‘DIY’ agent platforms are big tech’s latest gambit to drive AI adoption

In recent months there’s been a big push on agentic AI in the tech industry, in particular with regard to ‘do-it-yourself (DIY)’ platforms that give businesses the tools to design their own agents.

Just this week, Oracle launched its ‘AI Agent Studio’ that embeds agentic tooling within the Oracle Fusion Platform, allowing customers to both create and manage agents or teams of agents across enterprise platforms.

OpenAI dropped a new set of tools and APIs earlier in the month that serves a similar purpose, allowing users to build agent logic and orchestration processes for agents to complete tasks.

Other firms such as AWS, Salesforce, and Workday have released their own platforms, all with a focus on giving users the tooling to make agents rather than simply providing off-the-shelf agents for businesses to use.

The reason that the emphasis is being placed on user customization and direction speaks to the use cases of agentic AI, which according to Gartner analyst Pieter J. den Hamer are often less deterministic or pre-defined than with other technologies.

“You need to tailor them,” Hamer told ITPro. “You need to customize them to the business context at hand.”

That may be why AI vendors are ramping up offerings in this regard, Hamer noted, as they know that end-users will get the most use out of agentic tools if they have full control over how they work, and what they do. It’s all in their hands.

Agentic AI in the hands of the enterprise

With the onus on the end-user, this raises questions over how these platforms will compete when they all offer similar tooling and the end result is largely dependent on implementation.

From an enterprise perspective, this also presents a conundrum over which platform to choose.

Ultimately, businesses need to pay attention to the nuances of different agentic platforms and assess whether the controls they provide align with their needs in terms of management, security, and scale.

Ease of development will be a key differentiator, according to Hamer, and there will be differences in whether agents can be created via low-code or no-code processes, or whether more complex engineering is required.

Integration with other tools and workflows - or with other agents - will also be a deciding factor, he added, as well as the flexibility or openness of the framework that the agents are running on.

If the technology is being targeted at staff as a means of letting them create agents for their own personal tasks, for example, then ease of use and configuration capabilities also become more important.

“I think importantly it's also the financial costs that also need to be monitored,” Hamer added, referencing the potential for vendors to charge users on a pay-per-use basis.

“If you have 10 agents that may be okay, but if, you know, if you're successful you get hundreds or 1000s of agents, then this cost may quickly spiral out of control,” Hamer said.

The ideal agent platform should also be able to integrate with agents from different vendors or origins in Hamer’s opinion, as customers may eventually want to, for example, make their Workday agents collaborate with their Salesforce agents.

With this in mind, as the agentic AI space matures in years to come, its development may mirror that of the broader cloud computing space, which initially started as a rush to the public cloud, but more recently has evolved to a point where many enterprises adopt a multi-cloud approach.

The risk of agent sprawl

While agents bring with them great possibilities in terms of driving efficiency and productivity, there is potential for them to create problems, particularly when businesses get to the stage of mass deployment.

Businesses could find themselves facing ‘agent sprawl’, whereby different staff members and teams are using multiple agents that may be siloed or not configured to communicate correctly with one another.

There may be hundreds of different models operating on different devices, according to Forrester analyst Craig Le Clair, meaning the importance of proper governance becomes more vital.

“Keeping the number of platforms to a minimum, having them be approved by corporate IT - that's really a critical governance function that we find,” Le Clair told ITPro.

If proper housekeeping isn’t done on the agentic front through the removal of excess tooling, then businesses may find themselves facing sprawl issues, Hamer said.

“If you do nothing then your agent ecosystem will quickly become an agent jungle,” Hamer added.

The prospect of agent sprawl mirrors a trend already witnessed in the broader enterprise space in recent years, with firms across a range of industries reporting problems with tool or application sprawl.

Many had already developed oversized software ecosystems before AI came along, and research shows this can have a huge impact on both efficiency and security.

Red Canary’s 2024 Security Operations Trends report last year showed many businesses were contending with heightened costs as a result of tool sprawl, for example.

Discussing the looming prospect of agent sprawl with ITPro, Oracle’s VP of AI business value Neil Sholay made similar comments about the issue and the need for proper agentic management.

“My personal view is we are very quickly going to move to a world where you have agents coming from multiple vendors,” Sholay said.

“I think we need to very quickly start looking at things like portability, reuse, and some kind of standard framework for how you have agent-to-agent interoperability,” he added.

MORE FROM ITPRO

• OpenAI unveils its Operator agent to help users automate tasks
• AI agent announcements are a dime a dozen right now – here’s what Oracle thinks it’s doing differently
• On-device AI assistants are meant to be helpful – why do I find them so annoying?