Thank you for reading Antigravity Lab through May.
Reading back over the month, what ran quietly underneath was the slow shift from "trying out agents" to "delegating work to them and letting them run." Less than the glamour of new features, what dominated the month were the quiet foundations that keep production alive, and the operational records of slowly handing the night watch over to agents.
I've been running indie wallpaper and wellness apps on iOS and Android since 2014, and the combined download count has crossed 50 million. With both my grandfathers having been temple carpenters, my instinct around tool selection isn't "buy the loud thing" but "leave behind a structure that can be patched back together when something breaks." May's articles connected naturally to that instinct, so I want to take a single pass over the month here.
1. Designs That Don't Break in Production — HITL, Fallback, Caching, Knowledge Freshness, Record & Replay
The largest pillar of May was the series of pieces on keeping agents alive in production over the long term.
Designing an HITL Approval Pipeline That Survives Production — Routing Probabilistic Actions Safely With Antigravity Agent takes on how to combine probabilistic agent output with human approval. Neither "approve everything" nor "automate everything" — the perspective of building an approval design suited to a probabilistic partner, in a form that won't collapse in production, matters to everyone preparing to embed agents into real workflows.
Designing a 4-Tier Fallback Architecture for Antigravity Agents — Catching Model Degradation, API Outages, and Cost Overruns Across Layers is written from the premise that "agents will eventually break," and lays out a four-tier hierarchy. The instinct that systems with external dependencies can only be protected through a two-part strategy — lowering the probability of failure, and limiting the damage when failure happens — lines up exactly with what I've felt during long-run indie operations. It's the kind of piece I want kept beside me as a production blueprint.
Prompt Caching and Context Strategy for Antigravity Agents — Cutting 60-80% Off Monthly API Costs in Long-Running Production addresses a reality that's invisible during prototyping but becomes a business-level concern once long-term operation begins. The headline 60–80% figure looks dramatic, but the breakdown reveals quietly disciplined caching design layered carefully on itself — the kind of knowledge that reproduces well.
Designing Knowledge Freshness for Antigravity AI Agents — A Runtime Architecture for Model Cutoffs, Corpus Staleness, and Real-World Time Drift takes on something easy to overlook: time. Treating model cutoffs, the gradual aging of internal documentation, and real-world clock drift as three separate problems with three separate designs is a viewpoint that pays off the longer the system runs.
In the second half of the month, attention turned to the post-incident side as well. Record & Replay for Antigravity Agents — A Production Pattern for Reproducing Failures in Three Minutes and Forensic Design for Antigravity Background Agents — A Cloudflare R2 Audit Log Workflow That Lets You Reconstruct Decisions Six Months Later address audit log design good enough to reconstruct your own judgments six months later. If you're taking post-release incident review seriously, this isn't optional territory. May gained depth precisely because "design that doesn't break" and "design that lets you patch back together when it does" arrived together.
2. Handing Over the Night Watch — The Background Agent Series
Agents earn their highest value during the hours humans are asleep. May lined up several pieces focused on that territory.
Letting Antigravity Background Agent Watch the First 72 Hours After Release records the metrics-monitoring workflow forged across six wallpaper apps released in parallel. The design of watching crash-free users, ANR, eCPM, and Day-1 retention simultaneously comes packaged with warning thresholds calibrated against operations at a combined 50 million downloads — a structure easy to import into your own apps.
Three Weeks of Letting Antigravity's Background Agent Handle Nightly Wallpaper Asset Updates records three weeks of handing the everyday nightly asset update — work that happens every day but doesn't require a human — over to a Background Agent. The article wrestles with how to lower the temple-carpenter instinct of "when in doubt, put down the tool" into the design of a Background Agent.
The end of the month brought Six Weeks of Letting Antigravity Background Agent Review AdMob × Remote Config Diffs, An Operations Record of Auto-Recovering iOS dSYM Across Six Apps With Antigravity Background Agent, and Five Weeks of Letting Antigravity Background Agent Update App Store Screenshots Across Four Apps in Multiple Languages — a sequence that shows the territory of the Background Agent expanding from "observation" into "repair" and "distribution."
3. Six-App Parallel Operations and Sub-Agent Division of Labor
A distinctive movement in May was the sudden thickening of patterns built around sub-agents.
