Agents & Manager Articles
Multi-agent development with Antigravity — Manager Surface leadership, orchestration patterns, loop prevention, and monitoring unattended runs.
When Your Agent's Files Vanish Into .gitignore — A Pre-Commit Detection Gate
When an agent writes files that match .gitignore, the diff review looks perfect but nothing lands in the commit. Here is a gate script that catches ignored build output before you push, plus the tuning it needs.
Measuring the Rework Rate of What You Delegate to Agents: Drawing Delegation Boundaries with Numbers, Not Instinct
How much should you hand to an agent? I drew that line by instinct for a long time. Here is a practical way to compute a per-category rework rate from your git history and redraw the delegation boundary with numbers, with working code.
When Parallel Agents Ran the Same Task Twice and Quietly Doubled the Bill — Field Notes on Measuring and Stopping Duplicates
The bill for our parallel agents came in about 1.9x higher than expected — because multiple workers were running the same task twice. These are field notes on measuring the duplication, stopping it with idempotency keys, and attributing cost per task.
Never Miss a Managed Agent Completion: Pairing a Serverless Receiver with Polling Reconciliation
A cloud Managed Agent can finish while you are not watching, and the webhook that should tell you can quietly fail. Here is a serverless receiver on Cloudflare Workers paired with polling reconciliation, and a state machine that recovers every completion within minutes.
Protecting Your Agent Stack's Known-Good State with a Single Lockfile — Change-Budget Design for an Era of Simultaneously Moving Parts
When the IDE build, CLI, model, and dependencies all move at once, you can no longer tell which one caused a regression. Here is a change-budget design that pins your known-good state to one lockfile, with working code and operational logs.
Make the Self-Debugging Agent Walk the Logged-In and Post-Paywall Screens
By default, Antigravity 2.0's real-browser self-debug only sees the logged-out free view and reports success. To catch billing regressions, inject an authenticated session and paid state into the agent's browser and force coverage with assertions.
When Nobody Reads Your AI Code Reviewer Anymore — Field Notes on Measuring Actioned-Rate
Our production AI code-review agent quietly went hollow over six months. When the team started silently resolving every comment, we instrumented actioned-rate and false-positive rate to bring it back. These are the field notes.
Where to Put Evidence and Approval When Your Agent Self-Debugs in a Real Browser
Antigravity 2.0 launches a real Chrome mid-build, clicking buttons and taking screenshots to self-heal. It is fast, but shipping that as-is is risky. Here is how to capture evidence and draw the approval boundary.
Turning Last Night's Failed Runs into Tomorrow's Prevention — Designing a Postmortem Feedback Loop
Stop letting unattended failures end at a notification. A concrete design for classifying failures and feeding fixes back into Guide skills, gates, and schedules, with measured recurrence rates.
When the Tech-Debt Score Drops but the Same Files Keep Breaking — Field Notes on Instrumenting Fan-in and Churn
Letting Antigravity's architecture agent score technical debt is not enough — bugs often recur in the same files after refactoring. Here is how we instrumented the fan-in times churn that static complexity misses, and reconciled the score against real incidents.
It Worked Interactively but Went Silent Overnight — Making an Antigravity Agent Behave the Same in the Desktop and the CLI
An agent that runs perfectly in the Antigravity desktop app but does nothing when you schedule it through the CLI. This walks through absorbing the gap between interactive and unattended runs across four points — approvals, context, secrets, and runtime — with working code and a preflight check, so one definition behaves identically on both.
Detecting and Fixing Drift Between a Guide Skill and Your Code
Pin a procedure into a built-in Guide skill and it gets left behind when the code later changes. Here is an operational design that machine-checks the things a Guide references, catches drift early, and keeps the Guide thin.
Keeping Naming and Formatting Stable When Your Model Falls Back
When a model falls back mid-run, an agent's naming conventions and formatting drift quietly. Here is how I enforce a model-independent style contract plus a drift probe to keep output consistent.
