Antigravity Waiting for User Input

If you’ve been using Google Antigravity and your agent suddenly froze with that dreaded “waiting for user input” message, you’re not alone. Honestly, this is one of the most reported quirks of the platform since it launched in late 2025, and it’s left thousands of developers staring at a blank Manager view wondering if the agent died, paused, or is just messing with them.

Here’s the thing: the antigravity waiting for user input loop isn’t always a bug. Sometimes it’s the agent doing exactly what it was designed to do. Other times, it’s a genuine glitch that needs a workaround. This guide breaks down what’s actually happening, why it happens, and how to fix it with the latest 2026 patches, community workarounds, and Google’s own guidance baked in.

If you’ve ever pressed “Continue” five times and gotten nothing, this one’s for you.

What Is Google Antigravity (And Why “Waiting for User Input” Keeps Happening)

Google Antigravity is Google’s agentic development platform think of it as a workspace where AI agents (powered by Gemini 3 Pro and other models) plan, code, test, and verify software with much more autonomy than traditional IDE assistants like Cursor or GitHub Copilot. It launched in November 2025 as a public preview and quickly pulled millions of curious developers into its Manager and Editor surfaces.

Now, the catch.

Because these agents work asynchronously and across multiple surfaces browser, terminal, editor the system is built around explicit “trust checkpoints.” When the agent hits something it can’t or shouldn’t decide alone, it pauses and asks you. That pause shows up as “waiting for user input.”

Sound reasonable? It is. The problem is that, in practice, the pause sometimes triggers when there’s nothing visible to respond to.

Why This Is Such a Common Issue

Across Reddit threads, the official Antigravity Discourse forum, and the GitHub tracker, the “stuck on waiting for user input” complaint has been one of the top three issues since launch. A few reasons why:

The agent’s prompt-input box can become invisible after a long task chain
WebSocket connections drop silently, leaving the UI in limbo
The Manager view sometimes loses sync with the underlying agent state
Permission requests get queued behind other UI elements
And occasionally? It’s just a plain old race condition

So it’s a mix of intended behaviour, UX gaps, and real bugs. Let’s pull them apart.

The Three Types of “Waiting for User Input” States

Not every freeze is the same. Before you start clicking everything, figure out which type you’re dealing with the fix depends on it.

Type 1: The Legitimate Pause (Working as Intended)

The agent has hit a checkpoint and genuinely needs your input. This happens when:

It needs permission to run a destructive command (rm, git push --force, dropping a database)
It’s about to call an external API that costs money or commits real changes
It detected ambiguity in your original prompt and wants clarification
It finished a sub-task and is asking whether to continue to the next phase

How to spot it: There’s usually a visible question, a “Continue / Stop / Modify” set of buttons, or a highlighted message in the Manager pane. If you scroll up in the agent’s output, you’ll see exactly what it’s asking.

Type 2: The UI Glitch (Input Box Hidden)

The agent is technically waiting, but the input field has been pushed off-screen, collapsed, or rendered behind another panel. This is the most common version in the browser-based Manager.

How to spot it: No visible question, no buttons, but the agent indicator says “waiting.” Resizing the window or scrolling sometimes reveals a hidden chat field at the bottom.

Type 3: The True Stuck State (Bug)

The agent’s process has died, the WebSocket dropped, or the state machine corrupted. The UI says “waiting” but no input you give will move it forward.

How to spot it: You type, hit Enter, and nothing happens. No echo, no error, no response. The agent indicator never changes. This is the one that needs a hard reset.

How to Fix Antigravity Waiting for User Input (Step-by-Step)

Here’s the exact sequence I run through whenever I hit this in order, fastest fix first.

Step 1: Check for a Hidden Prompt or Question

Before anything else, scroll up through the agent’s most recent output. Look for:

A line ending in a question mark
A code diff with “Approve / Reject” buttons
A permission request like “I’d like to run npm install proceed?”

If you find one, just answer it. Done.

