My AI Engineering Team

I didn’t realize it at first, but over time I found myself building a team—of AI and large language model bots. It started simple: using ChatGPT to craft prompts for Lovable AI. ChatGPT was like my prompt engineer, shaping instructions so Lovable could code precisely what I needed.

As I kept working, this setup evolved. ChatGPT became more than a prompt factory—it turned into a deep research and context engine. I’d have it produce detailed product/prompt requirement documents (PRDs), broken into clear sections. Then I’d feed those sections to Lovable to execute focused coding tasks. What felt like talking to one AI became managing a workflow between specialists.

Eventually, I hit Lovable’s credit limits. Work had to keep moving, so I switched to using Lovable’s GitHub integration and picked up the thread in Bolt, another AI-assisted coding tool. Bolt wasn’t an exact match but close enough to maintain momentum without losing context.

Next came local development with Kilo integrated into VS Code. Now ChatGPT and Kilo work like an architecture team and senior developers. ChatGPT helps me clarify and decide what task to tackle next, while Kilo acts like a senior dev, reviewing code, suggesting improvements, and ensuring quality and consistency.

This isn’t just juggling random tools. It’s a layered engineering team: ChatGPT as architect and lead, Lovable and Bolt as mid-level devs, and Kilo plus local VS Code as senior devs. I’m the team lead, orchestrating handoffs, vetting outputs, and keeping everyone aligned.

This layered approach shows how AI really adds value. It’s not expecting one AI to do everything perfectly. It’s about matching tool to task, combining strengths, and managing transitions like a human team leader.

There are challenges: overlapping roles, miscommunications, switching contexts. Credit limits or quirks mean I sometimes have to switch tools midstream or adjust workflows on the fly.

At its core, this is a leadership exercise. Managing AI like developers requires judgment on when to push, pull back, and maintain consistent voice and direction. It’s a quiet orchestration—“holding the mirror” to keep the process steady.

If you think of your AI workflow as a team, not a solo tool, possibilities shift. It’s not “one AI to rule them all.” It’s a modular, layered system, where your role is conductor, setting tempo and tone rather than just pressing buttons.

That shift changed how I work and what I expect from AI. It’s not magic. It’s collaboration.