Loading...

🚀Join Scale24/6 — Israel’s Exclusive Community for VP R&Ds, CTOs & Engineering Leaders. Apply Now👉

This morning, 60 VPs of Engineering and Technology cleared an hour from their busy schedules in to talk about AI, and it’s not like they had nothing else to do.

We hosted an online session in our Scale 24/6 community with Omri Chen of Tailor Brands, who gave us a look at the ADLC processes they’re building and rolling out. It was one of the most interesting sessions I’ve taken part in recently. But what I found to be just as interesting as the content, was the reaction to it. The questions simply didn’t stop coming, to the point we ran out of time before everyone got to ask what they wanted.

And to me, that tells us a whole story.

What ADLC Actually Means?

It stands for AI Development Life Cycle. It isn’t SDLC with another tool built into it, it’s a shift in how work is divided within the cycle itself.

In practice, it looks something like this:

  • Requirements: Written in a way that an agent can work off of them directly, not just a human.
  • Writing Code: The agent produces the code and the developer directs and approves it.
  • Testing: Evals which measure the output quality, not just whether or not the code runs.
  • Production: Monitoring that covers model cost and output accuracy, not only performance and errors.

And then there’s the bottleneck, which is the real deal. It used to be all about how much code you could write. Today it’s about how many decisions you can make and verify. Once code generation stops being a constraint, decision quality and verification speed become the new constraint. That changes the way in which an engineering organization is structured, not just whichever tools it uses.

The Real Sign From That Session:

Even the most experienced people in the industry, the ones running large teams and leading engineering organizations for years, are in a serious learning curve right now.

I mean that as a compliment. When people at that level ask basic questions in public, it means the field is genuinely new, not that they’re behind. Anyone who feels alone during today’s uncertainty is wrong.

What This Means in Practice:

  • The first investment isn’t in tooling, it’s in requirements. An organization with sloppy specs won’t benefit from an agent that writes code quickly, it’ll just get the wrong code faster.
  • Testing is the new bottleneck. If code generation tripled and verification stayed manual, you gained nothing.
  • It affects who you hire. The profile that fits ADLC is someone who can make architectural decisions and validate output, not whoever types faster.

Why is This Happening in a Community?

One of the things I love most about my job is being able to connect people with rich experience who, despite everything they already know, stay curious and keep learning from each other. A huge thanks goes out to Omri, for a great session and for being willing to share everything. I’d also wish to thank everyone who joined, for your curiosity and questions. Next time we’ll clearly need to leave more time for them.

If you’re a CTO, VP R&D, or technology leader and you’re not part of Scale 24/6 yet, you’re more than welcome to apply. If you’re building a development team which will adapt this work structure, I’d be happy to talk to you.

Questions I Keep Getting:

Isn’t ADLC just giving every developer Copilot?

No. Copilot speeds up an individual developer inside an existing process. ADLC changes the process itself like who writes the spec, who generates the code, and who verifies it. You can roll Copilot out to the whole team in a week and see a local improvement without a single stage of the cycle actually changing. That’s exactly where most organizations stall and conclude that “it doesn’t work.”

Where do you start in an existing organization?

You start with one feature, end to end, and not by replacing the entire department’s process. Pick something of moderate importance, run it through the new methodology, and measure what happened to the time and quality. An organization that tries to swap out its whole SDLC at once finds out a quarter later that it has no idea what worked and what didn’t.

How do you know if it works?

Not by lines of code and not by how many features shipped. There are three metrics you can compare before and after. The time from idea to production, the share of bugs that reach the customer, and infrastructure and model cost per feature. If the first went down and the other two went up, you didn’t gain anything, you just moved the cost somewhere else.


Roma Timoshpolski is the founder and CEO of YozmaTech, which helps dozens of Israeli tech companies build global development teams. This column is based on his LinkedIn post.

You May Also Read:

Unsure Where to Start?

Get free consultation to find the right solution for you.

two male IT employees
chat circle
whatsapp icon green