PostHog made chat its default homepage. Linear embedded an AI agent across the app. Attio moved further towards agent-first interactions. Three of the most design-aware SaaS products in the market, all pushing conversational AI closer to the centre of the experience.
The response was immediate. Users did not want to trade dashboards, visual density and direct control for a blinking cursor. That reaction made sense. But the instinct behind the shift still holds. Traditional interfaces cannot serve every user in the same way, all the time. The opportunity is not to choose between chat and dashboard. It’s to build a new kind of interface, where AI shapes the experience around the user, and the design system governs how that experience looks, feels and behaves.
The chat bar backlash is not about AI resistance.
When PostHog made chat its default homepage and Linear introduced an AI agent throughout the product in early 2026, the reaction was loud. Users were not struggling with the idea of AI. They were reacting to what had changed.
A strong dashboard lets you take in a large amount of information at speed. Colour, scale, position and proximity all work together. Your eye spots the change before you have fully named it. That is the strength of visual density. It helps people process complexity quickly.
But a chat bar changes that dynamic. Instead of scanning a view, you need to know what to ask. Then you wait. Then you read. Then you piece together context across a series of responses. Information that once landed in seconds starts to unfold over minutes.
Users also lost direct manipulation. Filtering. Dragging. Clicking into a metric to explore what sits beneath it. Chat turns interaction into a sequence. You type. You wait. You read. That structure works for some tasks, but not for all of them.
The frustration here is grounded in the experience itself. Users lost tools that helped them move quickly and confidently. In return, many were given a text box with little context, little structure and no obvious starting point.
But the instinct is right.
The chat bar is a weak answer to a strong question.
Traditional SaaS interfaces come with a built-in constraint. Views are predefined. Dashboards are hardcoded. Workflows reflect what the product team anticipated a typical user might need. If you want a different way of seeing your data, someone has to build it first.
Agentic systems change that. They reason across your data, your tools and your context. They can connect actions that used to require multiple clicks, filters and saved views. What a user can do is no longer limited in the same way by what has already been shipped.
That is a meaningful shift in capability. The issue is not the intelligence. It is the interface. A chat bar gives you flexibility, but often at the cost of the qualities that made traditional interfaces effective. The information may still be there. The experience becomes harder to use.
Chat is stage one, not the destination.
The evolution of agentic interfaces has a clear direction. What we are seeing now is the starting point.
- Stage one is plain text chat. You ask. The system replies. Usually in text, sometimes with light formatting. The experience sits entirely inside a conversation thread.
- Stage two introduces inline generative UI. The system still responds in a conversational flow, but now with charts, tables, forms and interactive components appearing within it. The output starts to become visual and usable, not just descriptive.
- Stage three moves further. Chat becomes a way to create persistent views. You ask for a dashboard, a report or a workflow, and the result lives beyond the conversation itself. The interface begins to take shape from the request.
- Stage four is where this is heading. Embedded agentic UI. The interface composes itself around the user. Based on role. Behaviour. Context. Data. Chat remains available, but it no longer needs to carry the full experience. It becomes the support layer, not the main stage.
That is the longer-term opportunity. Software that adapts around the way you work. Reaching it means solving a few design problems that chat-first products often skip over.
Three problems to solve.
The blank page problem.
Many chat-first products begin with the same empty state: a blinking cursor without context, prompt or guidance. Traditional products solved this with dashboards, templates and guided onboarding. Chat often removes those cues. A better starting point uses the context the system already has. Role. Data. Usage patterns. The interface should begin with something useful and evolve from there.
The visual density problem.
Chat lowers information density. You cannot scan a conversation in the same way you scan a well-designed dashboard. Patterns, outliers and relationships become harder to spot when they are spread across sequential text. The answer is not more chat. It is a canvas-based layout where generated charts, tables and views take priority, while chat supports them from the side.
The brand and consistency problem.
If AI is generating the interface, the challenge is keeping the product feeling like itself. Different users may see different views, but the system still needs coherence. This is where design systems matter. The agent should not generate free-form UI. It should compose from a defined set of components, tokens and behaviours. That gives it room to adapt while keeping the experience recognisable and consistent.
What we recommend.
If you are integrating agentic AI into a SaaS product, start by protecting what already works.
Don’t just replace your interface with a chat bar.
Users chose your product partly because of how it behaves. Replacing that with a text box does not move the experience forward. It strips useful structure away. A stronger model embeds AI into the existing product, making it accessible when needed rather than placing it in the way of everything else.
Design hybrid interfaces.
Keep the core UI in place and introduce agentic capabilities alongside it. Contextual assistance. Inline suggestions. Smart actions. The system becomes more capable without asking users to abandon familiar patterns.
Build your design system for composition.
Your components need to support dynamic assembly, not just manual use by designers and developers. That means clearer rules, semantic tokens, constraints and specifications that a machine can understand as well as a person.
Use progressive disclosure.
Not every user needs agentic functionality straight away. Introduce it where it adds value. Let it appear when it helps someone move faster, see more clearly or complete a task with less effort.
Invest in contextual onboarding.
If chat is part of the experience, it should never feel empty. Show users what it can do through relevant starting points, prompts and examples. Let the system demonstrate value before it asks for trust.
Test with sceptics, not just enthusiasts.
Internal teams and power users will often adapt quickly. The more revealing perspective comes from users who do not think in prompts and queries. If the experience works for them, it is more likely to work at scale.
Design systems are the infrastructure for this.
This is not only a product design challenge. It is a design system challenge too.
Traditional design systems are built for people. Static component libraries. Documentation. Guidance for designers and developers. That model starts to stretch when an AI agent is composing the interface.
Agentic design systems need more structure.
They need machine-readable component specifications that describe what each component does, what data it takes and when it should be used.
They need semantic token layers that connect visual decisions to meaning, so the system can compose with intention rather than just assemble shapes on a screen.
And they need governance rules for composition. Which components belong together. Which layouts make sense in which contexts. How density, spacing and interaction patterns should adapt.
This is where investment pays off. Teams building design systems that are ready for agentic composition now will be in a stronger position to create AI-powered experiences that still feel native to the product. Without that work, generative interfaces risk feeling disconnected from the brand they are meant to serve.
The interface is still the brand.
It can be tempting to treat agentic UI as a pure technology problem. Something for engineering to solve. Something the model will work out on its own.
That misses where the real experience lives.
The interface is still the place where your users meet your brand. It is where trust builds. Where clarity lands. Where confidence grows or weakens. Every view the agent composes and every interaction it enables becomes part of that perception.
If the generated interface feels generic, your product feels generic. If it feels inconsistent, your product feels harder to trust. If it removes control, users feel it immediately.
The teams that will move this forward are the ones treating agentic interface design as a design system and brand challenge as much as a technical one. The technology is advancing quickly. The experience needs to keep pace.
Let’s get things Done & Dusted.
Agentic AI can make SaaS products more adaptive, more helpful and more powerful. But only when the interface holds together. Around the user. Around the task. Around the brand.
That’s where we come in.
We are a brand design agency working with SaaS and technology teams to design systems and digital experiences built for now and future-ready. From component libraries and token architecture to platform UX and interface design, we help you create products that evolve without losing coherence.
If your product is integrating agentic AI and you need a design system that supports dynamic composition while keeping the experience clear, consistent and unmistakably yours, we can help.
Contact us today and let’s assess what your brand needs to take its next step.
Ready when you are.
