By Holt Hackney
Organizations moving quickly to deploy agentic artificial intelligence should first determine whether proposed projects actually require autonomous AI agents, and whether their data and governance systems are prepared to support them, according to new research from Info-Tech Research Group.
The IT research and advisory firm has released a framework for evaluating agentic AI projects in data management, arguing that organizations risk wasting money when they fund projects before determining whether agentic technology is appropriate for the problem.
The research comes as businesses experiment with AI systems capable of performing multistep tasks, making decisions and taking actions with varying degrees of human involvement. The capabilities have generated interest in applying AI agents to data management, but Info-Tech argues that some projects described as “agentic” could be handled through conventional automation.
“The fastest way to waste an AI budget is to invest in agentic AI before confirming whether a use case actually needs an AI agent,” Jason Edwards, principal research director at Info-Tech, said in announcing the research. “Teams need to make sure the data and governance foundations are ready, understand how complex the use case will be to build, and confirm that an agent is the right solution before they move forward.”
Info-Tech identifies what it calls “agent-washing” as one obstacle to making those decisions. The term describes the practice of applying the agentic AI label to systems that may more accurately be characterized as traditional automation, robotic process automation or AI-assisted workflows.
That distinction can have architectural consequences. An autonomous system that makes decisions and takes actions may require different governance, security, oversight and escalation mechanisms than a deterministic workflow performing predefined tasks.
Info-Tech’s framework attempts to address the issue by requiring potential projects to pass three qualification gates before organizations rank them for investment: agentic fit, readiness and complexity.
The first asks whether the proposed use case actually requires an agent. Organizations are encouraged to define the trigger for the system, the actions an agent would perform and the intended outcome. Projects that can be addressed more predictably or economically through conventional automation can then be redirected before additional resources are committed.
The second gate examines organizational readiness. That includes whether the data, governance, controls and operating conditions necessary to support the proposed system are in place.
This step is particularly important for systems that are expected to act rather than merely provide information to a human decision-maker. Poor data quality, unclear ownership or inadequate governance can become more consequential when an AI system has authority to initiate actions.
The third gate assesses complexity, including the difficulty of building and operating the system and the confidence organizations should place in estimates of its potential value.
Only after projects have passed those evaluations does Info-Tech recommend comparing them according to expected benefits, implementation effort and risk.
The resulting projects can then be placed into a sequenced roadmap indicating which initiatives should proceed immediately, which should follow later and which require additional foundational work.
The approach differs from AI planning processes that begin by generating a large collection of possible use cases and ranking them primarily according to potential business value. Info-Tech’s methodology places qualification ahead of prioritization, effectively asking organizations to eliminate unsuitable or premature projects before deciding which remaining initiatives deserve funding.
The framework also reflects a broader enterprise architecture issue emerging with agentic AI: increased autonomy places greater demands on the systems surrounding the AI model.
Data quality, metadata, lineage, security, ownership, governance and escalation procedures become part of the architecture required to operate an agent reliably. An AI model capable of performing a task does not necessarily mean the enterprise is prepared to allow it to perform that task independently.
Info-Tech recommends that organizations document candidate projects consistently before evaluating them. Its methodology begins with identifying priority data-management areas and defining potential use cases. Candidates then move through the three qualification gates before qualified projects are scored and placed on a roadmap.
The firm’s accompanying tools include a candidate-definition workbook and a scoring and prioritization kit designed to standardize those evaluations.
For CIOs and enterprise architects, the larger implication is that selecting agentic AI projects may become as much a governance and architecture decision as a technology decision.
As organizations move from AI experimentation toward systems capable of taking actions inside enterprise environments, the question is increasingly not simply what AI agents can do, but where giving them that capability makes operational and economic sense.