EZOps - Agentic-AI Deep Dive

A project manager pulls up two estimates on his screen. Same developer. Same client. Same kind of work. Last year: 200 hours. This year: 20.

He checks the numbers twice. The developer shrugs.

“Yeah, that was a year ago. Now I have these AI-tools. It’s easier. It was fast.”

The number isn’t wrong. The world moved.

Thiago Maior knows this moment. He built a 70-person infrastructure company from zero external funding, scaling from freelance developer in Rio de Janeiro to CEO of a US-based firm over a decade of organic growth. He lived through the DevOps revolution. Now he’s watching AI compress that same transformation into months.

In this week’s episode, Thiago and I sat down to map the terrain. What he sees from the EZOps Cloud CEO’s chair should matter to every technology leader making investment decisions right now.

DevOps gave you seven years. AI gives you months.

Thiago started working with DevOps around 2014, four years after the movement began. He thought he was late. People around him resisted. “How can you automate creating a server? How can you automate a pipeline? No, you have to test manually to guarantee that it’s working.“ That resistance lasted years. Companies that ignored DevOps for seven years still survived. They caught up. Eventually everybody learned containers, pipelines, auto-scaling.

AI is a different creature.

“It’s like DevOps from 12 years ago. Everybody wanted to talk about that. Now everybody wants to talk about it. It’s DevOps happening all over again, even higher, bigger.”

But here’s what Thiago sees that most miss: DevOps was a technical shift. It changed how infrastructure got built. AI changes the DNA of an organization. It touches hiring, pricing, project estimation, quality assurance, client relationships, and the very business model that keeps the lights on. That is why seven years of delay, survivable with DevOps, could be fatal with AI.

The data confirms the terrain. 84% of organizations have adopted AI (McKinsey 2024). But METR research found that teams using AI coding tools were actually 19% slower on real engineering tasks, while believing they were 20% faster. A 39-point perception gap. The macro numbers say the revolution is here. The micro reality says most aren’t ready for it.

When the estimates collapse

Thiago’s project managers are watching it happen in real time.

“This project would take 200 hours. Now it’s taking 20, 30.”

Same developer. Same quality expectation. The compression is not theoretical; it’s showing up in invoices and statements of work. For any company selling hours of engineering time, the math is existential.

“Everyone that is selling hours of work is gonna have a problem. If they don’t have a problem right now, they will have a problem shortly.”

Repackaging hours into fixed-price engagements doesn’t solve it either. Clients see a package that used to take 200 hours now taking 20. The value perception shifts whether you label it differently or not.

But the real danger isn’t the compression itself. It’s what happens next. Companies feel the pressure. Boards demand cost reduction. Leadership enforces AI adoption. Layoffs follow. And then comes the moment that should keep every CTO awake: the people you let go are often the ones who could have governed what AI produces.

“They already had those layoffs. They already fired the ones that are actually the most experienced, the better guys. So they’re gonna need help.”

The governance question lands here, hard. When AI generates code at scale and the senior engineers who could review it are gone, who is checking the output? Thiago saw the evidence firsthand: a CTO told him his team estimated four months for a set of features. The CTO built it himself with AI in two days. That speed is extraordinary. But speed without verification is how AWS-level outages happen from vibe-coded infrastructure.

Not ready, but forced to move

I asked Thiago directly: are organizations ready to harvest the true potential of AI?

“They are not ready. But here’s the thing. The first ones that are actually using AI are their developers.”

The adoption pattern is bottom-up. Developers first replaced Stack Overflow with ChatGPT, and now they’re replacing it with Claude Code/Codex. They put code in, code comes out, it compiles, it works. They get faster. Their managers notice. The productivity signals flow upward. But the signals are perception, not measurement. Nobody is tracking whether the output is actually correct, secure, or maintainable at scale. The rework cost is often hidden.

This is the gap between adoption and readiness. Organizations are adopting because the pressure is real; market compression, board expectations, competitive anxiety. But adoption without readiness is where reputations, cash flow, and client trust come apart. People spending two clicks on tests & reviews that should take ten hours. People not reviewing because vibe coding “just works.” Companies choosing tools that were themselves vibe-coded and don’t hold up under production load.

You’re not behind because you hesitated. You’re navigating terrain that didn’t exist 18 months ago. The organizations that stumble here aren’t reckless; they moved at the speed the market demanded. The finish line moved before they could cross it.

This is what the AI Readiness Assessment measures across four dimensions:

Can your people focus under AI’s cognitive load?
Can they challenge AI on technical correctness?
Is product intent clear enough to generate unambiguous specs?
Are feedback loops fast enough to catch errors before they compound?

Most organizations score well on adoption. Almost none score well on readiness (and we ran hundreds of them).

The engineering discipline layer

Thiago’s team spent a year building Ace, their AI agent for infrastructure and development. The proof of concept was, in his words, “crazy good.” Then reality hit. Months of instability. Unpredictable outputs. Models responding differently to the same inputs (AI determinism is something you must deliberately build, as with Alessandro Di Gioia, we did with nWave.ai**).

What pulled them through was engineering discipline. Not a better model; a better architecture. Multi-agent systems where five or six specialized agents handle different aspects of every operation. Right models for the right tasks. Right prompts in the right context. Careful management of what information gets injected and when. Thousands of tests.

“AI also needs to be driven by humans. The fact that we have engineers that really understand the technology and know how to use it. That’s when they have the actual benefit out of it.”

Thiago arrived at this through a year of heuristic iteration. What’s remarkable is that the conclusion matches what every serious framework now converges on: ungoverned AI is expensive guessing. The answer is not less AI. It’s more discipline. Expectation-driven development where human intent gets translated into iterative, verifiable specifications before AI writes a single line of code. Deterministic execution where you can predict and reproduce agent behavior. Counter-agents that verify what creator agents produce, before humans get cognitive overloaded.

This is the shift from vibe coding to AI augmented engineering. And it’s the last mile that separates organizations riding the AI wave from organizations being swept under it.

The project manager’s estimate didn’t shrink because the developer got faster. It shrank because the world changed. The 7-year window that DevOps gave us has closed. AI moves faster, reaches deeper, and the organizations that wait will not get a second chance to catch up at leisure.

But the path forward is not starting over. You made the right investment. Your teams are already using AI. The terrain shifted before anyone could draw a complete map.

I’ve walked this last mile with hundreds of technology leaders. Not to sell frameworks, but to help them see clearly where they stand and what the next step looks like. The AI Readiness Assessment takes 30 minutes and gives you a map of your four dimensions; no strings, just clarity on the path ahead.

When you’re ready to walk the final stretch, I’m beside you. Your timeline. Your choice.

Listen to the full conversation with Thiago Maior

Take the free AI Readiness Assessment we integrated inside nWave https://nwave.ai/go/discipline

Next week: we go deep into the frameworks that explain why every major player, from Google to Anthropic, is building the same governance architecture into their AI systems.