Splitting Daily Crashlytics Triage Across Five Antigravity Sub-Agents is a design for triaging Crashlytics crash reports across multiple cooperating agents. The attempt to automate the triage layer for a volume of crash information beyond what a single developer can read comes out of the lived reality of running six apps in parallel.
Orchestrating Six-App iOS Updates With Antigravity Main / Sub Agents — A Real Account records the Main / Sub agent pattern that solidified during May's parallel update of four iOS apps and two Android apps. Giving the Main agent the choreography and distributing the work across Sub agents is a shape whose advantage only becomes visible at the six-app scale.
Implementation Notes for A/B Testing AdMob Mediation With Antigravity's Parallel Agents and A Design Note on Splitting Pre-Release QA Across Four Parallel Paths With Antigravity Sub-Agents apply parallel agents to A/B testing and pre-release QA — two territories where you want more passes but can't add more people.
Unifying In-App Review Prompt Conditions Across Five Apps With Antigravity Editor — A Few Days of Record and Running 50+ AdMob Mediation Groups With Agents as an Indie Developer — What to Delegate and What to Hold shed light on cross-app "configuration unification" and "configuration optimization" — territories whose value compounds the longer the operation runs.
4. Antigravity 2.0 and Choosing Among the Surrounding Tools
Mid-May, the article Combining Antigravity 2.0 With Chrome DevTools for Agents 1.0 — Migrating Lighthouse Audits, Extensions, and Memory-Leak Tracing into the Loop along with Chrome DevTools for Agents 1.0 Reaches Stable — What Bundling Inside Antigravity 2.0 Changes quickly mapped out the joinery between Antigravity 2.0 and its surrounding tooling.
On the CLI side, Trying the Antigravity CLI (agy) — Migrating From Gemini CLI and Reading the Slash Commands organized the migration path from Gemini CLI and the slash-command surface.
The judgment notes on choosing between tools also gained weight. Six Months of Running Antigravity and Codex CLI Side by Side — A Measured Comparison From Automated Posting Across Six Sites is a measured comparison against the concrete subject of running automated posting across six sites for six months. Paired with the end-of-month Choosing Between Antigravity and Codex in May — Practical Lines From Parallel Operation: An Indie Developer's Judgment Notes, the pair redraws the practical lines for indie developers as of May.
A Month of Switching Between Antigravity's Inline Edit and Agent Mode in Wallpaper App Operations is a tiny choice inside the editor, but a month of switching surfaces concrete lines like "Inline Edit for wallpaper app config-value changes, Agent Mode for feature additions."
On the local LLM side, Choosing Among Ollama, LM Studio, and Gemma 4 for Antigravity's Local LLM Switching — An Indie Developer's Judgment Notes writes out the practical lines for local LLM selection from the angles of bandwidth cost, electricity, and nighttime offline operation.
A Month of Operating Antigravity's Inspector — Reading Agent Behavior Through Observation is an organization of the agent-behavior reading patterns that surfaced from a month of running the Inspector observation tool. A quiet reaffirmation that without observation, there's no improvement.
5. Field Notes from 50-Million-Download Indie Apps
The thread of "honest records from real operations" carried weight in May as well.
Delegating iOS App Maintenance to Antigravity — An Honest 6-Week Report is a head-on record of what happened when I delegated. The willingness to write down failures that look bad from a marketing angle is exactly what makes it the kind of piece you come back to.
Firebase Apple SDK Migration from CocoaPods to SPM: 3 Pitfalls from 4 Real Apps is a practical record of the specific traps encountered while migrating four apps in anticipation of CocoaPods distribution ending in October 2026. For any indie developer using Firebase, it's an unavoidable migration, and the article functions as a field manual for it.
Surviving New iPhone Resolution Support with Antigravity — 29 Changes in DefineManager.h, One Honest Recap and Library Added via Antigravity Crashes Only on Older Android Devices: The coreLibraryDesugaring Blind Spot record the practical work of validating agent suggestions on real devices — from supporting new hardware on one end to keeping older devices alive on the other.
Where Is the Source of Truth for Billing State? Designing the ad-free Pattern with Antigravity frames the well-known fact that if billing state can be read from multiple paths, contradictions will eventually appear somewhere — and treats it as a Source-of-Truth pattern. With concrete implementation details for the composite judgment between BillingManager and AdFreeManager, it shines light on one of the most accident-prone areas in indie development.
