Gartner: Over Half of Enterprises Will Abandon AI Assistants for Agents by 2028

Research firm Gartner has issued a forecast that strikes at the very foundation of today's business AI tools market: by 2028, more than half of enterprises are expected to stop paying for AI assistants and copilots, tools like Microsoft Copilot and similar products that help people perform tasks and make decisions. In their place, agentic platforms that carry out tasks on their own are set to take over.

What's changing

At the heart of Gartner's forecast is a shift in the role software plays inside companies. Until now, AI tools such as copilots and internal assistants have served an advisory function: suggesting, summarizing, proposing the next step, but leaving the decision and the click to a human. The new generation of agentic platforms is set to reverse that relationship.

Software is shifting from being a tool people use to do work, toward a situation where people oversee intelligent systems that act on their behalf - Alastair Woolcock, VP Analyst, Gartner

In practice, this means employees stop being tool operators and become what Gartner calls an 'agent steward,' someone who supervises the output of autonomous systems. Instead of clicking and approving every step, the person defines the agent's goals and boundaries, then checks the results.

Where it hits first

Gartner points out that the first waves of change will affect processes built on clear rules and multiple approval stages, where response time matters. These include IT service management, legal and compliance support, procurement, CRM systems, revenue operations, and select functions within ERP systems. In these areas, an AI agent can independently carry a request from submission to resolution without waiting for further human sign-off.

Analysts issue a direct warning to vendors of customer service and service operations software: if they fail to build delegated task-execution mechanisms into their products within the next 18 months, they risk being pushed out of the market by competitors already offering full agent autonomy.

A new architecture of advantage

According to Woolcock, the ability to hand agents execution authority isn't just a product feature, it's an architectural decision, one that spans identity control, permissions, policy enforcement, access to a company's systems of record, and the ability to audit actions. The winning vendors, he says, will be those who build agent orchestration directly into companies' systems of record, offer policy-aware APIs, and enforce identity control and auditability at the platform level itself.

Gartner also stresses the importance of what it calls 'controlled openness,' meaning offering direct integrations and access via the Model Context Protocol (MCP) rather than closing off the platform to competitors. In this scenario, companies that restrict access to their data and systems stand to lose the most, as competitors' agentic platforms simply route around them using other vendors' open integrations.

What it means for Polish companies

For Polish businesses still learning to deploy traditional AI assistants and copilots, Gartner's forecast is a signal that investment in advisory tools may have a shorter lifespan than assumed when signing multi-year licenses. Companies planning AI software purchases should already be checking whether their vendor plans to move toward agentic task execution or intends to stick with a pure advisory model.

It's also a point in the broader discussion about the pace of AI adoption among Polish companies, where earlier studies found a significant share of businesses still lagging behind the EU average. If Gartner's forecast holds, companies only now buying simple copilots could face another wave of tool replacement with agentic systems within two years.

Gartner notes that this is about a market segment winding down, not the complete disappearance of AI assistants. Simple advisory tools are expected to remain in use where decisions require human judgment, regulatory oversight, or involve tasks too nonstandard to fully automate. The shift primarily concerns repetitive, rules-based processes.