How to Become an Agentic Architect: Carmelo Iaria's Path

TL;DR: Carmelo Iaria has been working in AI since 2017, before most people in enterprise tech knew what an agent was. Today he trains engineers, product leaders, and consultants from companies like AT&T, Morgan Stanley, and NATO to stop writing fragmented code and start orchestrating systems. The role is called Agentic Architect. The gap between people who understand this shift and those who do not is widening fast.

There is a new role emerging in software development. It does not require you to write more code. It requires you to stop thinking about code entirely and start thinking about systems.

Carmelo Iaria has been preparing for this role for most of his career. He just did not have a name for it yet.

Over 30 years across Europe, Silicon Valley, and Brazil, through 12 years at Cisco and 8 years driving innovation at Claro Brasil, Carmelo watched the same pattern repeat: organizations that treated technology as a tool to execute tasks fell behind organizations that treated it as a system to design. When AI arrived, not the AI of headlines but the real AI of production deployment, enterprise workflows, and multi-agent coordination, that pattern accelerated into a cliff.

He built a framework. He named the role. He started teaching it. His students now include a Cyberspace Technical Director at NATO, a Group PM at AT&T, and senior engineers from Experian, HP, and Morgan Stanley.

This is not a story about learning to code with AI. It is a story about learning to think differently about what building means.

The Career That Led Here

Carmelo started working with AI in 2017, when most enterprise teams were still treating it as a research topic. What he built since then is not just a course. It is a method forged from watching AI pilots succeed in demos and fail in production, over and over.

Carmelo Iaria is self-taught in the ways that matter most. Not self-taught as in no credentials: he holds an MSEE from Politecnico di Torino, a Data Science professional certification from DataCamp, and a Human-Centered Service Design credential from IDEO. Self-taught in the sense that his edge was always curiosity, not a degree program. He needed to understand how things were built, not just what they did.

That instinct drove him into AI in 2017, years before it became the defining professional skill of the decade. He built a Data Science and AI consultancy in Sao Paulo, secured projects at tier-1 financial institutions, and watched from the front row as the technology matured from models that could predict things to agents that could act on them.

What he noticed was the same failure mode playing out everywhere. Organizations were deploying AI pilots. The pilots looked good. Then they hit production and fell apart, because nobody had designed the system, only the demo. The agents had no clear roles. The workflows had no governance. The humans in the loop had no defined relationship to what the agents were doing. Speed had become technical debt.

That failure mode is what drove him to build the AAMAD framework: AI-Assisted Multi-Agent Development, a structured, repeatable operating model that goes from scope to architecture to deployment without losing control in between.

"AI value is won in execution: clear scope, reliable multi-agent workflows, and governance that survives production." — Carmelo Iaria

What an Agentic Architect Actually Does

The Agentic Architect is not the fastest coder on the team. They are the person who decides what gets built, how the agents divide the work, and how the system stays in control when things go wrong.

The term is new. The underlying skill is not.

An Agentic Architect designs multi-agent systems: collections of specialized AI agents that coordinate to complete complex tasks on their own. Think of it as moving from writing individual lines of code to designing a team of workers, each with a defined role, a defined scope, and a defined relationship to the others. You are not the one doing every task. You are the one who makes the system work.

Carmelo's framing: "What if the future of building complex products isn't about being the fastest coder, but about orchestrating a team of specialized AI agents, where your value lies in design, coordination, and vision, not just in the code you write?"

That question is no longer hypothetical. Job postings mentioning agentic AI skills grew by nearly 1,000% between 2023 and 2024, according to data tracked by The Interview Guys. Companies including Deloitte, Salesforce, Apple, and NVIDIA are actively hiring for the role. It has begun appearing on enterprise job descriptions at Cognizant, AT&T, and financial services firms under titles including AI Agent Architect, Multi-Agent Systems Developer, and Agentic AI Lead.

The problem is that most people entering this space are doing it backwards. They are learning tools. They are vibe-coding prototypes that demo well and break in production. They do not have a system. They have a pile of prompts.

That is the gap Carmelo built his course to close.

Why This Is Not a Technical Problem

Carmelo's audience are not beginners. They are experienced professionals: engineers, product managers, consultants who already know their domain. What they are missing is a mental model for how agentic systems are organized, not a coding tutorial.

This is the insight that separates Carmelo's approach from most AI education in this space: the constraint is not technical. It is conceptual.

Most people with a technical background can learn to use an AI coding tool in an afternoon. What they cannot do in an afternoon is answer the questions that determine whether the system they build will actually work: What is the scope of each agent? How do you define context an agent can execute on? Where does human oversight sit in the loop? What makes a multi-agent workflow reliable rather than unpredictable? When does speed become technical debt?