Two Weeks of Letting Antigravity Translate Localizable.xcstrings Across 8 Languages — Operations Memo, Running iOS Push Notification A/B Tests Weekly With an Antigravity Agent — A Self-Improving Loop on Copy, Timing, and Segments, and the late-May Four Weeks of Phased AppLovin MAX Integration on iOS Through Antigravity Editor record the three pillars of app operations — monetization, notifications, multilingual — written down as agent division of labor.
6. Quiet Troubleshooting
May lined up plenty of practical pieces too.
Why Antigravity Agents Hit Permission denied on git push, and How to Fix It for Good organizes a problem every developer juggling multiple GitHub accounts hits at least once — from diagnosis through permanent fix.
When Antigravity Agent Edits Break Diffs Due to Mixed CRLF/LF Line Endings and When Antigravity Ignores Your AGENTS.md — How to Diagnose and Fix It are mid-month entries about "hard-to-notice but real" trouble.
Toward the back of the month came When Antigravity's Workspace Indexing Won't Finish on Large Repositories, Antigravity Agents Hang on ssh / sudo / git rebase -i: Diagnosing the Missing-TTY Problem, Fixing Japanese Mojibake in Antigravity's Built-In Terminal, When Antigravity's AI Agent Misreads .env Values — Causes and Fixes, and When Antigravity's Google Sign-In Won't Return From the Auth Screen — May was a month where the "drawer of fixes" quietly grew.
When Antigravity Agents Still Reference the Previous Project After Switching — Investigating and Fixing Context Bleed and Fixing and Preventing Antigravity's AI From Overwriting package.json Dependency Versions translate naturally for indie developers running multiple projects in parallel.
7. Parallel Operation and Comparison Reviews
Articles reinforcing the choose-between judgment also lined up.
Three Weeks of Running Antigravity Browser Agent and Claude in Chrome in Parallel — A Clearer Division of Roles records three weeks of putting the Browser Agent and Claude in Chrome next to each other and watching a clear division of roles emerge.
Growing Antigravity's Walkthroughs Into a "Regression Replay Memo" — A Four-Month Operations Record and Two Weeks of Antigravity Walkthroughs During a Wallpaper App Refactor record from both the long and short ends the idea of treating Walkthroughs not as one-off procedural docs but as living operational documents that keep being grown.
Three Weeks of Operating Antigravity's Settings Sync — A Synchronization Design for Doing Indie Development Across Two Macs is a record of the synchronization design for sustaining indie development across two Macs. Switching between a main Mac and a sub Mac, the Settings Sync design choices visibly shape the next day's productivity.
The Weight of "35-Hour Autonomy" Shown by Qwen3.7-Max — Design Notes for an Indie Developer Combining Antigravity and Claude Code writes down how the new axis of long-horizon autonomous execution lands inside an indie developer's design judgment.
8. VS Code 1.121 and the Agent Host Protocol
Around the editor, VS Code 1.121 and the Agent Host Protocol — Organizing What's Happening Around the Editor as Agents Begin to Live Remotely organizes the new current of agents that "live" remotely, viewed through the surrounding editor motion.
Carrying Into June
Looking back over May, the themes I want to carry into June are:
- The implementation side of operational design for agent observability — combine Inspector, Record & Replay, and the R2 audit log into concrete templates that other indie developers can import
- Extending the Background Agent's territory further into "repair" — building on dSYM auto-recovery and App Store screenshot multilingual updates, hand more small post-release repair tasks to the night shift
- More "agreements" for six-app parallel operations — extending the In-App Review prompt unification, encode billing-state, notification, and ad-placement agreements together with agents
- Re-organize the joinery between Antigravity 2.0 and surrounding tools — polish the mid-month mapping through June's real operations
May was a month that took one step from "trying" agents to "delegating" to them. With temple carpenters in my family, I carry a strong instinct that "working" and "being trustworthy" are two different stories. To delegate, you have to design the patch-back-together path alongside the operating path — that's what May quietly confirmed.
In June, Antigravity Lab will continue writing pieces grounded in the texture of real production operations. Thank you for reading.