Jeff Niebuhr

Founder & Operator

This did not begin as an AI initiative.

It began because posting videos was brittle.

I was operating primarily in a TikTok-first world. Videos were handed to a human poster. Files were lost. Duplicates were posted. Tracking what had already gone out required manual tagging and increasingly complex folder conventions. Every attempt to fix the workflow locally made the system more fragile.

At first, I tried to solve this by tightening process. That failed.

The inflection point came when I stopped asking how do we post this? and started asking who decides what gets posted?

That question forced a shift from workflow to architecture.

Once decision-making moved upstream, posting became a downstream concern.

Instead of posters hunting for files, I introduced a selector whose only job was to decide what should go out next, independent of who or what was doing the posting. That worked briefly, but selection logic embedded directly in code proved brittle. Every change required edits. Every edge case multiplied complexity.

The next shift was philosophical: selection logic belongs in configuration, not code.

Song, video type, platform, timing, and later ranking were externalized. Behavior could be changed instantly without rewriting programs. This was the first moment the system felt alive.

Selecting one item at a time was still fragile.

If posting stalled, the system stalled. If priorities shifted, everything had to be recalculated. The breakthrough was the queue.

Instead of deciding the next video, the system continuously computes the next batch—sometimes 10, sometimes 100—based on explicit criteria. Posters simply consume from the queue.

This allowed:

• human and automated posters to behave identically
• priority shifts without chaos
• resilience to interruption

Queues turned decision-making into a buffered system, not a just-in-time guess.

Once selection stabilized, the next problem became obvious: is the system making good choices?

That question led to scorers.

Each platform exposes different metrics. Those metrics are noisy. But they are signals. I built per-platform scorers that retrieve performance, compare results to rolling baselines, decay priority after posting, and increase weight when content outperforms recent averages.

This is not optimization theater. It is a feedback loop that learns what to try next.

The system does not fight platform algorithms. It listens to them.

As output increased, inspection no longer scaled.

Supervision had to move from inspection to health monitoring.

Each subsystem emits factual status signals. Those signals are collected into a daily report that answers simple questions:

• Is the system producing?
• Is it blocked?
• Is something drifting?

Red, yellow, green. No narrative overhead.

Not everything is automated.

Some steps remain human because platforms require it—approvals, tokens, permissions. Others remain human by design. Creativity, artistic direction, and taste are not delegated.

JITR is not an AI artist stable. It is an AI company that supports real artists.

The goal is not replacement. The goal is leverage.

Finance entered the system sideways.

To produce a monthly fan update, I needed a factual answer to what actually happened this month. That required collecting every touched file and categorizing real work. That same reconstruction became the basis for billing artists.

Only later did another question emerge: what does it cost to run this company?

That produced a deliberate split:

• Billing accounting — what value the artist received
• Cost accounting — what it takes to operate the business

This separation makes scaling beyond a single artist possible.

None of this was possible pre-AI.

Not because the ideas were novel, but because the execution cost would have been prohibitive. AI turned architectural intent into running systems. It made failure affordable and correction fast.

This is not AI augmenting a development team. This is AI as the development team, supervised by an operator.

What emerges is not a toolchain, but an operating model:

• configuration over code
• queues over immediacy
• feedback over assumptions
• automation as leverage, not spectacle

This is what an AI-first business actually looks like when it runs.

The system was built in the context of a record company, but its structure is industry-agnostic. Product, distribution, feedback, finance, and reporting are universal business primitives.

JITR demonstrates how those primitives can be recomposed when AI is treated as workforce rather than tooling.