Source: Unsplash
At SAP Sapphire 2026, SAP unveiled the “Autonomous Enterprise,” deploying over 50 domain-specific AI agents that execute business processes without human initiation. Gartner predicts 40% of enterprise applications will have AI agents by end of 2026, up from less than 5% last year. This article breaks down what that shift actually means for product designers and why the real bottleneck is not the AI itself. It is the interface layer that sits between the agent and the human.
Last week at SAP Sapphire, SAP did something that should have landed louder than it did. They did not just announce a feature update or a new AI add-on. They announced the Autonomous Enterprise. Fifty-plus domain-specific AI agents built into their core suite, executing finance, supply chain, HR, and procurement workflows without waiting for a human to click anything. The product is called SAP Joule, and it is not a chatbot. It is a system that does the work.
I have been building enterprise software for most of my career. Forty-two products shipped, Fortune 500 clients, Apple-grade design standards. And right now I am watching the ground shift in a way that reminds me of the early SaaS wave, except this one is moving three times faster and the implications for how we design products are almost completely underexplored.
“Enterprise applications will achieve 40% deployment of task-specific AI agents by 2026, up from less than 5% in 2025. AI agents will transition from assistants to autonomous executors across enterprise software.”
— Gartner Research, August 2025
Let me put some numbers around what is happening. The agentic AI market is sitting at $8.5 billion in 2026 and is projected to reach $47 billion by 2030. Multi-agent deployments across enterprises spiked 327% over a four-month window earlier this year, according to a Databricks survey. Deloitte is projecting that up to half of organizations will put more than 50% of their digital transformation budgets toward AI automation in 2026. These are not trend-watchers being optimistic. These are enterprise CFOs signing contracts.
The shift that SAP just made explicit is a shift from software that enables work to software that does work. For twenty years, enterprise SaaS has been about giving humans better interfaces to execute decisions. The CRM screen, the ERP dashboard, the procurement portal. All of it was designed around a human sitting in front of a screen, reading data, clicking buttons, initiating actions. That model is ending. Not slowly. Fast.
The Real Design Problem Nobody Is Solving
Here is what I keep running into in conversations with product teams: everyone is focused on making AI agents more capable, and almost nobody is thinking about the experience layer that connects those agents to real humans at work.
Think about it from a user’s perspective. You are a procurement manager at a mid-size company. You used to log into a portal, review purchase requests, approve them manually, track supplier status. Now an AI agent does that. Great. But here is the question: what is your job now? When do you step in? How do you understand what the agent decided and why? What does “reviewing” an autonomous process even look like as a designed experience?
Most teams are building the agent first and then bolting on a UI as an afterthought. I have seen this pattern in at least six enterprise products in the past year. The agent works. The interface makes no sense. Users end up either micromanaging the agent because they do not trust it, or ignoring it entirely because they cannot figure out what it is doing. The design problem is not making AI agents smarter. It is making them legible, interruptible, and trustworthy to the humans they work alongside.
Via GIPHY
What Agentic UX Actually Needs to Do
When I talk about designing for agentic workflows, I mean something specific. It is not just “make a chat interface.” It is rethinking five core UX responsibilities that used to belong to the human:
- Intent declaration: The user needs a clear, low-friction way to tell the agent what outcome they want, not which steps to follow. This is fundamentally different from a form or a menu. It requires natural language input combined with structured confirmation.
- Progress transparency: When an agent is executing a multi-step workflow, the user needs to see what is happening, not just a spinner. Think audit trail as UX, not just as compliance. Show the decisions made, not just the status.
- Interruption design: Every agentic workflow needs a well-designed moment where the human can say “stop” or “not like this.” Most enterprise AI interfaces today have no graceful interruption pattern. The agent either runs to completion or errors out.
- Error and exception handling: Agents will encounter ambiguous situations. The experience of handing something back to a human should not feel like a system failure. It should feel like a deliberate, designed handoff. This is probably the hardest UX problem in the space right now.
- Outcome review: After the agent acts, the user needs to understand what happened. Not a log dump. A designed summary that gives them confidence (or appropriate pause) about the result.
None of this is theoretical. I ran into every one of these gaps building an internal procurement tool for a Fortune 500 client six months ago. We had great AI. The UX was a mess. We spent more time redesigning the human-agent interaction model than we spent tuning the model itself.
The 11% Production Problem
Here is a stat that tells you everything about where enterprise AI actually is right now: 79% of enterprises say they have adopted AI agents. But only 11% are running them in production. That gap is not a technology gap. It is a trust gap. And trust is a design problem.
When users cannot understand what an agent is doing, they do not trust it. When they do not trust it, they do not let it run. When it does not run, the ROI disappears. Gartner has already flagged that more than 40% of agentic AI projects could be canceled by 2027 due to unclear value and weak governance. Unclear value and weak governance are symptoms of poor product design. They are not AI problems.
The teams that will win this wave are not the ones with the best models. They are the ones who figure out how to make autonomous systems feel approachable, controllable, and trustworthy to non-technical users. That is the product design challenge of the next three years.
SaaS Is Not Dying. It Is Being Restructured.
I want to push back on the “SaaS is dead” narrative that has been floating around since early 2026. SAP, ServiceNow, Salesforce, and Workday are not going anywhere. What is changing is the unit of value they sell. It is moving from “seats and access” to “outcomes and actions.”
Deloitte predicts that by 2027, the majority of new SaaS contracts will include usage-based or outcome-based components. By 2030, 35% of point-product SaaS tools will be replaced by AI agents or absorbed into larger platforms. The tools that will disappear first are the ones that only exist to bridge a human and a database. Scheduling tools, data entry workflows, report generation. All of that is agent territory now.
What survives? Products with deep data moats, strong network effects, and complex enough workflows that humans genuinely need to stay in the loop. The design challenge for those products is not “should we add AI?” It is “what is the right human-agent collaboration model for our specific workflow?” That is a product strategy question, and it is the right one to be asking in 2026.
What I Would Tell Every Product Team Right Now
If you are building enterprise software and you have not yet mapped your core workflows to ask which steps require human judgment and which do not, you are already behind. That mapping exercise is not just for AI roadmaps. It is the foundation of every design decision you will make in the next two years.
Start with your highest-volume, lowest-judgment workflows. These are the ones where an agent can run end-to-end with minimal oversight. Design the exception handling first, before you design the happy path. Think about what your users need to see to trust the system, not just what makes the demo look impressive.
And here is the thing I keep coming back to. The enterprise users who will interact with these agents are not early adopters. They are procurement managers, HR coordinators, finance analysts. People who have used the same software for ten years and are now being told that software is going to start doing their job. The design empathy required to get that transition right is significant. It is not about making things look futuristic. It is about making people feel in control of a system that is moving faster than they are used to.
I wrote more about designing for AI trust and human-agent collaboration on my reloadux blog if you want to go deeper on the design systems side of this. And if you are working through this in your own product, I am always up for a conversation.
What is the hardest human-agent UX challenge you are running into right now? Drop it in the comments. I read every one and I am genuinely curious what the patterns look like across industries.
Sources: SAP News Center, SAP Sapphire 2026 Keynote (news.sap.com); Gartner Research, “Gartner Predicts 40% of Enterprise Apps Will Feature Task-Specific AI Agents by 2026” (gartner.com); Deloitte, “SaaS Meets AI Agents: Transforming Budgets, Customer Experience, and Workforce Dynamics” (deloitte.com); Databricks Multi-Agent Deployment Survey, 2026; OneReach.ai, “Agentic AI Stats 2026: Adoption Rates, ROI and Market Trends” (onereach.ai); Gartner, “Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027” (gartner.com)