Survey finds enterprises race to agentic AI, lag on scale

SOUTHWORKS has published new survey data suggesting many enterprises plan to source agentic AI through external platforms and service providers, while relatively few have deployed these systems across the whole organisation.

Conducted with theCUBE Research, the survey questioned 625 cloud architects, cloud decision-makers and cloud-native Azure professionals. It examines how organisations source and scale agentic AI, including more autonomous systems that can take actions and run workflows.

The findings suggest a gap between interest and operational readiness. Nearly all respondents said they see value in agentic AI capabilities such as AI reasoning, decision optimisation, AI assistants for task execution and fully autonomous agents. However, deployment and sourcing plans indicate many organisations lack the internal resources to standardise and scale these initiatives.

External sourcing

Asked how they plan to source agentic AI, 70.9% cited agentic AI platform vendors and 68.6% cited IT or consulting service providers. SaaS application integrations were also common (58.7%), followed by open-source frameworks and libraries (47.4%).

Only 31.5% said they primarily plan to build agentic AI capabilities in-house. The survey also found 50.7% of organisations still rely mainly on public AI tools for implementation.

SOUTHWORKS said the results suggest many technology leaders are prioritising speed of delivery and access to scarce skills over long-term internal development, but warned the approach could increase dependence on third-party tools and providers over time.

Johnny Halife, CTO at SOUTHWORKS, said organisations see the potential for agentic AI but struggle to deliver it with existing teams.

"This research confirms what we've been hearing from enterprise tech leaders as they work to operationalize agentic AI across complex cloud environments: they see enormous potential in agentic AI, but few can realize it with internal teams alone," said Johnny Halife, CTO, SOUTHWORKS.

Capability priorities

Respondents ranked AI reasoning that can plan, optimise and justify outcomes as the most important feature (71%). AI assistants that help execute tasks followed closely (70.7%).

Autonomous, goal-driven AI agents that take actions ranked next (57.6%). Agentic workflows led and orchestrated by AI came in at 51.7%, while multi-agent collaborative systems ranked at 48%.

Over the next 18 months, respondents expect continued investment across these areas. The reported order of expected investment was AI reasoning, autonomous AI agents that take action, GenAI assistants, agentic workflows and multi-agent collaborative systems.

Current usage still skews towards narrower applications. Respondents most often use AI agents to automate repetitive tasks (73.1%), assist workers in decision-making (67.5%) and diagnose and solve business problems (65.6%).

Deployment maturity

Despite high interest, deployments remain fragmented. Some organisations have implemented agentic AI at scale and with consistency, but many remain limited to departmental or siloed programmes.

About 29.8% said they have enterprise-wide agentic AI deployments built on a common framework. A similar share (29.1%) said deployments are limited to isolated departmental use cases. Another 23.8% reported siloed deployments across multiple business units without standardisation.

A further 17% said initiatives remain in pilot or experimental phases. Separately, 10.6% reported AI use across multiple business units without shared standards, while 20.2% said they have enterprise-wide AI deployments built on common frameworks.

Paul Nashawaty, Practise Lead and Principal Analyst at theCUBE Research, linked these patterns to trade-offs between speed and control.

"Our research shows agentic AI has crossed the curiosity threshold and entered an execution phase, but enterprise readiness is lagging ambition," said Nashawaty. "While nearly all respondents see value in agentic AI, only 31.5% plan to build these capabilities primarily in-house, and fewer than 30% have standardized enterprise deployments, forcing organizations to trade speed for long-term control as they rely on platforms and services to move forward."

Governance questions

The survey also highlights governance and security considerations as more organisations adopt agentic AI through external tools and partners. SOUTHWORKS said that as agentic systems become more autonomous and more embedded in production environments, gaps in standardisation and ownership can increase risk.

Halife said these systems can affect data and business processes across an organisation.

"Moving fast with AI only works if teams can keep control of what they build," said Halife. "Agentic systems touch data, workflows and decisions across the organization. Without clear ownership and governance, early speed can create friction as deployments scale."

SOUTHWORKS said it works with organisations through a development-on-demand model that embeds senior engineers within client teams. It described this as a way to increase delivery pace while keeping ownership of architecture, workflows and governance with the client.