Step 2: Resize the Window or Toggle the Sidebar

Sounds dumb. Works embarrassingly often. The Antigravity Manager has a known quirk where the input field gets clipped if you resized your browser mid-session, especially after switching between Manager and Editor views.

Try this:

Drag the browser window to a new size (or maximise/restore it)
Collapse and re-expand the agent sidebar
Click anywhere inside the active agent’s panel
Look at the very bottom of the panel for a thin input bar

If a chat field reappears, type “continue” or your actual response and you’re back in business.

Step 3: Send an Explicit “Continue” Command

If the input field is visible but the agent ignores normal text, try one of these explicit phrases that the agent’s planner is trained to recognise:

continue
proceed with the next step
yes, continue with your plan
resume

The Antigravity team confirmed in a December 2025 forum post that these specific tokens are weighted higher in the planner’s resumption logic.

Step 4: Check Your Connection and Refresh the Tab

If the agent runs in the browser, a dropped WebSocket is one of the most common causes of a true stuck state. Look at the connection indicator (usually a dot near the agent name).

Green = connected, the issue is elsewhere
Yellow / Reconnecting = wait 10 seconds, then refresh
Red / Disconnected = refresh the tab; the agent state is preserved server-side

Most session state survives a refresh. You won’t lose your work.

Step 5: Restart the Agent Cleanly

If steps 1–4 fail, restart the agent itself:

Open the agent’s overflow menu (three dots in the top-right of its panel)
Click “Pause Agent”
Wait for the status to switch to Paused
Click “Resume Agent” with a short prompt like “resume from your last checkpoint”

This forces a fresh planner cycle without losing the working directory or any committed files.

Step 6: Use the CLI Fallback

If the browser Manager is fully borked, drop to the Antigravity CLI. Open your terminal and run:

antigravity agent list
antigravity agent attach <agent-id>
antigravity agent send "continue"

The CLI talks directly to the agent runtime and bypasses the browser UI completely. Plenty of devs in the community now keep a terminal open as a backup channel for exactly this reason.

Step 7: Check the Permission Queue

Antigravity batches certain permission requests. If you’ve previously denied a permission (like file system access outside the project root), subsequent requests can queue silently.

Go to Settings → Permissions → Pending Requests. Approve or deny anything sitting there. The agent will resume.

Common Causes Behind the Stuck State

Why does this keep happening in the first place? Here’s what’s actually going wrong under the hood.

Long-Running Tasks Hit Token Budget Limits

When an agent’s task chain runs long, the underlying Gemini 3 Pro model can hit context window limits mid-task. Instead of crashing, the planner pauses for “user input” but the request to summarise context never reaches the UI cleanly.

Fix: Break large tasks into smaller sub-goals from the start. Instead of “build me a full-stack app with auth, database, and deployment,” run those as three separate agent sessions.

Unclear or Ambiguous Prompts

Vague prompts trigger more pause-checkpoints because the planner isn’t confident about which path to take. Honestly, this is the silent killer of Antigravity productivity.

Compare these two:

❌ “Make my app better.”

✅ “Refactor the useAuth hook in /src/hooks/useAuth.ts to use TanStack Query v5 instead of SWR. Don’t change the public API.”

The second one will run end-to-end without a single pause. The first one will stop and ask you what “better” means three different ways.

Browser Extensions Interfering with the Manager

Ad blockers, privacy extensions, and even some grammar tools have been confirmed to break the Antigravity Manager’s WebSocket frames. uBlock Origin and Privacy Badger have specifically been called out in community threads.

Fix: Whitelist antigravity.google.com in any extension that touches network requests, or run Antigravity in a clean browser profile.

A Server-Side Outage

Sometimes it’s not you. Check the Google Cloud Status Dashboard and the Antigravity status page before going down a debugging rabbit hole.

Antigravity Settings That Reduce Stuck-State Issues

A few configuration tweaks dramatically cut down on how often you hit the waiting-for-input loop.

Enable “Autonomous Mode” for Trusted Tasks

In Settings → Agent Behaviour, you can flip on Autonomous Mode for specific project paths. The agent will skip non-destructive permission requests inside those folders. Use this for sandboxes and personal projects not for anything connected to production.

