Skip to content
J
← Insights
AI-assisted Development6 min readBy Junmarvi Tampac

AI-Assisted Development: Shipping Real Software as a Solo Developer

The systems on this site — an operations Control Tower for EPC material work, a repository-assessment tool, reporting tools, an inventory prototype — are built by one person. Not because one person writes every line, but because one person directs AI-assisted tooling under rules strict enough that the output can be read, maintained, and defended.

Direction, not delegation

The failure mode of AI-assisted development is treating the model as a contractor: describe the feature, accept the code, move on. That produces software fast and technical debt faster. The approach I use inverts it — I own the architecture, the standards, and every accepted decision; the AI supplies speed inside those constraints.

In practice this means standards exist before projects do. My work runs on a documented foundation: engineering conventions, an AI workflow standard, and security rules that define what tooling may touch and where secrets live. Every project inherits the same context, so the AI works from the same rules I do.

The domain advantage

AI lowers the cost of writing software; it does nothing to lower the cost of knowing what to build. That is why the leverage goes to people with real domain knowledge. Years of material control mean I can specify a reconciliation tool down to its tolerance rules, then check the output against scenarios a generic developer would not think to test.

The honest claim is modest: requirements, build, and testing converge in one head, and the usual loss between 'what operations needs' and 'what was delivered' gets smaller. It does not make one person a team, and it does not make a prototype a production system.

What discipline buys

Documented standards mean a codebase reads consistently even though a model wrote much of it. Security rules mean AI tooling never sees production credentials. Written decision records mean that six months later, 'why does the matching tolerate 2% on bulk materials?' has an answer. That is the difference between using AI and directing it.

AI is part of how I build. It is not the thing I am selling. What I am accountable for is the same thing any developer is accountable for: software that does what it says, and that I can explain.