, and those that stall in pilots absorb the cost without the return.

Three Decisions That Determine Outcomes

1. Set governance before you set scale

Governance now carries more weight than model selection. OpenAI’s December 2023 paper, Practices for Governing Agentic AI Systems, recommends three disciplines that leaders can apply now:

  • Evaluate suitability for each task. Test the agent’s accuracy and behavior across the full range of inputs and situations it will encounter, not only the scenarios in the demo.
  • Constrain the action space and define approval points. Decide which actions an agent may take alone, which require human sign-off, and what information the reviewer needs to make that call quickly.
  • Set default behaviors. Design the agent to choose the least disruptive response when it encounters uncertainty. A default that pauses and escalates costs far less than a default that acts on a bad inference.

Capgemini’s data shows why this matters. Trust in fully autonomous agents fell from 43% to 27% in a single year. Governance builds that trust back. Leaders who treat each agent as a governed identity, with defined permissions, audit trails, and an accountable owner, put themselves ahead of those who treat agents as software features. Forrester’s 2026 analysis reaches a similar conclusion: scaling fails on task complexity more than on the number of agents, and organizations need orchestration in place before they add more of them (Forrester, June 2026).

Action: Assign an executive owner for agent governance. Require an approval-point map and an audit trail before any agent touches a production system.

2. Match the strategy to the function

A large e-commerce company runs automation in customer service and in its supply chain. Agents can serve both functions, but the requirements differ.

Supply chain agents need to deliver efficiency. They move inventory, adjust routing, and reconcile orders, and the success measure is throughput and cost. Customer-facing agents carry a different burden. They must train on carefully governed data to avoid biased outcomes in credit, returns, and recommendations. They must also speak in the brand’s voice, resolve cases the way the company would want them resolved, and escalate to people when a situation calls for judgment.

Customer-facing deployments demand the most rigor, and they also deliver the most durable advantage. Loyalty follows consistent, fair experiences. Tredence’s analysis of the agentic organization describes how enterprises can redesign core workflows around agents, rather than bolting agents onto existing processes, so that each deployment serves a defined business objective.

Action: Rank your agent use cases by business objective and risk. Fund the customer-facing deployments with the strictest controls, and move the internal efficiency cases first to build operating experience.

3. Build the operating model and maintenance discipline

Integration with legacy systems, data infrastructure, and talent development remain the familiar challenges. Agentic AI intensifies all three. Agents consume far more compute, because they analyze live data streams, reconcile outputs from multiple models, and act without waiting for instruction. Teams need people who can build, supervise, and maintain agents, and those roles did not exist in most enterprises five years ago.

The hardest problem arrives after launch. Markets shift, regulations change, internal processes evolve, and customer preferences move. An agent trained on last year’s conditions drifts, and its autonomous decisions lose accuracy long before anyone notices. Monitoring and retraining need a named owner and a budget line.

MIT Technology Review Insights, in a report sponsored by Uniphore, argues that process redesign should come before model selection, and that most enterprises discover too late that having data differs from having AI-ready data (MIT Technology Review Insights, October 2026). Bain’s 2025 executive survey supports the operating-model case. Companies that use AI for agentic workflow automation are twice as likely as assistant-only users to say AI exceeded their goals (Bain & Company, November 2025).

Action: Form a cross-functional team that includes technology, risk, operations, and the business owner of each process. Give that team authority over the agent’s scope, monitoring thresholds, and retraining schedule.

The Economic Case Is Real, and Conditional

Capgemini projects that agentic AI could generate up to $450 billion in economic value by 2028 through revenue uplift and cost savings. That estimate assumes semi- to fully autonomous agents. It also shows a sharp divide: organizations with scaled implementations project about $382 million in average three-year value, while others project about $76 million (Capgemini, July 2025).

The value belongs to the organizations that execute well. It does not flow automatically to organizations that buy the technology.

What Leaders Should Do in the Next 12 Months

  • Name an accountable owner for every agent in production. Ownership includes permissions, performance, and retirement.
  • Map every agent’s approval points before launch. Document which decisions the agent makes alone and which require a person.
  • Redesign the workflow before you deploy the agent. Automating a broken process produces faster errors.
  • Budget for monitoring and retraining as a permanent cost. Treat agents as systems that require ongoing maintenance.
  • Start with internal efficiency cases, then move to customer-facing work. Build operating experience before you put agents in front of customers.

Sources

  1. BCG, “AI Leaders Outpace Laggards in Revenue Growth and Cost Savings,” September 30, 2025
  2. McKinsey & Company, “The State of AI in 2025: Agents, Innovation, and Transformation”
  3. Capgemini Research Institute, “Trust and human-AI collaboration set to define the next era of agentic AI,” July 2025
  4. Harvard Business Review Analytic Services survey on agentic AI trust, reported by Fortune, December 9, 2025
  5. Gartner prediction on agentic AI project cancellations, reported by The Star, June 25, 2025
  6. OpenAI, “Practices for Governing Agentic AI Systems,” December 2023
  7. Forrester, Brian Hopkins et al., “The State of Agentic AI in 2026,” June 2026
  8. MIT Technology Review Insights (sponsored by Uniphore), October 2, 2026
  9. Bain & Company, “Executive Survey: AI Moves from Pilots to Production,” November 24, 2025
  10. Tredence, “The Agentic Organization: How AI Agents Will Rewire Your Business,” January 2026
  11. Tredence, “Agentic AI Explained: Autonomy, Purpose and Real-World Impact”

Questions people ask

What is the primary recommendation of Agentic AI Is Moving into Core Operations. Governance Will Decide Who Captures the Value.?

Self-driving cars put autonomous decision-making on the boardroom agenda a decade ago. Agentic AI now brings that same capability into the enterprise, where the decisions

Who is this research written for?

Enterprise CIOs, CTOs, Chief Data & AI Officers, and technical procurement teams.

Related reading

All thought leadership
Top AI Firms

Independent research on the firms building enterprise AI

© 2026 Top AI Firms. Independent research. We take no payment for placement.About · Privacy policy · Cookies policy · Terms of use