AI agents are software that can carry out an assigned task over several steps, using tools and data. In 2026, two predictions for this field should be read together.

Rapid adoption: In its August 26, 2025 release, Gartner predicts that 40% of enterprise applications will be integrated with task-specific AI agents by the end of 2026, up from less than 5% in 2025 (Gartner, 26.08.2025).

High attrition: On June 25, 2025, Gartner also predicted that over 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls (Gartner, 25.06.2025).

Read together, they say: agent technology is spreading, but projects that do not establish business value and risk management from the start are hard to sustain.

What to watch when starting

• One measurable goal: Instead of “efficiency”, pick a defined goal such as measuring and shortening proposal preparation time.
• A small pilot: Try the agent first in a limited process with limited data.
• Human approval: Leave the final decision to a person in steps involving money, customer communication or personal data.
• Access limits and logging: Restrict which systems the agent can reach and log every action it takes.
• Cost tracking: Monitor model and infrastructure costs throughout the pilot and weigh them against business value.
• Personal data: If the agent processes personal data, plan your information and security obligations under applicable data protection law from the start.

At Keywork we recommend starting agent projects with discovery and a small pilot. Meeting requests: support@keywork.ai