AI has become part of the modern software stack, and isn’t going anywhere. Product managers and designers have to rethink how users interact with SaaS products. The classic model—clickable buttons, forms, toggles—is still relevant and reliable. But it’s no longer enough.
In this article, we’ll look at how to design SaaS interfaces that feel natural in the age of AI.
This article has been written in collaboration with Nik Grishin, a CPO at Albato and product design expert with more than 6 years of experience. Enjoy!
From UI to outcome-oriented interfaces
Traditional SaaS interfaces rely on visual controls, such as dashboards, buttons, and input fields. Some of these interfaces are designed better, some worse.
However, the core value for users has always been the outcome. Good UI design can be one of the ways to simplify the customer journey. But AI can offer new, exciting possibilities.
In many cases, users don’t want to figure out how your product works or where to press to get the results. They want the work done for them. That’s why we’re seeing a shift toward two core interaction models:
- Classic UI-based (visual) interfaces. Products are built around fixed user flows, static dashboards, and clearly defined menus. The priority is to create intuitive navigation and clear instructions. Every task—whether generating a report, scoring leads, or drafting content—required a series of clicks and forms. Personalization is minimal, often limited to role-based dashboards or saved settings.
- Command-driven (textual) interfaces. Design is no longer about guiding users through pre-set steps, but about anticipating their needs and removing friction entirely. AI copilots, assistants, and agents change the software from a tool into an active partner. A marketing manager can now type “Create a performance report for last month’s campaigns,” and get results instantly—no configuration required. \
Both approaches will likely coexist. Think of it as “interface duality”: where a user can either click through tasks or simply ask for what they need.
Let’s take an example: modern developer tools like Cursor. It combines a code editor with a chat interface. The chat is not a help tool—it’s part of the UI. You can type “Refactor this function” or “Create a login page,” and it does the work. You’re not collaborating with AI. You’re delegating.
We’re seeing this pattern spread to many other types of software.
Design tools like **Lovable **use a split screen: the left side is where you describe the outcome, right side shows the generated result.
Workflow tools like Albato are integrating side-panel assistants. You can ask the assistant to “Create a data sync between X and Y.” The system generates a draft workflow, which you can then refine. This is the feature that we’re planning to add in the future.
Your users are likely already familiar with this approach in other products and expect your SaaS to offer the same level of functionality.
What makes this work: Structured data + system design
For AI interfaces to deliver real business value, they must:
-
Have access to **well-organized, relevant data
-
Use agents, triggers, and actions that are tightly scoped
-
Include evaluation layers to review outputs before surfacing them to users or systems
That last point is critical. Every AI agent or assistant should be paired with an evaluator—one generates the outcome, the other validates it. This dual-agent model ensures reliability and builds user trust. It should become the default pattern for AI-first SaaS products.
Key principles of design in the AI era
In this section, we provide a summary of what you need to keep in mind when building your SaaS product:
1. Data readiness
AI only works well if your data is clean and organized. For example, if you run an email marketing platform, your AI can only give accurate send-time recommendations if the customer engagement data is properly collected and labeled.
2. Continuous feedback loops
AI gets smarter when users give feedback. Small things like a thumbs up/down on a recommendation or a quick correction help the system learn. For example, if a sales assistant bot suggests the wrong lead, letting the user mark it as “not relevant” teaches the AI to do better next time.
3. Personalization at scale
AI can customize the experience for every user automatically. It can adjust dashboards, emails, or even in-app tips based on how users interact with the product. For instance, a marketing dashboard might highlight ad metrics for one user and email performance for another, depending on their role.
4. Automation
AI takes over repetitive tasks, freeing teams to focus on strategy and innovation. Imagine a CRM that automatically scores leads, schedules follow-ups, or generates reports — all without manual work.
How to design SaaS in the AI-first era
The shift to AI-first design isn’t just about adding a chatbot or recommendation engine. It’s about building smarter workflows that make users feel like the product understands their needs. For product managers, this requires a clear AI implementation strategy from day one.
Here are five key action steps we’ve learned from experience.
1. Start with AI in mind
Too many teams treat AI as a “bonus feature” added after the product launch. This approach rarely works. To create real value, AI needs to be part of the product vision from the start.
Tip: Before adding AI, map the user journey and identify friction points. AI should remove complexity, not add it.
2. Invest in data
AI is only as good as the data behind it. If the data is incomplete, messy, or poorly structured, the AI will give poor results. We learned early on that data readiness must be a top priority.
Without this preparation, the AI can’t deliver relevant or trustworthy suggestions.
Practical steps for PMs:
-
Make data collection a product feature, not just an analytics task.
-
Ensure that all events, actions, and user behavior are tracked in a standardized way.
-
Set up dashboards that monitor data quality in real time.
3. Keep users in control
AI should never feel like a “black box” that makes decisions without user input. We’ve found that users trust AI more when they have easy ways to accept, reject, or tweak its suggestions.
Key idea: Always offer a fallback. AI should propose, not impose. Even a small feature, like a thumbs-up or thumbs-down on a recommendation, can provide valuable feedback for both the user and the AI model.
4. Test and improve continuously
AI models are not static. They need constant tuning and validation. At Albato, we approach every AI feature like an experiment. We launch small pilot tests, collect feedback, and refine the model before rolling it out to all users.
Our process includes:
-
Running A/B tests to measure whether AI-driven features improve key metrics (conversion rates, time saved, or user satisfaction).
-
Analyzing user behavior to see how often AI suggestions are accepted or ignored.
-
Collecting qualitative feedback from beta testers to uncover unexpected pain points.
5. Focus on privacy and ethics
Trust is the foundation of any AI-powered product. Users want to know how their data is used and that it’s protected.
Best practices for PMs:
-
Add clear disclosures when content or recommendations are AI-generated.
-
Offer users an option to opt out of AI personalization if they choose.
-
Ensure compliance with privacy regulations like GDPR by default, not as an afterthought.
Summing up
Designing SaaS products in the AI era isn’t just about adding a chat box. It’s about rethinking what the interface is. In the end, users want less friction and more results. Whether through buttons or plain language, the goal is the same: minimize mental overhead and maximize business value.
Companies like Albato are already embracing this shift. By natively integrating AI assistants into their core UI, they’re not just experimenting with AI—they're redesigning the user experience around it. It’s not a feature layer. It’s the product itself evolving.
Read more:
