Episode 002: Orchestrating Your AI Engineering Workforce
Episode 002: Orchestrating Your AI Engineering Workforce
In this solo episode, CrewDeck founder Alper Tovi steps back from the product and talks about the shift underneath it: what happens to engineering work when the constraint stops being how fast can you write code and becomes how well can you coordinate the things writing it for you.
He argues that the skill that matters in the new economy is not knowing one stack deeply — it is being able to facilitate a team of AI workers well enough to ship real products across any platform.
Orchestration beats raw model power.
What's Covered
- Why engineers leaving large tech organizations are unusually well equipped to build and sell their own digital products
- The platform gap: fifteen years of depth on one stack, and how little of it transfers when you move to desktop or mobile
- AI code sprawl — when agents generate more code than any human can meaningfully review
- Why your cost per prompt climbs as your repository grows
- Treating every task as a hire: onboarding and offboarding an AI engineer thousands of times over
- Building guardrails so verification does not fall entirely on you
- Running the whole software development lifecycle locally, in isolated worktrees — in CrewDeck, up to twenty parallel sessions powered by Claude and Codex
In My Own Words
The following are the founder's original recorded notes, lightly edited for reading.
The shift already happening
It is remarkable that when companies lay people off from tech, those engineers are incredibly well equipped to build and sell their own digital products. That could be a mobile app, or a desktop app like CrewDeck. I think that is the future. People who know how to use these tools are going to be far more inclined to build their own digital products and contribute to the economy that way.
As we observe more of a shift from a human workforce to an AI workforce, the key is this: as long as a human being is able to facilitate the work of AI workers, they are more likely to succeed in the new economy. That is the direction things are going, and CrewDeck is built to solve that problem.
The platform gap
Say you are a developer. You know your platform, you know your stack. My stack is Salesforce, and I have been on that platform for about fifteen years. I always had a hard time grasping other technologies, because I was too busy with my work, or I did not have the time and opportunity to experiment.
If you are technical, you already know the concepts do not translate cleanly. Working on a cloud platform like Salesforce is not the same as building a desktop application in Electron. I had ideas I wanted to build. And then I started thinking about the new economy, and how individuals are going to need to build their own digital products.
So my problem became: how do I move from a very enterprise-focused CRM skill set to building web applications, desktop applications, and mobile applications?
Where it started to break
AI is tremendously helpful as it currently stands. But I found it genuinely challenging once I had to orchestrate multiple sessions and queue up work across them. That has been my challenge.
The tools are creating so much code that it is literally impossible for a human being to have the cognitive capacity to follow what is happening in that code — to catch problems and fix them. So there is a real need to build your own team of AI engineers that work in coordination, doing all of these tasks that used to be human tasks.
This is also an experiment, and an expensive one. When you create an application directly with a single prompt, your cost is likely to be small. But two things happen as your codebase grows. Each individual prompt now triggers the same kind of research all over again, so your cost climbs. And on top of that, you will have a hard time keeping standards consistent if you have never had to manage engineering resources before.
Every task is a hire
Working with AI engineers is similar to managing people. For every single task, you are hiring an engineer, you are onboarding that engineer, and you are offboarding that engineer. That applies to AI engineers exactly as it does to human ones.
But how do you handle that when you are working on an application that will require thousands of sessions? To give you a sense of scale: I have been developing CrewDeck for about six months, and in six months I have exceeded three thousand pull requests. The volume is incredibly high.
We need ways to create guardrails around what the AI is producing, and ways to check and verify it — and we do not necessarily have to do all of that ourselves. There are ways to achieve this.
What CrewDeck does about it
CrewDeck takes the entire software development lifecycle and puts it in front of you. It orchestrates your AI workforce in individual worktrees on your local machine. It does not do that on a remote server, and it does not hand your code to an external service — which is also what keeps the cost extremely low.
From the initial mock-up, to how you create the task, to how you eventually produce a reviewed result, all of it is delivered in the core functionality of the platform.
The goal is to give you the ability to orchestrate AI the way you actually envision it — with self-validation and self-checking around the work, so that as you gain confidence in the process, more of the verification burden comes off you.
What's next
There are features I am genuinely excited about — including having your agents hold a structured debate with each other about the right technical solution before anyone commits to it. I will be walking through these in future sessions. Watch this space.
The audio for this episode is a NotebookLM overview generated from the founder's original recorded notes above, so the narration is AI-generated and its wording differs from the text. The video version covers the same material.