When the Android CLI Got 3x Faster and Cut Tokens by ~70%, the Right Move Was More Verification Per Change — Not More Parallelism
Reading that the Android CLI agent runs ~3x faster while using ~70% fewer tokens, my first instinct was to ask how many runs to parallelize. But a faster agent doesn't change how much work ships — it changes where the queue forms. This walks through why, sizes the new bottleneck (review and verification gates) with Little's Law, enforces a WIP cap with a working Python admission controller, and reinvests the freed budget into depth per change — with measured results.
When the Agent Hands You 1,400 Replacements in One Commit, Ask for Batches Instead
Ask Antigravity to run a large codemod and you can get back one unreviewable commit. Here is a small design — ast-grep rules plus a verified batch driver — that splits a mechanical replacement into the machine's job and the human's check, with working code.
When Parallel Sub-Agents Fight Over One API's Rate Limit: A Shared Token Bucket That Caps the Aggregate
Run Antigravity 2.0 dynamic sub-agents in parallel and each one hits the same external API independently, pushing the aggregate rate over the limit and triggering cascades of 429s. Here is a shared token bucket that caps the aggregate proactively, with working code through a Redis version.
When Your Antigravity Agent Opens a PR That Just Says "Update files" — and a Gate That Forces a Reviewable Summary
Pull requests opened automatically by an Antigravity agent tend to carry empty descriptions like "Update files." Here is a validation gate, with working code, that estimates risk from the diff and rejects vacuous descriptions so a human can actually review them.
Spotting Agents That Are Alive but Stuck — Designing a Progress Heartbeat and Watchdog
The process is alive but the work isn't moving — the nastiest state for a background agent. Here is how to switch from liveness to progress monitoring to detect it, and how to stop it safely, with working code.
When Background Agents Run Twice — Stopping Double Execution with Leases and Fencing Tokens
The same scheduled job fires from two machines at once and they overwrite each other's output. Here is how to stop that failure mode at the root in Antigravity 2.0 background agents, using leases and fencing tokens, with working code.
When Your Antigravity-Written Tests Only Look Green: Measuring Effectiveness with Mutation Testing
Tests written by an Antigravity agent can pass and still fail to catch the bug that matters. Here is how to measure their real effectiveness with mutation testing and only adopt tests after they kill the surviving mutants — with working code.
A Review Gate Design for Safely Folding Parallel Agents' Diffs into One Branch
Antigravity 2.0 made running multiple agents in parallel practical, but verifying each agent's output and integrating it into one branch is left to you. Here is how to build a diff-level review gate in stages, with judgment criteria and scripts.
It Did Things I Never Asked For — Binding an Agent's Task Scope With a Contract
Ask it to fix a button color and you get a refactor, renames, and a dependency bump too. This is a scope problem, not a permission one. Here is a contract that stops at the scope boundary and asks.
Treating Built-in Guide Skills as Design Assets, Not Throwaway Prompts
Antigravity v2.2.1 added built-in Guide skills. Here is a concrete structure and set of judgment calls for running them as version-controlled, shared design assets instead of one-off instructions.
Prune an Antigravity plan before you approve it
Instead of approving a Planning-mode plan wholesale, cut the one risky step and keep the rest. A field-tested look at partial plan editing from a solo developer's desk.
The Day the Article I Asked It to Format Became the Agent's Instructions
When you run an unattended content-formatting pipeline with Antigravity CLI, instruction-like text buried in the file you are processing can hijack the agent. Here is how I separate the instruction channel from the data channel and add an output-scope acceptance gate to reject anything out of bounds.
Before Your dynamic sub-agents Branch Out Too Far — Designing a Depth Budget and Fan-out Cap
Antigravity 2.0's dynamic sub-agents can spawn their own sub-agents at runtime. Handy, but without depth and fan-out control they can burn through your quota overnight. Here are three guards, with concrete code.
Keep a Tamper-Evident Audit Log of Your Autonomous Agent's Actions
To record the decisions and actions an Antigravity agent takes autonomously in a form you can trace and verify later, design an append-only audit log whose hash chain detects tampering. Includes the implementation.
