Maintaining two unofficial #Java SDKs for fast-moving AI APIs used to mean hours of manual spec diffing.
Now, Google @antigravity automates my developer workflow every Monday at 9:00 AM via scheduled automations, with 2 Human-in-the-Loop gates! 🤖☕
Here is the setup 🧵👇
Are your AI subagents drowning in too many tools? 🛠️
By default, subagents inherit every active MCP server. This leads to context window bloat and degraded agent reasoning.
I wrote a guide on how to isolate your subagents' context in @antigravity using:
🤝 Shared Plugin Servers
🛡️ Filtered Global Servers
🔒 Private Inline Servers
⚙️ Tool-Level Filtering
Keep your subagents lean, focused, and equipped with only the tools they need.
Read the full guide here 👇
linkedin.com/pulse/agent-plu…#AgenticEngineering#AI#AgentPlugins
Yes! For example, I developed this CLI & library to handle my 64x64 LED screen, and #Antigravity took care of the reverse engineering of the protocol & web API:
github.com/glaforge/jixoo
It's very powerful!
Yeah, in a way!
I wouldn't develop & maintain those libraries otherwise. So it's actually more like extra work & contributions, than to really automate more of my existing work.
This morning, both automations spotted upstream updates, implemented the changes after my plan review, and helped me ship v2.1.1 & v0.3.3 before 10:00 AM!
Full walkthrough & configs:
👉 glaforge.dev/posts/2026/11/0…
Gate 2️⃣: Review the accomplished work before release.
Once approved, the agents update the code, deprecate removed schemas safely, write parity unit tests, and run ./mvnw test.
Human-in-the-Loop (with 2 human gates) 🛡️
I don't let unattended cron agents push code or publish artifacts on their own:
Gate 1️⃣: Review the implementation plan. The 9 AM run stops after diffing & planning. Over coffee, I review the proposed changes before clicking Proceed.
Using /automation (or the Automations panel), you define a cron schedule (sidecar.json) that spins up a fresh agent session on a recurring cadence.
Every Monday at 9 AM, 2 background agents wake up in my SDK workspaces & follow the AGENTS.md runbooks.
Le problème avec les cannelés que j'achète quand j'ai la chance de parler à @bdxio, c'est qu'ils ne restent pas longtemps dans la boîte, même quand j'en prends une grande !
Merci @LaToqueCuivree !
Built your own token optimizer? Fair. Measure it on total task cost, against a clean run. Some tricks break the prompt cache and raise the bill. Tokenade reduces token costs by 39% on average on long sessions (THOL, Sept 14). github.com/pi-infected/token…
[🇫🇷 version FR👇] A look back at atmospheric reentry, a rollercoaster both physically and emotionally!
🎥 @esa
Retour sur les sensations fortes de la rentrée atmosphérique, tant dans le corps que dans la tête !
#εpsilon • @esaspaceflight • @ESA_fr • @Space_Station
Cool 😎 Thanks for the recommendation 🙏
I've implemented some extra checks as per your scorer's recommendations. I went from 66 to 88!
But I can't get to 100 as I don't have control on the headers, as it's hosted on GitHub Pages.