A lot has been written about how agentic development platforms like Antigravity can boost coding productivity. A few months ago I wrote an article about how I’ve been harnessing Antigravity as a PM. Since then, I’ve continued to using it every day in my job, and in this post I’ll share two more examples of how it helps me tackle complex tasks.
Solving the mystery of expanding PR review time
For the last two years I have been working with the team that creates the documentation for Google Cloud products to help them harness a variety of AI tools (including Antigravity, as well as custom-made AI tools for technical writing). We break pull requests (PRs) for doc updates into two parts:
- Writing time: From when a PR is created until it goes out for review.
- Review time: From when a PR goes out for review until the change is merged.
So far in 2026, we’ve seen writing time drop significantly as the tech writing team adopts AI more broadly — particularly Antigravity — to automate significant chunks of the technical writing workflow. However, we haven’t seen the same drop in review time.
To get to the bottom of why the review time is stubbornly staying high, we needed to unpack a ton of metadata about thousands of PRs, including the number of reviewers, the category of reviewers (engineers, tech writers, PMs, etc.), the diff sizes, which AI tools were used, and a dozen other details about these PRs. If I had a week or two, I could collect this metadata using the APIs for a variety of internal Google tools. I didn’t have a week — I needed to narrow down some plausible root causes for sticky review times in an hour or two.
Antigravity came to the rescue by helping me to analyze thousands of PRs and then to isolate some changes in 2026 that correlate with longer PR review time:
- Bigger PRs
- PR reviews involving engineers
- PRs related to launches
Read on to learn more about how Antigravity got me from zero to potential solutions to the mystery of why PR review time expanded in 2026.
First, I got Antigravity to create a spreadsheet for me that listed all the PRs I needed to analyze (about 15,000 PRs in all) with the basic information for each PR:
- Month and year the PR was created
- Owner of the PR
- Number of lines added, modified, or deleted
Next, I used Antigravity to examine the descriptions of all these PRs to get a catalog of all the AI tools used in them and to update the spreadsheet to indicate which AI tool was used for each PR.
Now that I had a human-readable catalog of metadata about an interesting set of documentation PRs, I pointed Antigravity at it to analyze the metadata and identify key differences between the PRs prior to widespread adoption of AI and after the tech writing team embraced AI. Antigravity surfaced some notable patterns, including an increase in the rate of extra-large PRs and a decrease in the number of extra-small PRs.
At this point I tagged experts in the tech writing team to review what Antigravity had come up with so far. They had some great recommendations for additional metadata that could help to peel the onion and get to some of the root causes of longer doc reviews, including the following PR details:
- The product area
- The trigger for the PR (such as a launch, a bug fix, or some other reason)
- The number of comments coming from automated agents (vs. human comments)
- The time between the final comment and the submission of the PR
In the past it would have taken days to assemble this additional metadata. With Antigravity, it took a couple of minutes with a few prompts like this:
Update <spreadsheet> filling in the “Time between last comment response and
submission” column in both tabs with the elapsed time (in hours rounded to
one decimal) between the final comment resolution in the PR and when the PR
was submitted for each PR.
Here is an example of what a portion of the spreadsheet looked like after a few iterations with Antigravity:
As an aside, the data you see in the above sample has been redacted with the help of Antigravity.
Once I had incorporated the additional metadata in the spreadsheet using prompts like these, I turned back to Antigravity to do another round of analysis on the updated metadata:
Now that additional columns have been added to this table
(including but not limited to PR trigger, product area, number of
automated comments, Time between last comment response and submission)
go through both tabs of the spreadsheet and share
(a) your observations about the root causes for review time increasing
in 2026,
(b) your recommendations on what we should do to reverse this trend of
increased review time.
With this prompt, Antigravity uncovered a set of unexpected differences between 2025 PRs and PRs from the same period in 2026, including:
- More big PRs and fewer small PRs in 2026
- Higher correlation in 2026 between engineers being involved in PR reviews and long-running reviews
- Reviews of PRs related to full-blown launches (vs. bugs or one-off doc updates) were, on average, much longer in 2026
By combining these insights from the Antigravity-driven analysis with anecdotal information about doc reviews, we were able to come up with a set of concrete actions (including a review-oriented agent skill and a set of tweaks to our doc review process) that stand an excellent chance of reversing the tide of long-running doc reviews.
In the past, this exercise of collecting and analyzing metadata on thousands of documentation PRs to find potential root causes for longer review time would have taken weeks, if not months. Thanks to Antigravity, I was able to complete the task in a few hours and, as the team recommended new angles to explore, layer in additional metadata. With Antigravity I was able to explore new hypotheses quickly. I could iteratively identify new data points (like the number of agent-created comments), and incorporate the new metadata in fresh analysis cycles.
Stitching together the context from multiple meetings
As a PM, I spend a lot of my time in meetings. Thanks to Google Meet’s transcription and AI note-taking capabilities, participants get both a detailed transcript and a summary of key points. Instead of splitting my attention between listening and taking notes, I can stay fully focused on the conversation.
Recently, I met separately with three senior leaders to gather feedback on an internal tool. Once the third session wrapped, I needed to synthesize all of their feedback. Even with Google Meet’s transcripts and summaries, manually cross-referencing three separate conversations would have been a daunting task.
I wanted a single spreadsheet that captured every major point raised across the sessions — categorized by type (requirement or question), which leaders raised it, its current status (done, planned, or TBD), a concise title, and a short summary paragraph.
Here is the prompt I sent to Antigravity to consolidate the feedback from the senior leaders:
I need to get a consolidated list of unfilled requirements / questions from
<senior leader 1>, <senior leader 2> and <senior leader 3> from their reviews
of the project: <project design doc>. Update this spreadsheet <spreadsheet
with project item summaries> to consolidate the feedback from these sources:
<senior leader 1> (<senior leader 1 mtg summary>,
<senior leader 1 mtg transcript>),
<senior leader 2> (<senior leader 2 mtg summary>,
<senior leader 2 mtg transcript>),
<senior leader 3> (<senior leader 3 mtg summary>,
<senior leader 3 mtg transcript>).
Antigravity came back with a complete summary that accurately captured 90 minutes of meetings and neatly tied together the strands from all three of the senior leaders. Here is an example of the output (again redacted courtesy of Antigravity):
As you can see from this example, Antigravity identified recurring themes raised by more than one senior leader across the meetings. We can use this summary, which took a few minutes for Antigravity to generate while I was fighting another fire, to triage the requirements and land on a prioritized set of marching orders for the project. Pre-AI, it would have taken a solid morning of concentrated work to get to this point.
More takeaways from using Antigravity to solve PM problems
Ever since I published my first post on using Antigravity as a PM, I’ve wanted to write a follow-up — though at first I worried I wouldn’t have enough varied examples. After another six weeks of chaotic PM life with Antigravity as my daily sidekick, I had the opposite problem: choosing just two examples from more than a dozen ways it helped me save time or tackle previously intractable tasks. From diagnosing why PR review times were climbing to synthesizing executive feedback across multiple meetings, Antigravity keeps surprising me with what it can do. Give it another month or two, and I’m certain I’ll have even more examples of Antigravity being a PM’s best friend.
More on Antigravity as a PM’s best friend was originally published in Google Cloud – Community on Medium, where people are continuing the conversation by highlighting and responding to this story.
Source Credit: https://medium.com/google-cloud/more-on-antigravity-as-a-pms-best-friend-3fbd9fc62831?source=rss—-e52cf94d98af—4
About the Author