Pin Your Agent's Output With Golden Snapshots Before Switching Models
When Antigravity's engine moves to Gemini 3.5 Flash, an agent's output can drift silently. This walks through a golden-snapshot regression gate that catches the drift, with the actual test code and a migration-day checklist.
When Your Agent Automation Breaks: How Many Minutes to Recovery?
As Antigravity 2.0 adds desktop, CLI, and SDK surfaces, the things you must restore after a failure multiply too. As an indie developer running several sites on autopilot, I lay out a three-layer recovery design covering credentials, definitions, and state, plus a monthly restore drill.
Turning a throwaway prompt into a reusable sub-agent
When a one-off prompt to an Antigravity 2.0 dynamic sub-agent works beautifully, it usually vanishes into your chat history. Here is how to capture it as a reusable definition, with the actual file layout and the distillation steps.
Before a Major Update Silently Breaks Your Overnight Automation — Designing a Staged-Adoption Canary Gate
After a major update dropped my unattended run success rate from about 98% to 63% overnight, I built a staged-adoption gate that freezes the working setup, verifies a new version against a golden output in an isolated profile, and only then adopts it. Here is the design with bash and Python.
Before a Stray Instruction in a Fetched Page Drives Your Unattended Agent — Tainting Inputs to Downgrade Capabilities
So an unattended agent that reads external pages or PDFs can't be hijacked by an instruction hidden inside them: track the taint of every input and automatically downgrade side-effecting tools. With working Python and real operational numbers.
Your Antigravity Custom Tools Don't Break by Design — They Break on Re-execution: Field Notes on Idempotency and Error Contracts
Once you add a custom tool to an Antigravity agent, the real production problem is re-execution and duplicated side effects. Here are the idempotency keys, error contracts, health gates, and tool-sprawl checks that actually held up in practice.
Stop Hard-Coding Your Agent Concurrency: Let It Tune Itself From What It Observes
When you run several Antigravity 2.0 agents in parallel, a single fixed concurrency number is wrong twice: it stalls at 429s during the day and idles capacity at night. Here is an adaptive design borrowed from TCP congestion control — additive increase, multiplicative decrease — that moves your concurrency from observed signals, with working Python and field notes.
Taking Stock of the Dependencies Your Agent Added — A Design for Keeping License and Provenance Traceable
A few months of letting agents work, and your package.json quietly grows dependencies you don't remember adding. Here is a design for taking stock — recovering what was added, when, and why, in a form you can still trace later.
Letting a Background Agent Work Overnight Without Regretting It by Morning — Guardrails for Unattended Runs
When you hand overnight refactoring to Antigravity's Background Agent, the morning brings as much anxiety as convenience. From three angles — blast radius, completion criteria, and detecting silent regressions — here are the guardrails that let me run unattended jobs with confidence.
Keeping Unattended Agent Run Logs Long Enough to Debug — Without Filling the Disk
A scheduled agent is only fixable if you can reconstruct why it failed. Here is how to keep run logs around without filling the disk — tiered retention, schema-versioned records, and a compaction job — drawn from running four sites on autopilot as an indie developer.
When a Timed-Out Unattended Agent Leaves a Half-Written File Behind
When a scheduled agent gets killed on timeout, it can leave a half-written file that silently poisons the next stage. Here is the atomic write, stale-temp cleanup, and post-write content assertion I use to keep unattended pipelines from breaking.
Don't Lose Failed Agent Jobs: Designing a Dead-Letter and Requeue Path
Scheduled agents fail silently overnight and the work simply vanishes. Here is how to catch those failures with a dead-letter store and a staged requeue, drawn from running four sites on autopilot as an indie developer.
More Agents Won't Speed Up Every Part of Your Pipeline — Designing the Parallel/Serial Line
Antigravity 2.0's parallel multi-agent execution is powerful, but adding agents doesn't make everything faster. Here's how I decide which work to parallelize and which to keep serial, derived from invariants and a dependency graph, with examples from running several sites as a solo developer.