What Users Expect from Agentic AI

July 21, 2025 | By John Morkes, PhD

Some software companies are betting big on agentic AI (AAI) features and products without knowing users’ needs. I see quick guesses and assumptions without much data, driven by fear of missing out. 

A better approach is to figure out real user needs. This enables software companies to take advantage of AAI’s huge benefits while reducing the risk and cost of building the wrong thing.

Below are key expectations users have of AAI (based on my quantitative and qualitative research with users in the last three years).

Value 

Your AAI should enable or improve use cases of high value to specific users. Value is users’ top expectation. Do structured user research to define and prioritize the value of use cases you’re considering for AAI. This is consistent with a new report by Gartner (“Emerging Tech: Avoid Agentic AI Failure: Build Success Using Right Use Cases”). Gartner predicts that more than 40% of AAI projects will be canceled before 2028, due in part to unclear business value. 

Accuracy 

Imagine an AAI product for radiologists that misreads MRIs and misdiagnoses cancer patients. Users expect tasks to be completed accurately, with sound reasoning and no significant hallucinations. 

Visibility & Explainability

Because AAI operates autonomously with complex workflows, data and rules, users expect to see and understand what’s happening. “I want to know how it works, what it does, what it doesn’t do, what it just did, and why,” one enterprise user told me. This level of clarity lets users know what to do with your AAI and helps them spot and correct mistakes.

Trustworthiness 

Users need to trust AAI’s actions, decisions, and data, or they won’t want to use it. (Accuracy, Visibility, Explainability, and adherence to ethical principles help here.)

Control

Users need the ability to start, stop, adjust, and undo autonomous actions by AAI.

Usability

Users should be able to figure out how to use your AAI, and using it should be efficient.

Learning & Feedback Loop

AAI should learn and adjust continuously. Users expect to provide feedback when things go wrong, and AAI should integrate the feedback. The human element should be an important part of your evaluation toolkit.

Delight 

Users want to be satisfied and happy with their experience. If you do a lot of things right, then your users might be delighted.

Go to Blog Overview

How can we help?

Send us a message below.

We won’t sell or share your information. We don’t like spam, either.