Thanks for reading Antigravity Lab this week.
Looking back at the past seven days, one shift stands out clearly: last week, the conversation was about how to make multi-agent systems work; this week it moved entirely to how to sell what works, how to sustain it, and how to keep it from breaking. AgentKit 2.0 has reached a kind of plateau, and the ecosystem around Antigravity has stepped from the "implementation phase" into the "business phase." That's the honest read from where I sit.
Working solo, this transition is quietly heavy. The mindset for "making it run" and the mindset for "making it pay and stay up" require very different wiring. The articles this week are deliberately built to help that switchover.
Center of Gravity This Week: From "Make It Run" to "Sell It and Sustain It"
The most telling signal was the AgentKit 2.0 articles shifting their weight from implementation tutorials toward production-grade design.
"Designing a Context Compression Sub-Agent in AgentKit 2.0" addresses the history bloat problem you always hit on long-running tasks, by separating compression into its own sub-agent. Splitting out a "folder" agent feels obvious in retrospect, but the first move is the hardest — this article focuses on where exactly to draw that boundary.
"Building Self-Healing Antigravity Agents — Detection, Diagnosis, and Recovery in Production" and "Designing Production Incident Runbooks for Antigravity Agents" are best read as a pair. The first covers autonomous recovery on the agent side; the second covers the human-side operational framework. The shared theme is the line between what AI can pick up and what it cannot — and both articles work that boundary from different sides.
"Building Validation Loops in Antigravity Agents: 3 Patterns for AI-Verified Output" lays out three patterns — self-verification, cross-model verification, and rule-based verification — for using AI to check AI output. The accuracy conversation doesn't end with "use a better model"; the validation loop design pushes back the failure rate substantially — that's the field-tested take we wanted to share.
And "Antigravity Agent Shadow Mode Production Rollout Guide" covers a staged release strategy where new versions run in parallel against production traffic for observation, without affecting users. The pattern scales down to solo-developer territory more naturally than people assume.
Revenue and Business Design Filled Out Significantly
Alongside the engineering content, we deliberately thickened the monetization and business-design coverage this week. The bridge from "I can make it run" to "I can sustain a business with it" is what readers are asking for most.
"Selling Antigravity-Powered Business Automation as Contract Work" lays out a realistic path for indie developers who don't want to start with a full SaaS — entering through contract engagements instead. It maps three target areas (sales, accounting, marketing) against three acquisition channels (existing network, DX communities, SaaS subcontracting), giving a 3×3 grid to navigate. Just dropping the assumption that "SaaS is the only correct answer" opens up a much wider field of options.
The companion piece, "From Acquisition to Delivery: A Complete Guide to Antigravity Contract Work", walks through eight steps from intake to delivery, including contract clause templates. The core insight is that the monthly operations phase is where the real margin lives — something you only learn after running the work yourself.
We also published "A 90-Day Roadmap for Solo Devs to Generate Income with AgentKit 2.0", paired with a channel comparison piece, "Five Distribution Channels for Selling Agents Built on AgentKit 2.0". Whether to ride a platform or sell directly is a decision we wanted to support with data, not gut feel.
"Building a Stripe Dunning Recovery Pipeline with Antigravity AI Agents" is a concrete implementation guide for automating recovery from failed payments using AI agents. The quiet truth indie SaaS operators learn the hard way is that the dunning design is the last line of defense protecting your MRR — that's exactly what this article is built around.
The Cloudflare Stack Makes "Never Leave the IDE" Real
Cloudflare-related coverage was concentrated this week as well. The trio — Workers, Vectorize, Durable Objects — has come fully into view as the natural infrastructure layer for AI agents.
"Antigravity × Cloudflare Vectorize: Build a Production RAG Pipeline That Runs at the Edge" is a complete walkthrough of a production RAG configuration. The shift from "RAG running somewhere in the cloud" to "RAG running close to the user" is starting to matter on both latency and operational cost.
"Stateful AI Agents on Antigravity × Cloudflare Durable Objects" covers a one-session-equals-one-instance design for centralizing conversation, tool state, and cost. Once you lean into Durable Objects' characteristics, treating each agent's "personality" as instance-bound starts to feel natural.
And "Build a Complete SaaS in Antigravity — Stripe Billing, Auth, and Deployment Without Leaving Your IDE" ties it all together with a single-pane workflow: Supabase auth, Stripe checkout, Cloudflare Workers deploy. A development experience that goes "from prompt to production without leaving the editor" is no longer aspirational; it's an actual option you can pick today.
Related pieces — "Antigravity × Lighthouse CI: A Complete Guide to Catching Web Performance Regressions Automatically" and "Composing Event-Driven AI Workflows with Antigravity and Inngest" — round out the operations and quality-assurance side, both drawing steady traffic.
A Strong Week of Troubleshooting Articles
Less flashy but quietly important: troubleshooting articles got reinforced this week.
"Diagnosing and Recovering Lost Chat History in Antigravity", "Diagnosing and Fixing Breakpoints Not Hitting in Antigravity", "Antigravity Keeps Asking Permission for the Same Command? Fix the Approval Dialog Loop", "Antigravity Model Picker Showing No Models — How to Diagnose and Recover", and "Fixing Antigravity's Slow Performance and Broken File Watching on WSL2" — every one of these covers a symptom most readers have hit at least once, broken down into root cause analysis plus a recovery procedure.
Troubleshooting can look like a low-traffic category, but it's actually one of the highest-trust article types: the moment a reader is stuck and your article gets them unstuck, they remember the site and come back for the next thing. That's the quiet, important design we kept in mind while writing them.
A Small Adjustment to the Author Voice
This week we also touched the writing itself, not just the content. The author intro has been standardized to "Masaki Hirokawa, indie developer and artist" rather than just a casual greeting. A bare "hello, it's Masaki" wasn't conveying what the site is built around for first-time visitors.
It's a small change, but a deliberate one — Antigravity Lab is a place run by a solo developer and artist, and we want that grounding to be visible in the writing itself.
Coming Up Next Week
Next week's lineup:
- Deeper dives on agent observability — OpenTelemetry and metrics design specific to agents
- AgentKit 2.0 integrating with established services — Salesforce, HubSpot, Slack and other practical scenarios
- Continuing the indie monetization series — Incorporation decisions, tax design, and real contract examples
May spans some long holiday weekends, so we're leaning into deeper articles you can sit with over a longer block of reading time.
Thanks again for reading. Because of everyone who keeps showing up, this site is becoming what it's meant to be: a place a solo developer can return to and not feel they've outgrown. I'm grateful.