Adobe's 2026 AI and Digital Trends Report looked at how organizations and consumers feel about AI agents doing real work. 78% of organizations expect agentic AI to be handling customer support within the next 18 months.
The open question is how customers feel when it does. Those numbers tell a different story.
The Comfort Gap Is a Design Brief
Consumers are noticeably more comfortable when an agent works alongside a person (45%) than when it decides alone (21%). Businesses are more permissive than their customers on every one of these measures: 55% versus 45% for agent-plus-human, and 35% versus 21% for autonomous final decisions. On large, complex purchases handled entirely by an agent, it's 29% of businesses versus 16% of consumers.
Read that as a spec for agents: let the agent do the work and let a human own the consequential decision.
Where an Agent Should Run Alone
- Low-stakes, reversible steps such as scheduling, rescheduling, status updates, and order lookups
- Record updates and data entry that can be audited later
- Drafting replies and summaries for a person to review
- Routing and triage to the right team
Where a Human Should Stay in the Loop
- Refunds, credits, and pricing exceptions above a threshold you set
- Complaints and emotionally charged conversations
- Anything legal, financial, or hard to reverse
- Any case where the agent's confidence is low or the customer asks for a person
Make the handoff a feature. "Let me bring in a teammate who has the full context" beats a dead end every time, as long as the agent actually passes the context along.
Design the Handoff Along With the Agent
The weak point in most customer-facing agents is the transition. A good handoff passes the full conversation, what the agent already tried, and what the customer needs, so the customer never repeats themselves.
Set explicit escalation triggers: low confidence, negative sentiment, repeated rephrasing, or a direct request for a person. And tell customers when they're talking to AI. Trust is easier to keep than to rebuild.
The Quiet Opportunity: Being Found by Other Agents
The same Adobe report found that only 13% of organizations have embedded agentic AI organization-wide for brand discovery, meaning content optimized so AI-powered search can interpret and surface it. It also found that roughly half of organizations say an agent's ability to interact with other agents is a crucial factor when choosing vendors.
As customers' own assistants start researching on their behalf, being clearly described, well structured, and easy to cite becomes part of being found. Clear service pages, structured data, an llms.txt file, and content that answers the exact questions customers ask are a practical place to start.
A Safe Rollout Pattern
Start with one customer-facing workflow. Let the agent prepare and draft, and have a human approve. Track resolution time and customer satisfaction for a month. Where the numbers hold, widen the agent's authority one step at a time.
Where to Start
Pick the customer interaction where your team loses the most time to repetition, and map which steps are reversible. Automate those first, put approvals on the rest, and let results decide how much autonomy the agent earns.
Want agents your customers actually trust?
We design customer-facing agents with the right handoffs, approval gates, and escalation rules built in from day one.
Book a Free Call