Designing the first five minutes

For any team building an agent, factors like accuracy, latency, hallucination, and context handling are top of mind. And rightly so – they shape every interaction. But users don’t form first impressions based on output quality alone. Their experience begins from the moment they open the agent, and it’s important that we treat early interactions with just as much care as what comes after.

To understand what makes an effective first impression, our user experience research focused on those crucial opening minutes with users trying a sales agent for the first time.

The agent was built for sales managers and reps who wanted to scale personalized outreach. Before users could see any output, they had to complete onboarding: providing knowledge sources, configuring the agent, running simulations… all through a chat-based interface.

When we observed users trying the agent for the first time, we found that their impression was significantly shaped by how the onboarding experience framed value, capability, and role. Based on what we learned, we arrived at five principles for designing those early moments.

01

First impressions:

Deliver the “aha moment” up front

When onboarding leads with setup – connect accounts, set preferences, describe your role – users are asked to invest before they’ve seen what the agent can do. In our research study, sales managers who had to provide multiple knowledge sources before seeing any agent output found it hard to stay motivated through the setup.

The fix isn’t eliminating the setup. It’s reordering it. Show a useful result first, then ask for context once the user can see why it matters. If the agent needs knowledge sources, generate a baseline draft from minimal input first, then show how adding sources improves quality.

Sequencing matters:

Offer small steps with immediate payoffs

Principle 1 addresses when users see value during setup. But let’s also consider how they get there. What is the most effective structure and pace for each setup step? In our study, the sales agent required different knowledge sources and prospect data before it could simulate outreach by drafting emails. Asking managers to complete all of that made the onboarding feel overwhelming.

Breaking configuration into small actions with immediate payoffs keeps users moving forward. Designing partial completion, where the agent produces useful output even with missing inputs, means users aren’t blocked by incomplete setup. Defaulting to “save and resume” for longer flows, and showing what’s done and what’s next, keeps the momentum informational rather than performative. Let users see what the agent can do with what it has, then surface what could be unlocked with additional inputs.

03

Visible momentum:

Give clear signals that the agent understands and is working

When an agent takes time to process, the absence of feedback creates its own problems. We noticed that participants running simulations of the sales agent’s outreach had no clear signal of what the agent was doing during processing. A lack of progress cues led to repeated prompts and uncertainty about whether the system had understood the request.

Progress cues don’t need to be detailed. Even plain-language descriptions of what the agent is doing (“analyzing prospect data,” “drafting outreach,” “running simulation”) go a long way toward keeping users oriented. When users can see that the agent understands their request and is actively working on it, they’re less likely to retry or second-guess. It also helps to set rough time expectations where possible, and to make end states clear so users know whether something is finished, partially completed, needs their input, or has failed.

Personalization without friction:

Start with simple output that users can refine later

Agent output that reflects a user’s audience, constraints and preferences will always feel more useful than something generic. But the more context an agent asks for upfront, the heavier the setup feels. Although sales managers expected the agent to tailor outreach to their prospects and selling style, they understandably resisted lengthy configuration flows to make that happen.

Effective personalization happens through learned preferences, inferred context, and smart defaults, not lengthy onboarding conversations. Starting with sensible defaults lets users refine after seeing initial output rather than configuring before they’ve seen anything. Asking for context only when it will materially change the result, retaining it once provided, and showing the impact through before-and-after comparisons (for example, an outreach email with and without prospect-specific details), makes the value of sharing information concrete rather than abstract.

Set expectations early:

Show users where the agent can offer them the most value

Users are used to tools like Copilot and ChatGPT, and those experiences shape what they try first. In our study, sales reps assumed the agent could pull information from the web and answer general questions. But it was designed to draft and send sales emails autonomously, and onboarding was required to get there.

Showing a few example tasks upfront helps users understand what the agent is designed for. Keeping the language simple and action-oriented makes it easier to try. After each step, suggesting what to do next keeps users moving without needing a full walkthrough.

The compounding effect

Each of these principles builds on one another. When users see value early, they’re more willing to invest. When progress is visible, they don’t second-guess the system or disengage. When personalization builds over time instead of upfront, getting started feels lighter. Surfacing capabilities helps users engage with what the agent is actually built for. And when setup is broken into smaller steps, users keep moving instead of dropping off. Together, these shape whether the first five minutes build trust or erode it.

While these principles came out of onboarding research, they extend to any moment where users interact with an agent for the first time or where the agent’s behavior shifts. Every new capability, workflow, or behavior change is another first impression. The goal isn’t simply to explain how the agent works, but to help users gain trust, feel in control, and have confidence in taking the next step.

Further reading

• Microsoft Research. “Guidelines for Human-AI Interaction.” HAX Toolkit. https://www.microsoft.com/en-us/haxtoolkit/ai-guidelines/

• Microsoft Design. “UX Design for Agents: Microsoft Principles and Guidelines for Building Agentic Experiences.” April 2025. https://microsoft.design/articles/ux-design-for-agents/