# MaGa Srl — Agentic AI Consulting (full) > Fractional CTO & Agentic AI Consultant. I put AI agents into production for teams with real, repeatable processes — and I'm the senior technical call for founders who need it without a full-time hire. On Google and Claude. Production, not demos — and an honest call on whether AI even fits. Matteo Gazzurelli — Fractional CTO & Agentic AI Consultant, Brescia, Italy. Over 20 years building software, 250+ projects shipped across mobile, cloud and distributed systems. Google Partner through MaGa Srl. Organizer of GDG Brescia. Site: https://gazzurelli.com · Email: matteo@gazzurelli.com --- ## Positioning I focus deliberately on two platforms. In production, depth tends to beat breadth: knowing two platforms well — their real tradeoffs, where they crack and where they shine — is usually worth more than knowing eight frameworks shallowly. The two platforms are equivalent choices with real tradeoffs. There is no universal winner; the right call depends on the case, and I'll tell you which. --- ## The problem I solve Most companies experimenting with AI agents get stuck somewhere between the demo and the deploy. The proof-of-concept works in a notebook, then breaks the moment it meets real users, real latency, real edge cases. The team picks a framework based on a Twitter thread, and six months in there are three half-working agents and no clear path to production. I help teams move past that wall — choosing the right platform for their case, designing systems that handle failure gracefully, and getting agents into production where they actually do useful work. --- ## The stack I work with ### Google — Gemini Enterprise Agent Platform The evolution of what used to be called Vertex AI. Agent Runtime, Memory Bank, Agent Identity, Agent Gateway, plus the Agent Development Kit (ADK) for code-first builders. A strong default for enterprise: identity, governance, and the GCP ecosystem. If the client is already on GCP, the choice is often made. ### Claude — Agent SDK and Code Anthropic's Agent SDK for building production agents on Claude, plus Claude Code for the development workflow. A strong choice when the model itself is the differentiator — long context, instruction-following, agentic coding — or when the team runs outside Google's ecosystem. ### Protocols and supporting tech MCP for tools, A2A for agent-to-agent communication. Firebase, Cloud Run, BigQuery for infrastructure. Flutter and Next.js for mobile and web. ### My take on the tradeoff Google gives more enterprise rails out of the box, at the cost of tighter coupling to GCP. Claude gives model quality and a cleaner platform surface, but it's a newer offering with a thinner ecosystem. Both are valid. I default to Google for enterprise work and reach for Claude when the model's strengths drive the case. --- ## How I work - **Discovery & Strategy** (typically 2–4 weeks): map your use cases, evaluate Google vs Claude for each, design the architecture and deliver an implementation plan. Output: a document your team can build from, and a clear go/no-go on each use case. - **Build & Deploy** (typically 8–16 weeks): hands-on implementation alongside your team — architecture, code, deployment, observability, evaluation. I work as a Fractional CTO embedded in your team, not as an external vendor handing over a black box. - **Ongoing Advisory** (3–12 months): for teams already building who need a second pair of eyes on architecture, code review, hiring and roadmap. Monthly retainer, no long contracts. --- ## What you get - **An honest assessment.** I'll tell you when an agent is overkill, when a workflow is enough, and when you should wait six months. - **Production-ready architecture.** Patterns tested across real implementations — not slideware. - **Knowledge transfer.** Your team learns; I don't build dependency on myself. - **Vendor honesty.** I'm a Google Partner and Google is my default, but I have no hidden kickbacks. I'll recommend Claude when Claude fits, or tell you "build it yourself, don't hire me" when that's the right answer. --- ## Credentials - **Google Partner** through MaGa Srl — Google Cloud Select Co-sell Partner and Google Workspace Select Co-sell & Services Partner. Verify on the Google Cloud Partner Directory: https://cloud.google.com/find-a-partner/partner/duckma - **GDG Brescia organizer** — I run the Google Developer Group in Brescia, building the local agentic AI community. - **Google Cloud & AI certifications** — multiple, kept current. - **Over 20 years in software**, 250+ projects shipped. --- ## FAQ **Which platform should I use — Google or Claude?** It depends on your case, and I'll tell you which. Google is my default for enterprise clients on GCP; Claude tends to win when the model itself is the differentiator or you run outside Google's ecosystem. **Do you still work with CrewAI?** I worked hands-on with CrewAI and most other agentic frameworks for a couple of years — good for breadth, useful for prototyping. I've since narrowed my consulting stack to Google and Claude. CrewAI is a reference point in my background, not part of the stack I lead with today. **Do agents really save time?** Sometimes a lot, sometimes not enough to justify the effort — it depends on the process. Any cost or time figure I give early on is a rough estimate to frame the conversation, not a guarantee. The honest way to find out is a small, contained build with measured before-and-after. **What does an engagement cost?** Cost depends on scope. I'll give you a concrete number after Discovery, not a fabricated range before I understand the problem. No long contracts, clear stop points. --- ## Contact Email: matteo@gazzurelli.com — I read and respond to every message personally. Based in Brescia, Italy. Agentic automation for teams in Northern Italy and Italian-speaking Switzerland (Canton Ticino), on-site when it helps — see https://gazzurelli.com/consulente-ai. Fractional-CTO work is remote, with international teams — see https://gazzurelli.com/fractional-cto. Note on naming: "Vertex AI" is now the Gemini Enterprise Agent Platform. CrewAI, LangGraph and AutoGen appear here only as historical references, not as part of the current stack.