Think about launching a brand new AI agent to deal with key buyer requests. It seems smooth, is built-in into your workflow, and it largely works. However, then you definately discover that adoption is sluggish, customers aren’t getting what they want, and the agent generally fails to ship on its promise.
We’ve been there.
At Salesforce, we use our inside Agentforce platform to energy the whole lot from gross sales assist to billing questions. And like lots of our prospects, we’re constructing these experiences in actual time, studying what works – and what doesn’t – as we go. Whereas my staff focuses on creating brokers for our staff to make use of, a number of the insights we’ve collected in our inside Agentic Interface Tips can assist our prospects of their efforts to implement Agentforce.
Right here’s our discovery course of and learnings:
Use Agentforce to be taught
Draw from collective design rules
Apply 5 suggestions for higher AI agent design
Preserve a newbie’s mindset
Use Agentforce to be taught
As “buyer zero,” Salesforce makes use of its personal merchandise not solely to run the corporate but in addition to find out how we would enhance the expertise for finish customers. Particularly for Agentforce – our service that helps corporations construct and deploy AI brokers at scale – it’s essential to have a suggestions loop that informs how we constantly refine our merchandise.
AI brokers are clever, task-oriented assistants that may make choices, automate workflows, and work together with customers. They can assist customers draft emails, floor key information, reply questions, or escalate requests all inside your CRM. However constructing a useful AI agent is extra than simply including a mannequin behind a button. It’s a design problem.
For instance, when my colleagues in buyer success noticed that the Agentforce implementation on Salesforce Assist wasn’t performing in addition to anticipated, they reimagined the expertise design – and launched it in 5 days.
There’s a lot to study the best way to design for all of the potential methods agentic AI can be utilized.
Draw from collective design rules
Discovery is a key step in design. To that finish, we relied on the breadth and depth of person expertise (UX) and product designers throughout our firm. These groups develop requirements and assist preserve consistency, usability, and accessibility throughout all merchandise. From visible components, like colours and typography, to interplay patterns and content material tips, our designers are key to constructing nice experiences for our staff and prospects.
For our functions, there wasn’t a have to reinvent the wheel. We drew from current sources to assist form the steering that will serve our particular use instances. For instance, we borrowed from the extensively accessible Agentic Experiences Patterns from Salesforce Lightning Design System 2 (SLDS 2 beta), in addition to Salesforce branding and advertising model guides, supplies from Slack, and agentic expertise patterns from our web site design system.
In developing with a framework for our staff to design simpler, intuitive, and reliable AI brokers, we found some insights we wished we had identified earlier.
Apply 5 suggestions for higher AI agent design
We’ve codified our learnings into an inside design framework we name the Agentic Interface Tips. Under are 5 actionable insights that proceed to form how we design agentic experiences and how one can enhance your personal.
1. Restrict agent names to 10 characters to spice up readability and consistency.
Naming appears easy till it isn’t. In our early builds, lengthy or imprecise names broke layouts, cluttered dropdowns, and confused customers. We established a naming conference: use “Identify + Agent,” and preserve the primary half below 10 characters and one-to-two phrases.
Brief, scannable names work higher throughout UI elements and enhance recognition. We additionally realized that specificity issues: “Ops Agent” is perhaps too broad, whereas “Billing Agent” instantly communicates goal. Clear naming means sooner belief.
Character limits assist guarantee consistency. [Salesforce]
2. Use heuristics to identify design issues early.
To judge the standard of our brokers, we leaned on heuristics, or easy analysis strategies, to catch usability points. We tailored a set of inside agent high quality heuristics to deal with the next:
- Consistency: Do brokers comply with interplay and language patterns customers anticipate?
- Error tolerance: Can customers recuperate when the agent offers a nasty response?
- Teachability: Can customers rapidly perceive what the agent does (and doesn’t do)?
With out these, we constructed brokers that seemed good on paper however confused actual customers. Heuristics helped us ask the proper questions early and keep away from expensive rework later.
3. Outline the construction of agent conversations to enhance usability.
What an agent says and when issues. We established content material tips to present every agent a transparent, constant conversational construction. Each agent expertise now features a:
- Welcome message that units expectations.
- Disclosure assertion to construct belief.
- Listing of instance questions to information customers.
- Clear scope boundaries for what it will probably or can’t do.
- Path to give suggestions or escalate.
This construction reduces friction, helps customers get began sooner, and builds confidence in what the agent can deal with. A clean enter discipline is intimidating; guiding content material helps take away guesswork.
Defining the construction of an agent and offering disclosure textual content enhance the person expertise. [Salesforce]
4. Embody disclosure textual content to make sure transparency.
AI isn’t excellent, and customers understand it. However we nonetheless have to say it.
Each agent, particularly external-facing ones, consists of commonplace disclosure language like:
“I could make errors, so overview my responses for accuracy.”
Place this assertion within the welcome message or simply under the enter discipline. Don’t bury it.
For inside brokers, we adapt this message based mostly on the viewers’s familiarity and danger tolerance. These disclosures aren’t simply authorized cowl, they assist set sensible expectations and construct person belief.
5. Use constant branding to sign agentic experiences.
Constant “feel and look” assist floor the person within the context. [Salesforce]
Branding is greater than logos it’s how customers acknowledge and belief the instruments they’re utilizing. We developed a set of visible identification tips particular to agent experiences, overlaying:
- Icons: Differentiate between advertising visuals and in-app agent avatars.
- Shade utilization: Guarantee accessible, constant palettes.
- Voice and tone: Align with our model whereas being clear and conversational.
We’re additionally phasing out legacy icons and standardizing agent avatars to create a constant expertise throughout all surfaces. A constant look alerts to customers: “You’re speaking to an AI agent and right here’s what to anticipate.”
Preserve a newbie’s mindset
Whether or not you’re designing AI brokers for inside use or customer-facing assist, these design rules can assist you keep away from widespread pitfalls and construct with confidence.
Our personal course of remains to be evolving however we consider in studying out loud.
Again to the highest
