Disclosure: This article began as a conversation with an OpenAI model in Codex and was edited for clarity and publication. The author reviewed and directed the final piece.
Substack had been nearly incomprehensible to me.
That is not because publishing is a mysterious activity. I know what I want to say. I can recognize a good headline, decide whether a paragraph sounds like me, and tell when an article ends too abruptly. The difficulty was everything surrounding the writing: sections, settings, confirmation emails, welcome emails, headers, footers, subscriber options, and buttons whose meaning was not always obvious.
The usual answer would be to learn Substack.
I am beginning to think that is the wrong answer.
A publication assembled through conversation
I recently had a conversation with an AI about AMD Ryzen AI, NVIDIA DGX Spark, and CUDA. The exchange seemed useful, so I decided it should become an article.
From there, the work expanded. We edited the conversation, created an image, saved the assets, published a canonical version on Kurka Labs, prepared versions for Medium and Substack, and distributed trailers through social channels. Then we turned to the longer-term question: how should a reader become a subscriber?
That led us back into Substack.
Together, we renamed the publication Conversations with AI, clarified its description, enabled email confirmation, revised the email header and footer, and rewrote the welcome email for new subscribers. The AI could inspect the live interface, find the relevant controls, make many of the edits, and verify the result. When the browser or platform required my direct action, it told me exactly what to do.
At one point I could not find the blue Save button. The reason was simple: the change had already been saved and the editor had closed. The AI reopened the setting and verified that the new welcome message was intact.
That small moment captured the larger idea.
I did not need to become an expert in the machinery of Substack. I needed to remain responsible for the publication while allowing the AI to become competent with the machinery.
Delegation is not abdication
“Let your AI learn Substack” does not mean turning over your identity or publishing whatever a machine produces.
The human work remains essential:
- Choose the ideas worth publishing.
- Decide what is true, useful, and genuinely yours.
- Set the tone and reject language that feels artificial.
- Approve public actions and commercial choices.
- Notice what readers respond to.
The AI's role is different:
- Learn the interface and remember the workflow.
- Turn a conversation into a coherent draft.
- Adapt one article for several destinations.
- Maintain links, images, metadata, and publication records.
- Check that what was supposedly saved or published is actually there.
- Reduce a maze of settings to the next clear decision.
This is not autopilot. It is a working relationship with a useful division of labor.
Stop collecting interfaces
Modern work asks us to learn an unreasonable number of interfaces. Every service has its own vocabulary, navigation, permissions, publishing sequence, and promotional machinery. Much of that knowledge is temporary. The service changes its design and yesterday's expertise becomes today's confusion.
AI changes the economics of learning those systems.
Instead of spending hours memorizing where Substack placed a particular setting, I can state the result I want: every new subscriber should receive a clear welcome, understand what Conversations with AI is, and have a useful place to begin. The AI can translate that intent into platform operations, while I review the consequential choices.
The durable skill is no longer knowing every button. It is knowing what outcome you want, how to describe it, and where human judgment must remain in the loop.
What we are building next
The goal is a publishing pipeline that begins during ordinary work.
An idea appears. A conversation develops it. The useful part becomes an article. The article is adapted for Kurka Labs, Substack, and Medium. Shorter trailers are prepared for LinkedIn, X, Facebook, and Instagram. The archive, images, links, and subscriber invitation remain consistent.
I should not have to stop the work, change projects, reconstruct the context, and operate five unrelated dashboards every time an idea deserves an audience.
The promise is not “one click and the AI does everything.” Some platforms still require confirmation, judgment, or a human hand on the final control. The promise is continuity: the publishing operation can stay beside the thinking instead of interrupting it.
So perhaps the next time a platform feels incomprehensible, the answer is not another tutorial.
Don't learn every corner of Substack. Teach your AI what you are trying to accomplish, supervise the decisions that matter, and let it master the interface for you.
Frank Kurka
[email protected]
kurkalabs.dev