In September 2026 I posted a series of LinkedIn updates showcasing a wave of open‑source AI‑agent projects. I highlighted the massive community traction of Awesome LLM Apps, introduced the lightweight Julia 1 decision‑routing model, demonstrated compliance tooling like ZeroDrift, and even revealed a refund‑attack demo that harvested $160 from a $149 order.
Awesome LLM Apps has been forked over 20,600 times on GitHub.
The repository contains 1,280 commits.
Julia 1 is a 144 M‑parameter model that runs on a plain CPU.
Training Julia 1 on cloud GPUs cost $104.
ZeroDrift’s first compliance verdict took 3,963 ms.
The refund‑attack demo extracted $160 from a $149 order.
Google's Gemini Agent is built for high‑ceiling work where EACH task gets its own computer.
Each Coworker agent gets its own Workspace account: an agent email, calendar, Drive, and presence in your company directory.
It keeps one set of memories, context, and personalization graph across every device and channel.
Layers of memory that improve over time: Session, Semantic, Procedural, Episodic.
Proactive by design: auto‑flags delegatable emails, ranks inbox items by importance, explains ranking.
Tools and skills, with company registries: Salesforce, ServiceNow, Jira, Git, Confluence, BigQuery, Databricks, Postgres, Snowflake, any MCP server.
Governed like an employee: cryptographically attested identity, least‑privilege permissions, audit trail.
Costs you can actually cap: Smart Routing selects cheapest model; hard spend limits auto‑pause agents.
Runs fully LOCAL using the new EmbeddingGemma 2 and Gemma 4 models.
EmbeddingGemma 2, a 740M parameter on‑device model that maps text, images, video, and audio into one shared vector space.
RRSI lets the agent change any part of its harness and puts guardrails on the search itself: limits change bundle size, logs failures, avoids benchmark‑specific tricks.
With Gemini 3.5 Flash, the agent went from 64.6% to 78.7% on Terminal‑Bench 2.1.
I grew the AI hardware‑software developer community to 3,000+ active developers in six months.
Retention was 70% quarterly thanks to bounty programs, open‑source grants, and academic partnerships.
Partnering with Koyeb slashed onboarding time by 90% and drove 3x growth in developer sign‑ups.
I was recognized as a Microsoft Most Valuable Professional (MVP), Jan 2024.
In my current role as Senior AI Product Manager at Google Cloud I lead the Tech Evangelism team for AI agents, while my side projects Unwind AI and Awesome LLM Apps continue to educate and empower a community of over 400,000 developers.
I am a Senior AI Product Manager at Google Cloud, open‑source maintainer, author, and community builder focused on practical AI agent development and developer education.
My title is Senior AI Product Manager @ Google.
I created Awesome LLM Apps, an open‑source GitHub repository with 140,000+ stars.
Awesome LLM Apps is #1 AI Agents GitHub repo with 140k+ stars.
The repo contains 100+ working AI agents, Agent Skills, and RAG applications.
Across Unwind AI and my social channels, I have built a community of 400,000+ AI developers and builders.
I’m also the author of two books.
My prior companies are Arcesium, 256 Network, Jina AI, Tenstorrent.