Enterprise benchmarks
in AI and agentic 
orchestration 2026

How ownership, integration and governance are shaping the next phase of AI adoption

About this research

This report draws on a survey of 300 senior enterprise technology, risk and AI leaders across the United States, Canada and the United Kingdom. Respondents span more than a dozen industries, including software, retail, manufacturing and finance. Nearly half hold C-suite titles, and 69.7% represent organizations above $1 billion in annual revenue, with 30.4% above $5 billion. Every respondent has hands-on involvement in designing, implementing, deploying, governing or evaluating AI initiatives.

Top

Executive snapshot

AI has become central to enterprise digital transformation. Adoption is widespread, but many still lack the connective layer to run it safely at scale.

What’s working

What’s not working

Top

AI is everywhere, but maturity is not universal

Most organizations have moved past experimentation. Far fewer have reached full maturity.

Where enterprises sit on the AI maturity curve

Enterprise Benchmarks Graphic 2

Scaling means having multiple AI use cases fully deployed across teams; mature means AI is embedded enterprise-wide across core processes.

Between “scaling” and “mature” is where AI adoption gets harder. It’s also where pressures like integration, ownership and governance start to surface.

Top

Agentic AI is accelerating

Agentic AI is moving fast, although it’s earlier in its maturity curve than overall AI. Deployment grew at 89.4% of organizations in the past 12 months.

54% describe their agentic AI adoption as scaling or mature, well behind the 77.6% who say the same of their AI adoption overall.

report image 2 scaled

CDOs report the most mature 
agentic AI adoption of any role.

The balance between traditional automation and agentic AI is shifting

Today

Expected in 2 years

Almost entirely traditional automation

15.3%

13.7%

Almost entirely traditional automation

45%

25.7%

Almost entirely traditional automation

18.7%

29.7%

Almost entirely traditional automation

15.7%

19%

Almost entirely traditional automation

4.7%

11%

Almost entirely traditional automation

0.07%

0.03%

Today, 45% run mostly traditional automation with some agentic AI, and only 4.7% are almost entirely agentic. Within two years, enterprises expect that to shift substantially: for the share running mostly or entirely agentic AI to shift from 20.4% today to 30% in two years.

Agentic AI is moving fast enough that gaps in ownership, integration, governance and trust will not stay theoretical for long.

Top

Nobody fully owns agentic AI in the enterprise

No single function has taken full ownership of agentic AI:

donut 1

44% of organizations put IT Operations in charge of agentic AI initiatives

33.7% assign it to a dedicated AI or automation team

16% leave it with software engineering

4.7% allow individual business units to handle agentic AI

1.7% say no team owns agentic AI at all

When responsibility is split, integration and governance decisions inherit that same fragmentation.

The approach to building agentic AI is just as divided:

donut b

39.7% primarily buy pre-built agents

32.3% use a mix of pre-built and custom

25.7% build custom agents in house

2.3% don’t use agentic AI

Neither decision, who owns it or how it’s built, has converged on a default. Most organizations are running a mix of bought and custom agents, with no single function accountable for how they work together.

Top

Legacy and siloed systems are the 
load-bearing walls

While organizations know what they want to do with AI, they’re running into major barriers. With no single team accountable end to end, the standout technical challenge — connecting AI to the systems that already run the business — has gone unsolved.

Top barriers to scaling AI, ranked

Integrating AI with existing and legacy systems

22%

Governance, risk, security or compliance concerns

18%

Data quality, availability or accessibility issues

17%

Siloed systems or platforms that can’t easily communicate or share data

10.3%

Lack of talent, skills or expertise

8.7%

Unclear ROI or business case

8%

Lack of executive sponsorship, organizational support or budget

6.7%

Not every respondent ranked every barrier.

When asked why AI initiatives stall before reaching everyday use, three reasons stood above the rest:

Difficulty integrating 
with existing systems

Dollar

Data quality issues

Dollar 1

Governance concerns

Only 41.7%

report a strong ability to connect AI agents to legacy or core systems

Teal AI 8

55%

report the same for cloud and 
SaaS applications

Turqouise cloud 1 1
rw logo

Redwood customers were 1.2x as likely as those who do not use Redwood solutions to report a strong ability to connect to both legacy and cloud/SaaS applications.

Top

Proving ROI is still hard, even where adoption is strong

Even where AI is technically connected well enough to run, proving what it’s worth is a separate, unresolved problem.

Confidence in measuring the ROI of AI is split almost exactly down the middle:

48.3%

47%

4.6%

48.3% are very confident their organization is accurately measuring AI ROI

47% are only somewhat confident

The remaining 4.6% are not confident at all, or aren’t measuring ROI

The most common ways organizations measure ROI?

person

Productivity or efficiency gains

55.3%

dollar sign

Revenue growth

47.3%

gauge

Faster time-to-market

42%

There’s real doubt about whether AI spend is worth it.

43.7%

say AI spend is delivering strong value

21%

say it delivers value, but they’re spending more than necessary

14.7%

are increasingly questioning whether it’s justified

11.4%

have already begun scaling back investment

8.7%

say it’s too early to say

down arrow

11.3%

of organizations have already begun scaling back AI investment due to cost or ROI concerns, which is a clear sign that enterprise AI spending is facing real scrutiny, not just early enthusiasm.

