The job
“We should publish more. Everyone agrees. Nobody has the afternoon.”
Content marketing fails on cadence, not on ideas. The post that would have been good in week one is still a Slack thread in week six. And the failure is invisible — nobody files a ticket saying “we didn’t publish again”, so it never competes for attention with things that page someone.
The naive fix — point a model at a topic and paste the output — fails differently, and worse. It produces text that is fluent, on-topic, and unmistakably generic. Publishing that costs you more credibility than publishing nothing.
What Forge does
A scheduled workflow that ends in a reviewable pull request:
The scout decides what is worth writing about — what shipped, what broke and got fixed, what an operator learned the hard way.
The writer is a named character , chosen by topic. That matters more than it sounds: the character carries persistent memory of its own past posts, so the voice is consistent across months rather than regenerated from scratch each time. The byline on every post is a real member of the team.
The hero image is generated through the same media API the rest of the platform uses — no separate tool, no stock photography.
The distribution copy is written in the same run, so the cross-post is not a link dump bolted on days later.
Then it opens a pull request. A human reads and merges.
Evidence
This is the easiest claim on the site to verify, because the evidence is public:
- Twenty of this blog’s commits were authored by
simple-container-forge[bot]— the majority of the posts you can read here. - The commit titles say what each run produced: “Publish a new simple-forge.com blog post + hero image + LinkedIn draft”.
- The posts carry a human byline by design — the character writes in that person’s voice, and that person’s name is on it.
We are not describing a capability we built for someone else. We stopped writing our own blog, and this is what replaced us.
What you control
- Cadence and subject sources. What gets watched, how often the workflow fires.
- Voice. Which character writes what. Adding a voice is adding a character, not retraining anything.
- The gate. The default is a pull request a human merges. Publishing straight through is possible and deliberately not the default — a blog is a public voice, and that is the one place where an unreviewed mistake is most expensive.
The honest limits
- A human still merges. We do not recommend otherwise, and we do not do otherwise ourselves.
- It needs a voice to imitate. Cold-starting a brand voice from nothing produces the generic text this use case exists to avoid. Two or three posts you consider on-voice are enough to anchor it.
- Subject picking is the weakest link. A scout watching thin sources writes thin posts. The quality of what gets published tracks the quality of what it is told to watch.
Under the hood
- Autonomous & Scheduled Workflows — the cadence and the guardrails
- Character Agents — the voice, and the memory that keeps it consistent
- Media Generation — hero images through the same gateway
- Git-Backed Audit — every draft and revision is a commit