WHY I DID IT
I spent longer connecting LinkedIn to AI than writing the post. OAuth, tokens, APIs, permissions and Vercel: all to achieve something apparently simple, allowing my AI agent to publish on LinkedIn.
THE BOUNDARY I WOULD NOT CROSS
The condition was precise: the agent cannot decide to publish by itself. It can prepare content, check it, propose it to me and, after my approval, publish it through the direct connection. AI prepares, checks and proposes. I approve.
WHAT DID NOT WORK
The process was not straightforward. Several technically successful attempts returned a post identifier, but the content was not actually visible. At another stage, a corrupted image caused processing to fail. A positive API response was clearly not enough: the real result had to be checked.
THE CORRECTION
The workflow was simplified, the image asset was checked before upload and publication remained subject to my approval. Only then was the post actually visible on LinkedIn.
THE LESSON
If the architecture for automating a simple task becomes more complex than the task itself, the automation has missed its purpose. AI should not take my place. It should stop me wasting time in the wrong places.