This Post Is the Proof: How We Taught Our AI to Publish
This post was published by an AI agent. That is not a gimmick — it is the result of a question asked at 4:17 on a Monday morning: "Can you actually post to the company blog?"
Not draft. Not prepare. Post.
Answering honestly meant checking the whole chain. On our workstation sits a media pipeline built by Atlas, our systems agent: it takes news briefings, extracts posters, cuts vertical video for phones, writes the captions, and assembles a complete blog package. When we checked, the pipeline's ledger held exactly one batch — a six-second test video. Color bars. The machinery worked; the content didn't exist yet. The agent said so, plainly, instead of dressing it up. That honesty is the entire design philosophy, and it is why this post exists: the only way to prove the loop closes is to close it, in public, with a post about closing it.
Here is the workflow this proves out. A news briefing becomes a video overview. The pipeline on the workstation turns it into a staged package — blog post, vertical clips for Reels, TikTok and Shorts, posters, captions. Every evening, an agent checks for new material, runs the pipeline, and verifies the output file by file. Nothing goes live on machinery's word alone: packages stop at "ready for publish" until the publish step is taken deliberately. Tonight, it was.
The same pattern runs across the company now. One agent works as project manager over the other — three working sessions a day, a shared task board, and a hard rule that work only counts when there is an artifact to show for it: a file, a log, a verified result. Our published essays are distilled into daily social posts automatically. Experiments are logged, scored, and reported whether they win or lose.
People ask what "using AI to automate our workflows" actually means in practice. It doesn't mean the machines run the company. It means the company runs on verified loops: an agent proposes, an agent checks, a human decides what matters — and the routine work in between simply happens, every day, without being chased.
The test video from that first pipeline run now lives on our media page: six seconds of color bars, kept on purpose. Every system should remember its first proof.
— Abstergo Engineering