Set a Default Continue Policy

Antigravity now lets you set a default response for ambiguous prompts. Under Settings → Planner, set “Default on ambiguity” to “Ask once, then proceed with best guess.” This keeps the agent moving instead of pausing every five minutes.

Adjust the Checkpoint Sensitivity

The slider under Settings → Trust controls how often the agent pauses to check in. Default is “Balanced.” Drop it to “Low” if you find yourself approving too many trivial steps. Bump it to “High” only when working on shared or production code.

Comparing Antigravity’s Pause Behaviour vs Other AI Coding Tools

Tool	Pauses for Input	Typical Stuck-State Cause	Recovery
Google Antigravity	Frequent (by design)	Hidden UI elements, WebSocket drops	Refresh tab or use CLI
Cursor (Composer)	Rare	Token limit hit	Auto-resumes
GitHub Copilot Workspace	Moderate	Plan revision required	Manual continue
Claude Code	Rare	Tool permission denied	Re-grant permission
Devin	Frequent	Plan deviation flagged	Owner approval needed

Antigravity’s pause-heavy behaviour is intentional it’s positioned as a safer, more verifiable agentic platform but it does mean stuck states show up more often than in less cautious tools.

You may also read: How Google Antigravity Compares to Cursor and Claude Code in 2026

Real-World Scenarios: When Antigravity Pauses Make Sense (And When They Don’t)

Scenario 1: Building a Feature Branch

You ask Antigravity to add user profile editing to your app. The agent plans the database migration, drafts the API route, and writes the React form. Then it pauses with “Ready to run migration on dev database proceed?”

That pause is doing its job. Migrations are destructive. You confirm, it continues, you ship.

Scenario 2: Refactoring Across Many Files

You ask for a sweeping refactor across 40 files. Halfway through file 18, the agent pauses with no visible message. Your console says “waiting for user input.”

This is the UI glitch version. The planner is technically waiting for a checkpoint approval that never rendered. Quick refresh, send “continue,” and it picks up from file 18 without redoing anything.

Scenario 3: A Failed External API Call

Your agent is integrating with Stripe. The test API key fails. Instead of erroring out, the agent pauses asking how to handle it.

This is technically a useful pause but it’s also avoidable. If you’d told the agent up front “if any API call fails, log and continue,” the pause never happens.

Best Practices to Avoid the Antigravity Stuck Loop

A handful of habits will save you hours.

Write specific, scoped prompts Vague goals produce more pauses
Set fail-handling rules upfront Tell the agent what to do on errors before it starts
Keep tasks under 15 minutes of agent runtime Break larger ones into chunks
Run a clean browser profile for Antigravity sessions to avoid extension conflicts
Keep the CLI installed as a fallback, even if you live in the browser
Save your work to git frequently Antigravity does this by default, but verify
Watch the connection indicator Catch dropped sockets before they become stuck states
Don’t chain unrelated tasks in one session Each fresh agent gets a clean planner

What’s Changed in Antigravity (Recent Updates)

Google has shipped several updates targeting the waiting-for-input issue specifically:

December 2025 patch: Improved WebSocket reconnect logic, fewer silent drops
February 2026 update: Added the visible “Resume Agent” button to the Manager
March 2026 release: New “Default on ambiguity” setting
April 2026 rollout: Pending permissions now show a notification badge instead of being silent

If you’re still on an older Antigravity build, update via Settings → About → Check for Updates. Several of the stuck-state complaints from the early launch period are no longer reproducible on the current build.

That said, the platform’s still in public preview as of May 2026. Always verify the latest patch notes on the official Antigravity changelog Google’s been pushing updates almost weekly.

Frequently Asked Questions

Why does Antigravity keep saying “waiting for user input” when I haven’t done anything?

Because the agent hit a checkpoint and is asking for your approval, but the prompt may be hidden behind a UI element or off-screen. Scroll up in the agent panel, resize your window, or send “continue” in the chat field at the bottom. If nothing happens, refresh the tab your session state is saved server-side.

