https://iamulya.one/A blog written by Amulya Bhatia, connecting technical innovations to business wins. 2026-07-03T22:47:37+02:00 Amulya Bhatia https://iamulya.one/ Jekyll © 2026 Amulya Bhatia /assets/img/favicons/favicon.ico /assets/img/favicons/favicon-96x96.png 2026-07-02T13:00:00+02:00 2026-07-02T13:00:00+02:00 https://iamulya.one/posts/understanding-the-policy-priority-model/ Amulya Bhatia

This article is part of the Antigravity Engineering Series. You have 15 policies. The agent calls run_command with npm install lodash. Which policy fires? Why that one and not the three others that also match run_command? If you’ve ever debugged a firewall rule set or traced a message through a content-based router, you already know the shape of this problem. Multiple rules match the same...

2026-07-02T13:00:00+02:00 2026-07-02T13:00:00+02:00 https://iamulya.one/posts/testing-strategies-for-automating-antigravity-workflows/ Amulya Bhatia

This article is part of the Antigravity Engineering Series. You built the hook scripts. You wrote the policies. You configured the sidecar. You shipped it all and went to bed. At 3 AM, a one-character typo in your PreToolUse hook emitted malformed JSON, silently failing open. The agent — now ungated — ran npm publish against your production registry. You woke up to 47 Slack messages. The...

2026-07-02T13:00:00+02:00 2026-07-02T13:00:00+02:00 https://iamulya.one/posts/systems-instruction-architecture-in-antigravity/ Amulya Bhatia

This article is part of the Antigravity Engineering Series. Your agent writes code like a chatbot — overly polite, hedging every suggestion with “you might want to consider…”, emitting markdown explanations where you need raw code. You want it to act like a senior engineer: direct, opinionated, testing everything it ships. This is a configuration problem, not a training problem. The syste...

2026-07-02T13:00:00+02:00 2026-07-02T13:00:00+02:00 https://iamulya.one/posts/orchestrating-multi-agent-teams-with-antigravity/ Amulya Bhatia

This article is part of the Antigravity Engineering Series. Picture a refactoring task that touches 47 files across 3 modules. A single agent starts at file 1. By file 30, its context window is saturated with the memory of files 1–29 — code it no longer needs but can’t discard. Coherence degrades. The agent begins making contradictory changes. By file 40, it’s undoing work it did at file 1...

2026-07-02T13:00:00+02:00 2026-07-02T13:00:00+02:00 https://iamulya.one/posts/hook-contexts-in-antigravity-sdk/ Amulya Bhatia

This article is part of the Antigravity Engineering Series. You’re building a token budget hook. It needs to track total tokens across the entire session. But it also needs to track tokens per turn to catch retry loops. And it needs to time individual tool calls to detect slow operations. Three different lifetimes. Three different scopes. If you use instance variables, you end up resettin...