AI as a Collaborator, Not an Autopilot
A two-track AI workflow — one for turning interviews into content, one for implementation — with human review as the seam between them.
A recent example is the rebuild of this professional site itself.
I use an AI content partner to interview me about projects and career experiences, extract the engineering decisions and outcomes from those conversations, and turn them into structured content packages. A separate AI-assisted implementation workflow then consumes those packages as backlog and content inputs.
The separation is deliberate. The content process isn't allowed to invent accomplishments, and the implementation process isn't responsible for inventing my professional story. I review the facts and decisions before they become published content, while implementation work is tracked through normal GitLab work items and source control.
That workflow gives me much of the speed of AI-assisted development without giving up human judgment. AI is useful for accelerating research, organizing information, drafting, and identifying gaps. It's still my responsibility to decide whether an assertion is true, whether a technical recommendation is appropriate, and whether the finished work actually represents what I intended.
A second example: Terraform across multiple domains
The same pattern shows up in infrastructure work, not just content. While managing Terraform across multiple domains and environments — including the serverless email platform I built for goodintechnology.com — I used an AI assistant to ingest the existing Terraform and state context, compare proposed infrastructure changes against current state, and validate architectural assumptions before I applied anything.
The AI didn't design that architecture. It acted as a review and validation partner, reducing the amount of second-guessing required when reasoning across interconnected state — while I stayed responsible for the actual decisions and implementation.