I built two open-source tools faster by letting AI write most of the code

Source: Dev.to

What actually changed (and what didn’t)

I didn’t “let AI build a product”.

I still:

• define the architecture,
• decide what exists and what doesn’t,
• control data models,
• review every line,
• say no very often.

What changed is simple: I don’t type most of the code anymore. The AI does.

This is not about prompts or magic

This has nothing to do with:

• clever prompts,
• autonomous agents,
• “vibe coding” without thinking.

It works because:

• I know exactly what I want,
• I know when something is wrong,
• I know how the system should behave.

The AI is fast. I am precise. That combination matters more than prompts.

How I actually use AI to code

My workflow is closer to this:

1. I describe very precisely what needs to be implemented.
2. The AI writes the code.
3. I read it like a strict reviewer.
4. I ask for corrections, refactors, deletions.
5. I integrate or reject.

The AI writes faster than I ever could. I think slower, but better. It feels very similar to:

• the first smart autocompletion,
• then IDE refactoring tools,
• then real-time linting.

Same shift, bigger scale.

Two different projects, same approach

Ackify

Ackify is an open-source tool to handle internal document acknowledgements—proof that people actually read internal policies, procedures, or mandatory documents.

It required:

• a clear domain model,
• strong constraints,
• no feature creep.

AI helped write:

• handlers,
• storage layers,
• repetitive logic.

I stayed in control of:

• scope,
• semantics,
• guarantees.

SHM

SHM (Self-Hosted Metrics) is much smaller. It answers one question:

“Is this self-hosted app actually used?”

It was built in days because:

• the scope was tiny,
• the rules were strict,
• the AI handled most of the boilerplate.

Different scale. Same method.

AI as a force multiplier, not a decision maker

The biggest misunderstanding about AI‑assisted coding is thinking it replaces thinking. It doesn’t. It replaces:

• typing speed,
• mechanical repetition,
• obvious glue code.

The moment you stop knowing what you want, the output degrades immediately. AI is not creative in architecture; it is efficient in execution.

Why this matters for solo developers and open‑source

For solo developers, time is the real constraint. AI doesn’t give you ideas. It gives you throughput. That makes it possible to:

• explore ideas faster,
• kill bad ones earlier,
• finish small useful tools instead of polishing one forever.

That’s exactly how many open‑source needs are met: small tools, precise scope, fast iteration.

Closing thoughts

I don’t feel replaced by AI. I feel amplified. I still design. I still decide. I still review. I just don’t type as much anymore. And that’s fine.