A survey of 625 IT professionals with cloud computing expertise, published today, finds most are opting to buy artificial intelligence (AI) capabilities rather than build them on their own.
Conducted by theCUBE Research on behalf of SOUTHWORKS, a provider of software engineering services, the survey finds 71% of respondents are relying on some type of agentic AI capability provided by a platform vendor, with 59% working with a software-as-a-service (SaaS) application provider. More than half (51%) still rely primarily on public AI tools for AI implementation.
By comparison, less than half (47%) are working with open-source frameworks and libraries and only 32% report plans to primarily build agentic AI capabilities in-house.
Regardless of approach, a total of 69% are also working with an IT service provider or consultants to acquire, build and, hopefully, safely deploy AI agents that are being used to automate repetitive tasks (73%), assist workers in decision-making (68%) and help diagnose and solve business problems (66%), the survey finds.
However, long-term survey respondents are specifically looking to invest in AI reasoning tools and platforms that can plan, optimize and justify outcomes (71%), AI assistants that help execute tasks (71%), autonomous, goal-driven AI agents that take actions (58%), agentic workflows led and orchestrated by AI (52%) and multi-agent collaborative systems (48%).
Less than a third (30%) claim to have plans to deploy AI agents that are based on a common framework, compared to 22% that have one in place. In comparison, 29% said agentic AI is limited to isolated departmental use cases or in siloed deployments across multiple business units without standardization (24%).
In general, most organizations today are still working with the first generation of AI copilot tools, says SOUTHWORKS CTO Johnny Halife. In fact, there is still a lot of confusion over what exactly constitutes an AI agent versus what is really just an implementation of, for example, Microsoft Copilot, he adds.
Additionally, many of the initial use cases involve automating routine tasks that might be better handled by existing automation platforms, says Halife. The main difference now is the conversational interface provided by AI tools is much more accessible so there is now a lot more experimentation with AI, he notes. How much of those efforts will later be shifted to a more robust automation platform remains to be seen, especially once the total cost of relying on large language models (LLMs) is calculated, adds Halife.
The truth is much of the cost of using LLMs to automate tasks is being subsidized by investors in the providers of these platforms that are today often operating at a loss. “Some of the use cases people are creating today use a lot of tokens,” says Halife.
In the meantime, there is little doubt that more tasks than ever will soon be automated. Less clear is to what degree all those instances of automation are going to provide a return on investment that provides a truly sustainable competitive advantage for the business. In most cases, it may turn out that AI is simply creating a new set of table stakes for remaining competitive.