As far as why some think they’re overspending, it’s not price or vendor choice. They point to AI usage scaling faster than cost controls, followed by limited cost visibility.

Top

Governance hasn’t kept pace with deployment

AI governance policies are common, but enforcement is where it gets tricky.

58.3% have a formal AI governance policy that is consistently enforced; 31.7% have one, but it’s inconsistently enforced.

77%

have experienced a negative consequence from an ungoverned AI agent

Unexpected or excessive cloud and computing costs are the most common consequence.

dollar sign 2

What ungoverned AI agents have cost organizations

Unexpected or excessive cloud/computing costs

34.7%

Operational disruption or system outage

31%

A compliance or regulatory issue

30.7%

Unauthorized access to sensitive data

28%

A reputational or customer-facing issue

28%

An incorrect or costly automated decision

24%

None of the above

18.7%

Not applicable

4.3%

Respondents could select multiple consequences.

CIOs report more types of ungoverned AI consequences than any other role, and VPs/Directors of IT most commonly point to unexpected or excessive cloud costs.



Just 18.7% of all respondents (not just heavy agentic AI users) report no consequences at all.

“Governance failures are an execution problem.”

– Charles Crouchman, Chief Product Officer, Redwood Software
Crouchman Charles Medium

By the seat: How this differs by role

45%

Error

of CDOs reported a reputational 
or customer-facing issue

41.9%

Error

of CIOs reported a reputational issue too, with compliance, cost overruns and unauthorized access clustered close behind

36.6%

Error

of CTOs reported an operational 
disruption or outage

37.6%

Error

of VP/Directors of IT reported 
unexpected or excessive cloud costs

Top

Trust breaks down where stakes are highest

Governance failures explain why trust isn’t uniform.

Trust shield

Combined trust (full trust + trust with human review) is 69.7% for financial transactions and 70.3% for production systems, compared to over 80% for internal workflows and IT infrastructure tasks.

Trust in AI agents to operate autonomously

Domain

Fully trust

Trust with human review

Limited trust

Do not trust

Financial transactions

Fully trust

22%

Trust with human review

47.7%

Limited trust

26.3%

Do not trust

3.3%

Production/data base systems

Fully trust

23%

Trust with human review

47.3%

Limited trust

24.3%

Do not trust

4.7%

Regulatory or compliance reporting

Fully trust

23.3%

Trust with human review

48.7%

Limited trust

24.3%

Do not trust

3%

Customer-facing communications

Fully trust

24.3%

Trust with human review

52.7%

Limited trust

20.3%

Do not trust

2.3%

IT infrastructure management

Fully trust

21%

Trust with human review

59.7%

Limited trust

16.7%

Do not trust

2%

Internal workflow or task orchestration

Fully trust

26.7%

Trust with human review

53.7%

Limited trust

18.7%

Do not trust

1%

Observability and stronger guardrails are nearly tied as the top two changes that would increase willingness to deploy autonomous AI agents on business-critical processes, cited by 49.3% and 48.3%, respectively.

Enterprise leaders also note observability and the ability to integrate with legacy systems as the two factors that would most increase their confidence to expand agentic AI use overall, cited by 62% and 58% respectively. Both point toward true orchestration as the way forward, before adding more AI.

By the seat: How this differs by role

Turqouise misc 17 1

73.2% of CTOs and 67.4% of CIOs cite observability as what would most increase their confidence, both well above the 62% overall figure

Turqouise architecture diagrams 2 1

68.2% of VP/Directors of IT and 60.5% of CIOs cite legacy-system integration, both above the 58% overall figure

Top

The next milestone: Orchestration

Self-reported AI readiness looks strong on its face, but the specific capabilities orchestration depends on are still missing at roughly half of organizations.

54.3%

say they’re fully prepared to orchestrate AI workloads across existing systems, data and processes

42.3%

have a centralized registry of deployed AI agents

54.7%

have a defined, funded plan already in motion

51.3%

have real-time monitoring of agent activity in place today

What would increase confidence to expand agentic AI use?

cover 2 scaled

Observability

cover 1 scaled

Legacy intrgration

Autonomy does not equal autopilot.

– Charles Crouchman, Chief Product Officer, Redwood Software
quote 2

Top

What sets the most prepared organizations apart

Most enterprises have the ambition, adoption and early results to show AI is working. What’s still missing is the layer that connects AI agents to the systems, guardrails and audit trails that make autonomy safe to expand.

Respondents using Redwood Software solutions to orchestrate mission-critical business and IT processes are

1.4x

as likely as those using other solutions to have a fully prepared, funded plan to connect AI to existing workloads already in motion.

icon

They didn’t wait for autonomy to outpace their infrastructure. Instead, they’re treating orchestration, integration, ownership and governance as prerequisites for scaling AI.

quote close

It’s no longer a question of whether AI can be used in a process. It’s whether a process enhanced by AI can run at scale with the level of security, observability 
and auditability that an enterprise needs.

– Charles Crouchman, Chief Product Officer, Redwood Software

Explore what agentic AI readiness looks like in practice.

All data represented in this report was gathered by Redwood Software and Leger Opinion, a third party whose panel was used in whole for data collection.

Top