For years, software design has been dominated by the SaaS model—dashboard-heavy, admin-focused applications packed with tables, forms, and validation flows. These CRUD apps (Create, Read, Update, Delete) are everywhere, powering businesses by letting users manage structured data.

But they come with an inherent problem: they’re expensive and complex to build. Every interaction requires a bespoke interface—custom reporting views, multi-step wizards, intricate validation rules. We, as designers and developers, have spent years crafting these systems to let users interact with their data in highly structured ways.

With AI becoming more sophisticated, I believe this paradigm is about to change. Instead of designing interfaces that let users manually input and manipulate data, we’re moving toward systems where users describe what they want to happen, and agents handle the rest. This isn’t just about chatbots—it’s a fundamental shift in how software works and how people interact with it.

The Future of Interfaces in an Agentic World

Interfaces aren’t going away—they’re just changing dramatically. As designers, we won’t be spending as much time crafting step-by-step wizards or static dashboards. Instead, we’ll be building more generalized, flexible environments that adapt to user needs in real-time.

Users will still need interfaces to interact with agents, but they won’t look like what we’ve been designing for the past decade. Instead of a fixed dashboard, applications might offer configurable workspaces, AI-generated summaries, or conversational interfaces.

Take an accounting SaaS, for example. There will still be an invoices section and a customers section—some structures will remain familiar. But these views won’t just be static lists. They’ll include AI-enhanced search that allows users to ask questions and surface insights. When viewing a customer profile or an invoice, there will still be well-designed overview pages, but right alongside them, users will have the option to interact with AI to extract key information or take suggested actions.

These shifts don’t eliminate UI—they make it more dynamic and responsive. We’re moving from designing rigid workflows to designing adaptable, AI-powered environments.

Designing for Agents: Emerging Patterns

Agentic design introduces new challenges and new interface patterns. Below, I’ll go through some of the key patterns that are already emerging and will likely define the next generation of interfaces.

1. Collaborative Workspaces

One of the biggest shifts we’re seeing is the move from form-driven interfaces to AI-assisted workspaces. Instead of rigid input fields and buttons, users collaborate with AI in a shared environment, where the system acts as a persistent assistant rather than a static tool.

Perhaps the best examples of this currently are ChatGPT's Canvas feature, and Cursor's side panel, both of which allow seamless collaboration between AI and human where both can make changes simultaneously and the interfaces helps to communicate and resolve conflicts.

A small side note here: Multiplayer interfaces like in Google Docs or Figma, where multiple users can edit at the same time, are an obvious place to see agents working as well – essentially behaving as part of your team. Though I've not seen anyone ship this just yet.

2. Transparency, Audit Trails, and Showing Work

A core challenge of agentic design is trust. If an AI is making decisions on a user’s behalf, the system must clearly communicate what it’s doing and allow for human oversight through showing reasoning behind actions and clear audit trails of what took place.

Showing Work – Interfaces must highlight AI-generated changes, similar to Windsurf's builder agent.

Showing Work & Reasoning – AI should visibly document the steps it took, whether scraping data, conducting research, or making structured decisions. Deepseek goes as far as explaining – at the model level – how it's approaching a given task.

4. Immutability and Confirmatory Interfaces

AI agents should avoid taking irreversible actions without user confirmation. Many emerging designs ensure that AI systems present their proposed actions for review before executing them.

Sources/Citations – Being able to see where an agent sourced its information from (be it your own data or a web search) is critical to building trust and ensuring audit ability. RAG (Retrieval Augmented Generation) is a key component here, allowing large datasets to be utilized by an AI in a way that's fully traceable.

Confirmatory Interfaces – When browsing or automating actions, the AI stops before executing irreversible steps, requiring explicit confirmation. ChatGPT Operator for instance asks a user to confirm before it takes any critical action in a flow such as clicking a 'buy' button or in the case below, following someone on X.

Operator coined some interesting new interaction patterns including the ability to 'take control' for the AI. I highly recommend trying it out.

Version Control & History – Storing past versions of records and being able to easily roll back changes made by AI ensures when inevitable errors occur, either a human or agent can rectify them.

5. Task and Notification Interfaces for Human Oversight

As agents become more proactive, users need structured ways to review AI-generated tasks and take action. I haven't seen any production-ready interfaces around these yet, but am excited to see where this goes.

Task Lists from AI – Instead of users manually working through processes, AI agents generate structured task lists.

Summaries & Change Reporting – An AI should be able to provide summaries of what it's done while you've not been looking

Where This Is Headed

The shift from traditional SaaS interfaces to agentic design is already underway. As AI systems improve, we’re going to see less reliance on rigid, manually-driven CRUD applications and more emphasis on dynamic, AI-assisted workflows. This will require an entirely new approach to design—one that prioritizes collaboration, transparency, and trust over static UI elements.

For designers, this means rethinking everything. How do we communicate AI intent? How do we make sure users stay in control? How do we build interfaces that feel like working with a competent assistant rather than fighting against an unpredictable system?

The future of interface design isn’t about more polished forms and dashboards. It’s about designing environments where humans and AI collaborate seamlessly.

If you have any ideas or suggestions for more patterns we should be looking into or adding to this article you can reach me on X @mrjeremyblaze