Those are systems design questions, not coding questions. And they require a mindset shift from execution thinking to orchestration thinking.

One of Carmelo's students, Alberto, a Cyberspace Technical Director at NATO, put it precisely: "Once you understand the underlying software engineering principles, AI stops feeling like a black box and becomes a powerful engineering tool."

That shift, from black box to system you can reason about, is what Carmelo's course is built to produce. His target learner is not someone starting from zero. It is someone who senses that AI is changing their job and wants a systematic way to work with agents rather than just reacting to each new tool release.

"This is not a course made of slides, nor one designed simply to tick a training box. It is a course to learn by building, to practice on a meaningful real-world project, and to grow as a software engineer." — Alberto, Cyberspace Technical Director, NATO (verified Maven review)

How He Tried to Solve It Before Maven

Before teaching on Maven, Carmelo built a platform to solve the same problem at enterprise scale. He was right about the need. The timing and the format were wrong. That failure sharpened everything he built next.

The path to Maven was not direct.

Carmelo had spent years watching corporate AI training fail. Not because the content was wrong, but because the format was wrong. Organizations would run a workshop. People would leave with slides and notebooks. Then they would return to their real work and none of it would stick, because they had never actually built anything under real conditions.

He built a platform called The Village in 2024, an AI-infused workplace learning platform designed to solve exactly this problem at the corporate training level. It was a serious attempt. It failed to gain traction. What he learned from it was that the format that actually worked was smaller, more intensive, and more personal than enterprise training budgets wanted to fund.

Maven gave him the format he had been looking for. A small cohort. A real capstone project. Weekly live sessions with direct feedback. No slides designed to be forgotten. Just a framework, a project, and six weeks to ship something production-ready.

The course was named a Maven TOP100. Students from AT&T, HP, Morgan Stanley, T-Mobile, and Cognizant have applied his method. His LinkedIn following has grown to 2,600 subscribers. He is now an O'Reilly instructor and a Pearson SME. The World Economic Forum has featured his work.

None of that would have happened through enterprise training. It happened because he found the right format for the right audience.

DATA TABLE: The Agentic AI Career: Key Numbers (2026)

Sources: HeroHunt AI (herohunt.ai/blog/fastest-growing-ai-roles-in-2026), The Interview Guys (blog.theinterviewguys.com/top-10-agentic-ai-jobs), McKinsey State of AI, Gartner 2025.

What Carmelo Is Actually Teaching

Six weeks. One capstone project. One repeatable framework. Students ship a production-ready multi-agent application and leave with a method they can apply to the next one immediately.

The course is called Become an Agentic Architect. It runs six weeks, one live session per week, with four hours of capstone project work between sessions. It is deliberately small. That is not a limitation. It is the design.

The AAMAD framework structures the work into three phases: Define, Build, Deliver. Each phase has clear deliverables, validation checkpoints, and templates students keep after the course ends.

What students build in the Define phase is not a wireframe. It is a project scope document with research and requirements that AI agents can actually execute on: not vague descriptions but structured context that gives the agents something to work with. This is the foundational skill most people skip, and it is why most agentic pilots fail.

The Build phase covers architecture, component design, and integration: from a multi-agent runtime configuration through frontend, backend, and third-party tools. Students use real infrastructure: Cursor or Claude Code as their agentic IDE, OpenAI API, GitHub, AWS. Not sandboxed environments. Real systems.

The Deliver phase adds what enterprise environments demand: human-in-the-loop checkpoints, observability, automated quality assurance, and deployment. Students present their capstone publicly and leave with a portfolio artifact they can show a hiring manager or a client.

Matt, founder of SAVANT-AI: "Five stars for course logic, course structure, and course value."

Satyan, Group PM at AT&T: "The concepts taught here make the difference between building a product that works and one that doesn't."

How to Evaluate Whether This Course Is Right for You

Carmelo's course is deliberate about who it is for. If you need a slow introduction to AI concepts, this is not it. If you have a technical background, a real project, and are ready to build something production-grade in six weeks, this is exactly it.

The course prerequisites are specific by design. You need to own delivery outcomes in tech, product, or design. You need comfort with Git, GitHub, and basic APIs. You need a real project idea or a willingness to find one in week one.

What you do not need is deep coding expertise. The course was designed to be accessible to people with a light technical background, the audience Carmelo spent years trying to reach through corporate training. The bottleneck is not coding fluency. It is the ability to think architecturally about systems.

The honest signal of whether a course like this delivers on that promise is not the sales page. It is what students say after they finish. Not what they rated it. What they described.

You can read verified student reviews of Carmelo's course at Carmelo Iaria on AllPros. The AllPros Score is a third-party signal built from real outcomes, not marketing, which is exactly the kind of verification this category needs. You can also browse agentic AI courses verified by real students on the AllPros platform.