Why My AI Coach Doesn't Have Tabs

30-second version

Tabs are how non-AI software organizes capability. AI software does not need them — the conversation is the index. Removing tabs from an AI coach product was the single highest-leverage UX decision: it forced every feature to be invokable from chat, which forced every feature to be designed as something a user would actually ask for. Half the features failed that test and got cut. The remaining half are all the product needed.

The default that nobody questions

Every SaaS product I’ve used in the last decade has the same shape:

A sidebar (or top nav) with the major modules
Each module is a “page”
Each page has its own state and its own information density
Settings is a separate area; help is a separate area

This is correct for non-AI software, because the user has to discover what the software can do before they can use it. Tabs are a discovery mechanism.

AI software inverts this. The user types what they want. The system either does it or doesn’t. There is no discovery problem to solve at the navigation layer — there is a discovery problem to solve at the conversation layer.

I built the first version of my AI interview coach with the standard shape. It had:

A “Practice” tab where you picked a question type
A “Mock Interview” tab where you ran a full mock
A “Dashboard” tab with your scores and trends
A “Profile” tab with your background
A “Settings” tab with… settings

It worked. It also felt exactly like every other AI product, which is to say, it felt like a CRUD app with a chat box bolted on.

The decision

The change was three sentences in a design memo:

No tabs. No navigation. The conversation is the entire product. Everything else is invoked from inside the conversation, or not at all.

What that meant in practice:

“Practice a question” → user types “I want to practice estimation” in the chat. The agent finds a question and starts.
“See my dashboard” → user types “How am I doing on system design this week?” The agent fetches the data and replies in the conversation.
“Update my profile” → user mentions “I just got an offer at a fintech startup,” and the agent updates the profile inline.
“Mock interview” → user types “Can we do a 30-min mock for a senior PM role?” The agent triggers the mock mode in the same window.

No “Practice” page. No dashboard page. No settings page. One conversation. Everything inline.

Why this is harder than it looks

Three things I underestimated:

1. Every “feature” had to justify itself. When you remove tabs, features can no longer hide behind navigation. A feature only exists if there is a natural utterance that invokes it. Half the features in the v0 design did not have a natural utterance. They got cut.

The features that did pass the test are the only ones that mattered anyway. Removing the dashboard page did not remove the dashboard data — it just forced the data to surface when the user asked for it, which is the only time the data actually matters.

2. State management gets harder, not easier. With tabs, each tab owns its state. Without tabs, the conversation owns all state, and state has to flow through messages. This required rebuilding the agent loop with a state machine that tracked where in a multi-step flow the user was — practicing? in a mock? mid-feedback? — and disambiguated user intent based on that state.

The reward for doing this is huge: the user never has to context- switch. The cost is real: you have to design the state machine carefully and the agent has to be aware of it.

3. You lose the easy “more features” axis. Tabs let you bolt on new features by adding a new tab. Without tabs, every new feature either fits an existing utterance pattern or has to introduce one, and introducing a new utterance pattern means teaching the user something new — in the conversation — which has a much higher cost.

This forces you to be deliberate about new features. That is not a cost. That is the point.

What this unlocked

The interesting effects of removing tabs:

Onboarding compressed to one screen. No tour, no walkthrough, no tooltip cascade. The first message tells the user what the product is and asks what they need.
Feature density per square inch went up. The conversation is vertical and infinite. There is no UI real-estate competition. The agent can produce a structured feedback report inline that, in the tabbed design, would have lived behind a “Reports” tab.
Mobile became trivial. Tabbed products require navigation redesigns for mobile. A chat product is already mobile. We shipped mobile in two days, not two weeks.
The product feels like a person, not a tool. This is the part I cannot quantify but felt the most. Users describe the product as “her” and “she” within the first session. They never did with the tabbed version.

The design pressure this imposed

Removing tabs is not a UI choice. It is an architecture choice. It forces:

Every feature to be a tool the agent can call from the conversation
Every state transition to be inferable from natural language
Every output to be renderable inline (not “click here to see your dashboard,” but “here’s how you did this week” with the chart embedded)
Every decision the agent makes to be explainable in the same conversation, because there is no separate “logs” tab to show why

This is the work most AI products avoid by hiding capability behind tabs. The tabs let the underlying system stay incoherent. Removing the tabs forces the underlying system to actually be a single coherent thing.

When this is wrong

Two cases I can think of where tabs are genuinely better:

Multi-user collaboration. Slack, Notion, Linear — the tabs are not navigation, they are contexts (this channel vs that channel, this doc vs that doc). The chat is the tab.
Visual creation tools. Figma, Photoshop. The tabs are tools, not destinations. A chat-only Figma would be slower for everything that matters.

For an AI coach — a 1-on-1 conversation with a stateful agent — none of those apply. The default should be no tabs.

What other AI products are doing

Most production AI products in 2026 still inherit the SaaS sidebar + tabs layout: ChatGPT (Plus/Team) keeps a sidebar with chat history and GPT directory; Claude.ai is moving toward a chat-as-hub but still has project pages; Perplexity keeps tabs for spaces, history, and discovery. The chat-only architecture I describe is a product-form choice, not a technology constraint — the technology supports it, the product habits don’t yet. I bet on the chat-only direction because I think tabs will look like browser bookmarks-bars look today in three years: a vestigial UI element from a previous era. Cross- reference: Agent Framework Landscape.

How I would pitch this in an interview

“How did you decide to remove navigation?” — sometimes asked, more often the interviewer just notices the product is unusual.

The answer:

Tabs are a discovery mechanism for software the user doesn’t yet know how to use. AI changes that — the system asks the user what they need. So tabs become friction. We removed them. The forcing function was that every feature had to justify itself by being invokable from chat. Half didn’t. We cut those. The remaining half is the product.

That answer is short, opinionated, and demonstrates the willingness to cut features. All three are rare. The interviewer remembers the candidate who cuts.