We shipped 335k+ lines of code and closed $4.5k in 3 weeks

We shipped 335k+ lines of code and closed $4.5k in 3 weeks.

Here is how we are building an AI native company.

Most customers flat out don't believe a 2-person team shipped OXYGEN in a couple of weeks.

Before I get comments from angry devs: We know that LOC's is not the right KPI to measure.

Here's what AI-native means inside our company, end to end.

1) Development

We run 4-6 coding agents in parallel. Claude Code and Codex inside Warp or Superset.

Building the first 80% is the fun part. The last 10% is where it gets brutal, and that's where the systems earn their keep.

We burn ~$1.5k/month on tokens. Coding plus code review agents (more on that later).

2) The product itself is built for agents

If you're AI-native, your product has to be too.

We built OXYGEN CLI and MCP-first, so agents can actually operate it.

Built for agents. Built by agents.

3) Quality runs through a pipeline

Early on it was messy. Probably normal for every startup.

Now every production merge runs through:

→ deterministic tests in the PR → 3 review agents in the background: Greptile, DeepSource, Cubic

Bugs in prod dropped hard.

4) GTM runs the exact same way

Our content engine and our outbound engine are both powered by OXYGEN, which is itself driven by Claude Code.

Newsletters, content, outbound. Same agents, same token budget.

5) Tool selection has one rule

We only adopt tools we can drive via CLI or MCP.

Newsletter platform, internal ops, all of it. If an agent can't operate it, it doesn't make the stack.

PS: I absolutely pretended to work in this picture. A friend told me that you gotta sit in a café and make pictures of yourself pretending to work and this will lead to a lot of engagement on linkedIn. Should I tell him he is an idiot?