Serban Petrescu

SVP of Eng at Trilogy. Former TA & start-up founder. 🚀
Tech chameleon: SAP, Spring, Node, AWS, React. 🌐
Exploring GenAI in development. 🤖

#GenAI #TechLead

[Serban Petrescu @TechLeadWAi]

AI raises the bar in software engineering. It’s an amplifier, not a replacement. But only disciplined teams thrive. Here’s how we tackle the AI era:
https://t.co/ve4k8IJ19d

[Serban Petrescu @TechLeadWAi]

The university system was built for a slower world. It prioritizes research, not teaching. It moves too slowly. Bootcamps aren’t perfect, but they’re evolving fast.

If academia doesn’t adapt, it will lose its place, and it's a damn shame.

[Serban Petrescu @TechLeadWAi]

Is a CS degree still worth it? A LinkedIn study says: maybe not. After a year, 92% of CS grads and 92% of bootcamp grads land software jobs.

Same outcome. But one takes 4 years and heavy debt. The other takes a few months. So what are we really buying?

[Serban Petrescu @TechLeadWAi]

I’ve also taught in academia. Third-year students couldn’t model basic software systems. They studied C and physics, but had no clue about architecture, layering, or real-world engineering. I had to fill those gaps myself.

[Serban Petrescu @TechLeadWAi]

Both models require migrations. In SQL, they’re explicit. In NoSQL, you hide them in app logic.

More code, more risk. Unless you're operating at Amazon scale, schema-less isn’t freedom; it’s complexity with no payoff.

[Serban Petrescu @TechLeadWAi]

An exec told me this week: “Just dump JSON into the database. No need for schemas.”

That mindset sounds agile, but it breaks the moment your app needs to read that data. And let’s be honest, every app needs to.

Schema-less just means schema chaos.

[Serban Petrescu @TechLeadWAi]

NoSQL isn’t about speed. It’s about scale. DynamoDB and Cassandra shine when you know access patterns and need global throughput. Most apps don’t.

For evolving products, SQL gives clarity, safety, and flexibility without scattered schema hacks.

[Tom Dörr @tom_doerr]

build real-time voice and vision AI agents

[Serban Petrescu @TechLeadWAi]

Leaders need to build cultures where curiosity is rewarded, even when experiments fail. A tool that didn’t work still saves time if we learn from it.

What we need isn’t 100 solo experiments. We need a system where learning compounds across the team.

[Serban Petrescu @TechLeadWAi]

“Everyone should spend 2 hours a day exploring AI tools” sounds smart. In reality, it’s a wasteful solo grind.

In my team, we’ve seen better results by structuring exploration, rotating who scouts tools, and sharing findings. Everyone contributes, no one duplicates effort.

[Serban Petrescu @TechLeadWAi]

Our model is bottom-up. Anyone can suggest tools worth exploring.

We track ideas in a shared backlog, assign scouts to evaluate, and let adoption emerge through peer evangelism, not top-down rules.

AI tools spread faster when they’re proven useful by peers, not imposed.

[Serban Petrescu @TechLeadWAi]

The illusion: AI saves time. The reality: it moves the cost to validation. Outputs look polished but carry hidden bugs and weak decisions.

AI is not a brain. It's a high-speed assistant. Treat it that way or watch productivity collapse under a mountain of invisible tech debt.

[Serban Petrescu @TechLeadWAi]

Some say AI is a developer productivity boost. Others cite the 2024 DORA report, which shows a drop in release frequency and quality. I've seen papers claiming a 19% decline. At the same time, GitHub and Anthropic say AI writes 80% of the code.

So what's the reality?

[Serban Petrescu @TechLeadWAi]

In my team, per-capita productivity has tripled, but only because we kept humans in charge.

AI is great at typing, scaffolding, and research. But the moment you let it decide things, especially small architectural choices, you lose context, and everything downstream suffers.

[Serban Petrescu @TechLeadWAi]

AI is a multiplier, if you already know how to think. For beginners, it’s become a disguise. A shortcut that robs them of real skill.

We don’t need to ban AI to avoid this. We need to rethink how we teach. Because without foundational thinking, we're not training engineers.

[Serban Petrescu @TechLeadWAi]

AI didn’t just kill LeetCode. It’s quietly collapsing how junior engineers learn. And almost no one is talking about it. 🧵

[Serban Petrescu @TechLeadWAi]

The result?

Many juniors today are indistinguishable from casual ChatGPT users. Same prompts, same output, same shallow grasp.