CIOs Plan to Rebuild Their Tech Stack for AI. Only 27% Have Planned for Agents.

Most large companies are about to rebuild a big part of their technology stack. Very few of them are planning that rebuild around the thing that is about to use it most: AI agents.

That is the headline finding buried in Accenture's first CIO Pulse survey, published on November 3, 2026. Accenture asked 300 CIOs and senior technology leaders at organizations with more than $1 billion in revenue, across 19 industries, about where their AI programs actually stand. The survey ran in August and September 2026.

The answers describe companies that have put AI on almost every desk, and have not yet built the plumbing that lets it do real work across the business.

AI Is on Every Desk. It Isn't Running the Workflows Yet

Accenture's first finding is a split between breadth and depth. 64% of CIOs say they are scaling enterprise AI deployment, but only 7% say they are optimizing at scale.

The difference shows up in how AI is used. 70% say AI is mainly used for individual productivity: copilots that help one person draft, summarize or track something. Only 37% are using agents for end-to-end orchestration, where AI coordinates connected tasks across teams by pulling data, routing decisions and updating systems. Even in IT operations, 49% still run agents with manual approvals at each step.

Accenture puts it plainly: the improvements "largely sit on single desks." A thousand people each saving twenty minutes a day is real, but it is not the same as a quote-to-cash process or a ticket queue that runs itself with people stepping in only where judgment is needed.

The Redesign Is Coming. Agents Aren't in the Plan

Here is the gap the title refers to. 61% of CIOs expect a significant transformation or fundamental redesign of their enterprise architecture within three years. But when Accenture asked which parts of the architecture were in scope, the agentic layer came last.

• Data architecture: 68% have it in scope
• Security architecture: 67%
• Cloud and infrastructure architecture: 66%
• Agentic architecture: just 27%, the lowest of the eight domains surveyed

Accenture's warning is direct: organizations that redesign data and cloud without the agentic layer "are building the infrastructure for the AI that already exists, not the AI that is coming." Its recommendation is to get orchestration, access control and memory into the design now, while the broader rebuild is still moving, because that "is significantly less expensive than retrofitting."

In practice, the agentic layer is the set of decisions most stacks have never had to make before:

• Which systems can an agent reach, and with whose permissions?
• Where do the credentials an agent uses live, and who approves their use?
• How does an agent hand work to a person, and how does a person stop it?
• What does an agent remember between tasks, and who can see it?

None of those are answered by a faster data warehouse or a new cloud region.

Governance and People Are the Real Bottlenecks

The survey also shows why so many CIOs are moving carefully. 54% name governance, risk and responsible-AI concerns as their top barrier to scaling AI, yet only 32% have reached Advanced or Leading AI-governance maturity. And 53% say user adoption and workforce training is their single biggest execution barrier, the highest of the seven barriers Accenture tested.

Data is the other drag. 53% cite poor data readiness and quality alongside sovereignty and regulatory constraints, and 51% cite the lack of a trusted, governed data foundation. Accenture's advice here is pragmatic: don't wait for a perfect data estate. Build a knowledge layer that tells agents what data exists and what it means, and let guardrails improve as the data does.

The encouraging part is that the controls are starting to be built. 70% of CIOs are building agent performance monitoring, 66% human escalation mechanisms and 61% agent governance and access controls.

Sign up for our newsletter

Get the latest updates, product launches, and helpful tips delivered straight to your inbox.

Email addressSubscribe

Costs Now Arrive by the Prompt

One more number explains why CIOs are being asked to prove value fast. 97% of organizations saw token consumption rise in the last 12 months, by a mean of 47%. Unlike a software license, AI cost accrues with every prompt and every agent action. At the same time, CIOs expect core operations to fall from 48% to 41% of IT budgets over two years, while spending on innovation and AI rises from 28% to 39%.

That combination, rising usage costs and a budget shift that has to pay for itself, is why Accenture keeps returning to one message: start with the workflows that justify the investment, and measure workflows transformed rather than licenses handed out.

What This Means for Teams Deploying AI Agents

For companies that want agents doing real work, not just drafting, the Accenture data points to a short list of things to settle early:

1. Pick the workflow before the platform. Name the process, the systems it touches and the outcome you'll measure.
2. Design permissions and credentials as part of the build. An agent that can reach five systems needs clear limits on each one, and a way for a human to approve sensitive actions.
3. Build the hand-off to people in from day one. Escalation and the ability to stop an agent are part of the design, not a later patch.
4. Keep model choice flexible. With token bills rising, the ability to use a cheaper model for routine steps and a stronger one where it matters is a cost lever.

This is the gap Workmate was built for. Workmate designs and deploys custom AI agents around a company's own workflows, connecting them to the apps and data a team already uses and setting their access and permissions as part of the build. Admins control which apps and skills each agent can use, credentials sit in a vault that can require human approval, and each agent runs on its own separate computer with private storage. The platform is model-agnostic, so each task can use the model that balances capability and cost, and agents work where people already are: Slack, Microsoft Teams and email. In other words, the agentic layer that only 27% of CIOs have planned for comes built in.

Frequently Asked Questions

What is Accenture's CIO Pulse survey?

It is a new recurring Accenture survey of senior technology leaders. The first edition, published November 3, 2026, covers 300 CIOs and IT leaders at organizations with more than $1 billion in global revenue across 19 industries and three regions, surveyed in August and September 2026.

What is "agentic architecture"?

It is the layer of a company's technology stack that lets AI agents work across systems safely: orchestration, integration, access control, memory, evaluation and guardrails. Accenture calls it "the connective architecture that makes scaled AI possible."

Why does it matter that only 27% have it in scope?

Because 61% of CIOs expect a major architecture redesign within three years. If agents are left out of that redesign, companies will likely have to retrofit the controls later, which Accenture says costs significantly more.