How do I make Antigravity stop pausing so often?

Go to Settings → Trust and lower the checkpoint sensitivity to “Low.” Also enable Autonomous Mode for trusted project folders, and write more specific prompts. The clearer your initial instructions, the fewer pauses the planner triggers.

Is “waiting for user input” a bug or normal behaviour?

Both, depending on context. It’s normal when the agent legitimately needs approval (destructive commands, ambiguous goals, external API calls). It’s a bug when the input field is invisible or the WebSocket has dropped which is fixable with a refresh or CLI command.

Can I lose my work if I refresh during a stuck state?

No. Antigravity persists agent state, working directory contents, and conversation history server-side. A refresh reconnects you to the same session. The only thing you might lose is unsaved chat input you were typing.

Does this issue happen in the Antigravity desktop app too?

The desktop app is more stable than the browser version, but yes the same kinds of pauses occur. The CLI fallback and “Resume Agent” button work the same way in both.

What’s the fastest fix when Antigravity is stuck?

Press the “Resume Agent” button in the Manager, or refresh the tab. These two cover roughly 80% of stuck-state cases. If neither works, drop to the CLI and run antigravity agent send "continue".

Why is the input field invisible in my Antigravity Manager?

Most often, a window resize or sidebar toggle clipped it. Maximise the window, or collapse and re-expand the agent panel. If that doesn’t work, browser extensions (especially privacy ones) can interfere try a clean browser profile.

Can I configure Antigravity to never pause?

Not entirely, and you wouldn’t want to. Some pauses are tied to safety checks that protect your code and data. You can drastically reduce pauses by enabling Autonomous Mode for trusted folders, lowering checkpoint sensitivity, and writing detailed prompts that anticipate edge cases.

How do I report a true stuck-state bug to Google?

Use the in-app feedback button in the Manager (the small flag icon near the agent’s name). Include the agent ID, your browser version, and a screenshot. Google’s Antigravity team has been responsive on the official Discourse forum as well.

Will this issue go away in future Antigravity versions?

Probably yes, at least for the UI glitch type. Google’s been actively patching the Manager’s rendering issues since launch. The “legitimate pause” version isn’t going anywhere it’s part of the platform’s safety design.

Does Antigravity’s stuck-state issue affect billing or token usage?

A paused agent does not consume tokens. You’re only billed for actual model calls during planning and execution. So even if you’re stuck for an hour, you won’t burn through your quota staring at the screen.

Is there a keyboard shortcut to resume a paused agent?

As of the May 2026 build, Ctrl + Enter (or Cmd + Enter on Mac) inside the agent’s chat field sends a “continue” command and resumes the agent. This is the fastest manual recovery option.

Wrapping Up: The Antigravity Waiting Loop, Solved

The antigravity waiting for user input issue isn’t going to disappear overnight, but it’s also not the showstopper it felt like during the platform’s first weeks. Most of the time, you’re either dealing with a legitimate checkpoint or a tiny UI glitch that takes ten seconds to fix.

Key takeaways:

Three types of stuck states: legitimate pauses, UI glitches, and true bugs diagnose before fixing
Most stuck states resolve with a window resize, a “continue” command, or a tab refresh
Use the Antigravity CLI as your backup channel when the browser Manager misbehaves
Specific prompts and fail-handling rules upfront reduce pauses dramatically
Keep your Antigravity build updated Google’s patched many of the early stuck-state bugs

Next steps: If you’re hitting this issue right now, run through the seven-step fix above in order. If you’re proactively trying to avoid it, jump into Settings and set up Autonomous Mode plus a sensible default-on-ambiguity policy. Both take five minutes and pay off across every future session.

For the latest patches and known issues, bookmark the official Google Antigravity Discourse forum it’s where Google’s PMs post real-time fixes and where the community shares workarounds the moment they’re found.

Other Antigravity Issues : Antigravity Stopped Replying After Recent Update