all work

01Emergent · Deploy experience · 2026

From a deploy popover to the Manage panel

I changed how users publish and manage apps on Emergent. Emergent is an AI app builder. Users make full-stack apps through chat and then publish them.

role
Product designer
company
Emergent (emergent.sh)
period
May to September 2026
scope
Deployment popover, agentic publish flow, Manage panel IA, actions during publish
output
Interactive HTML prototypes, job stories, IA specification

problem

The deployment panel did not scale. It started as a small popover. Then users needed domains, resources, database, secrets, billing, logs, and alerts. All of these went into five tabs in a narrow side panel.

  1. 01

    A publish takes 5 to 8 minutes. The popover did not let users leave and come back. Users did not know if the deploy continued after they closed it.

  2. 02

    Then Emergent added an agentic deployer. The agent does each step and repairs failures itself. The old UI showed only fixed steps. It could not show an agent at work.

  3. 03

    The items in the panel also overlapped. Stripe had two entry points. Keys did not show after users added them. Secrets, security, and integrations had no clear boundary.

+ media
The old deployment popover: five tabs in a narrow side panel.

goals and constraints

goals

  • Show the deploy status at all times, also when the panel is closed.
  • Give one location for all tasks after publish.
  • Show the work of the agent in a way that users can trust.
  • Keep the default view minimal. Show more detail only when users ask for it.

constraints

  • A publish takes 5 to 8 minutes. If a step fails, the agent starts again from step one.
  • Each change to a domain, secret, or database starts a new publish.
  • Free, Standard, and Pro users see different items.
  • The panel must work before and after the first publish.

process

The work had four phases. Each phase showed a problem that the next phase solved.

01Divide the popover by user intent

I defined four states: before deploy, during deploy, success, and after deploy. Each state has a different user intent: decide, wait, celebrate, manage. Each state got its own level of detail and tone.

  1. before deploydecide
  2. during deploywait
  3. successcelebrate
  4. after deploymanage
  • The time estimate shows before users start the publish.
  • The popover minimizes to a pill that stays on the screen.
  • If users close the popover, the deploy continues. The UI tells them this.

resultManagement after deploy did not fit in a popover. It needed its own surface.

+ media
The four popover states: before, during, success, and after deploy.

02Make the Manage panel

I moved Manage Deployments out of the side panel. It became a full surface with a sidebar. One structure holds the states before and after publish:

Appoverviewdomains
Connectintegrationspayments
Configsecretssecurity

Items become active when they are necessary. They do not move after launch. Secrets is the only store for keys. Integrations write their keys into it.

+ media
The Manage panel: one sidebar for the states before and after publish.

03Show the agent at work

The Publish button changes into a loader. A monospace line shows the current tool call. “View deployment logs” opens the full agent log. A back chevron goes back to the compact view.

When the agent repairs a failure, the step shows in amber. The agent explains the repair in plain words. Users can also publish from chat. The agent then posts a live status message in the thread.

+ media
The publish loader with the live tool call, and the full agent log.

04Control actions during publish

A domain link starts a publish of 5 to 7 minutes. Any other change during this time starts one more build. I made publish a background state, not a mode.

  • A banner stays on all tabs.
  • Only controls that change the app are locked. A tooltip gives the reason.
  • Read-only tabs, such as Analytics and Monitoring, stay fully usable.
  • The domain control shows “Linking…” progress in its position.
  • A timer shows the time used, not the time remaining.
+ media
Publishing in the background: the banner, locked controls, and “Linking…” in place.

key decisions

Manage gets its own surface
Five tabs of settings are a workspace, not a popover.
One structure before and after publish
Users do not learn the navigation again after launch.
Agent repairs show in amber, not red
The agent fixes the problem. Red tells users that the deploy failed.
Lock only the controls that change the app
A dimmed panel looks broken. Users can still read their data.
Timer shows time used
The agent can start again from step one. A countdown can then be incorrect.
Secrets show pending and live values
Users see what changes at the next publish.
Agent placement follows the stakes
Above the composer when the agent waits for the user. In the thread when the agent does the work. Full screen only for money and go-live.
Domain setup tries automatic linking first
Auto-Link through OAuth is the first tier. Guided manual DNS is the second tier.

outcome

The deployment popover became the Manage panel. The panel now holds deploy, domains, integrations, secrets, database, and billing in one structure. The agent shows its work in the panel and in chat.

to addWhat shipped and the results. For example: publish completion rate, support tickets about deploys, time to link a custom domain.

open questions

  • When the agent starts again from step one, should the progress bar go back to zero? This shows the truth but can alarm users.
  • Where do agent approval requests show: above the composer, or in the panel?

what i learned

  1. 01

    A long task needs a status that stays visible. The status must not depend on the surface that started the task.

  2. 02

    Agent UI needs a small set of patterns. Five patterns covered all deploy actions: handle in the thread, pick one, fill in a value, confirm, and report status.

  3. 03

    Show the truth, but set the correct level of alarm. Users must see what the system does. Color and copy tell them if they must act.