OpenCode ships 6 tab primitives in 9 days, then rebuilds the desktop around them

OpenCode Desktop v1.18.0 shipped yesterday. The "Introducing Tabs" copy is 4 sentences: OpenCode Desktop is now built around tabs. Start a new session in a tab. Open a new tab when you are starting something new, and close it when you are done. Download the latest to get started.

The work behind those 4 sentences is not short. Since July 6, OpenCode has shipped 6 desktop releases, plus the v1.18.0 launch and a v1.18.1 bugfix the same day. Every one of them is a tab primitive.

• v1.17.14 (Jul 6): "Added reopening for closed tabs and background tab opening." The undo button for tabs. The thing that makes tabs safe to use.
• v1.17.15 (Jul 7): "Show unread state on tabs with pending questions." Tabs now have the same attention affordance as a browser tab.
• v1.17.16 (Jul 9): "Routed the new session shortcut through tabs in the new layout." New chats go to tabs. Not the old single-session model.
• v1.17.17 (Jul 9): "Added a dismissible tabs intro popup and refreshed the help entry point." Onboarding assumes tabs are the default.
• v1.17.19 (Jul 13): "Added middle-click to open sessions in a new tab." Power user affordance. Shipped before the launch.
• v1.18.0 (Jul 14): Desktop v2 migration complete. Tabs as the architecture. A settings toggle to switch back to the old layout during the transition.

The pattern is not subtle. The new tab is the unit of work. Each tab is its own session, its own context, its own close button. The chat client is becoming a browser for AI work.

What the tab model actually changes

The single-session chat client assumed one long conversation per day. You titled it carefully. You scrolled back to find the context. You hoped the model still remembered what you were doing 4 hours ago. That model is dead.

The tab model assumes many short tasks. Research in one tab. Code review in another. Copy drafting in a third. Each tab closes when the work ships. Each tab opens with a clean context window. The mental cost of context switching goes down because the model itself does not carry state between tabs.

For founders and operators, the shift is sharper than it looks. The way teams actually use AI is parallel. Research, drafting, debugging, all running at once, in different windows of attention. The tool that supports that workflow wins. The tool that forces one long chat loses.

The catch

Three things to flag before you switch.

The old layout still exists. OpenCode shipped a transition setting. You can flip back to the single-session model if the v2 layout breaks your workflow. Most AI tools will not be that generous. Cursor still runs one chat per workspace. Claude Code still runs one chat per project. Codex is somewhere in between. If your team's AI workflow has been "one long session, carefully titled," you have a migration to plan, not just a feature to adopt.

The tab model requires a different mental model of context. With one long session, the model carries state. With tabs, you re-establish context each time you open a new tab. The cost is per-tab mental overhead. The benefit is context that does not bleed between tasks. Net is faster, but only if your team adapts.

The third thing is naming. Tabs need labels. Many of them, fast. If your team does not adopt a naming convention on day one, the tab bar becomes a graveyard of "New session - 2" titles within a week. Set the convention before the tabs multiply.

What we are doing this week

We turned on the OpenCode v2 layout in the YDN workspace this morning. The shift was immediate. We went from 2 to 3 long sessions per day to 8 to 10 short tasks. Each one closed when the work shipped. The convention we set on day one: tab title equals the deliverable, not the prompt. "Q3 Lifted report" not "weekly reports." "Babylon action plan" not "deck draft." The tab is the unit of work. The name is the unit of accountability.

Source