For the last three weeks, my morning has started at 5:30 with the AdMob dashboard open on one display and Antigravity's Browser Agent on the other, walking through the overnight logs it left for me. The trigger was simple: the wallpaper apps I have been running since 2014 started to lose a few points of revenue as the mediation networks quietly shifted their auction behaviour. When you are an indie developer juggling several apps, the act of flipping between browser tabs to compare numbers eats more time than you would expect. Putting Antigravity's Browser Agent and Claude in Chrome to work in parallel turned out to fit my mornings far better than I had assumed, and the split between the two became surprisingly clean.
This article is a plain account of what each one ended up doing, where I hit walls, and how I plan to keep using them together. I hope it helps if you are an indie developer running multiple apps with limited time and attention.
Why I Bothered Running Both
I will be honest. I started out assuming one of the two would be enough. Antigravity ships with a Browser Agent inside the editor, so you can drive a browser without leaving your code. Claude in Chrome lives directly inside the browser and can read the context of whatever tab you happen to have open. If the two overlap, it should be fine to pick a favourite.
The catch surfaced when I started working through actual AdMob maintenance for apps I have shipped since 2014. The work of staring at eCPM trends in the dashboard sits entirely outside the codebase. Updating mediation SDKs for UnityAds or AppLovin MAX clearly belongs in the editor. When I tried to do both with one agent, my attention kept switching modes, and the day felt fragmented.
With apps totalling 50 million downloads, even a modest eCPM improvement moves the monthly number. That was reason enough to be careful about how attention gets spent.
Week One: Trying Both Agents on the Same Task
For the first week I gave the same instructions to both agents and compared what came back. A typical example was pulling last week's eCPM comparison out of the AdMob dashboard and turning it into a small table.
When I asked Antigravity's Browser Agent, the table landed directly in the daily operations note I keep inside the editor. That suited the part of me that wants every observation to be greppable later. Claude in Chrome, by contrast, kept me inside the dashboard and let me ask follow-up questions in place. I could probe trends over thirty days, compare against last week, and isolate the slots that were sliding.
By midweek a quiet realisation settled in: what I needed was not a single agent that did everything, but two clearly separated rhythms. Recording belongs to the editor. Thinking belongs to the live page. Asking either agent to do both was forcing them into a shape that did not match my own working cycle.
Walls I Bumped Into
The second week brought the practical friction worth writing down.
The first was authentication. The AdMob dashboard requires a Google login, and my main profile carries several operational accounts including Claude Lab and Gemini Lab. Antigravity's Browser Agent runs inside its own browser instance that the editor launches, so once the session expires it asks me to log in again. Claude in Chrome runs inside my regular browser and reuses the existing session. For any page that assumes "you are already signed in", Claude in Chrome was simply less painful.
The second was the round trip back into code. The flip side of the above is that when I ask Claude in Chrome to translate dashboard numbers into a project config file, it cannot reach my local files. I have to copy and paste the output into the editor. Antigravity's Browser Agent can edit the file in place, right up to the commit.
The third was nightly jobs. Since 2014 I have been the kind of indie developer who likes overnight batches to do the quiet, repetitive checks. Antigravity's Background Agent plus Browser Agent can sit there overnight and leave a tidy log. Claude in Chrome, by design, is for the hours when I have the browser open.
The Numbers That Moved
A few quantitative notes for what changed in three weeks.
My morning routine — the ad dashboard, Crashlytics, the App Store review list — used to take forty to fifty minutes. After three weeks it sits comfortably between fifteen and twenty. Roughly half of that gain comes from the agents doing work I no longer touch by hand. The other half is the friction I have removed by not bouncing between "thinking screens" and "recording screens".
Reordering the mediation waterfall has settled into a clean two-step: form hypotheses with Claude in Chrome during the day, and let Antigravity's Browser Agent post the next morning's comparison report. The eCPM lift is a single-digit percentage and I do not have enough days to call it stable. What feels sustainable, though, is the fact that every morning I see the same quality of summary without burning the same amount of attention.
Open Threads
A few pieces I am still working on.
One is keeping the Claude in Chrome conversations reviewable later. Right now I export them on weekends. I have started writing a light bridge that picks out the decisions and drops them into the Antigravity workspace as plain markdown.
The other is the Crashlytics auto-fix loop. That is a separate experiment I am pushing alongside this one. Claude in Chrome reads the dashboards, Antigravity's Background Agent proposes patches. The seam between them is still being designed, but the split looks promising: judgement on the dashboard, mechanical patching in the editor.
My Current Split
After three weeks the routine has stabilised like this. Anything that lives inside an open dashboard tab — AdMob, Crashlytics, App Store Connect, Google Play Console — goes to Claude in Chrome. Anything that needs to land as a record in the editor — config updates, release-note polish, operations notes — goes to Antigravity's Browser Agent. The seam between the two, where dashboard numbers have to flow into a config file, runs through me: Claude in Chrome writes a short summary that I paste into the Antigravity conversation. Two steps instead of one, but the authentication loop never bites.
What I Want to Try Next
The next thing I want to lock down is the morning handover from Antigravity's overnight Background Agent logs into the Claude in Chrome conversation. Today I bridge it by hand. If I can land both ends on a thin, shared markdown report, the role split stays intact while the connection between them tightens.
When I started learning to code over the internet in 1997, the idea that I would one day be an indie developer using two AI agents to keep several apps healthy was nowhere on the horizon. The habit of stacking small, deliberate tools, though, is the same one I picked up back then. Running both of these agents at once might look indulgent, but in the daily reality of keeping a 50 million download portfolio quietly alive on my own, sharing the load between two well-defined tools has been gentler than trying to force one of them to do everything.
If you are also an indie developer running several services or apps at once, I hope some of this carries over.