<start of post>
Most AI implementations fail because they try to replace the person,
instead of the part of the day the person hates.
I was looking at a workflow for a legal team last week. They wanted an agent to 'write the brief.' It sounds like the dream. One click, job done.
But when we sat with the associates, the writing wasn't the bottleneck. The bottleneck was the four hours spent hunting through three different databases to find the one ruling that mattered. The 'glue' between the research and the draft.
So we didn't build a writer. We built a scout.
The scout doesn't touch the draft. It just lives in the sidebar and surfaces the files they used to have to hunt for. It takes the mechanical friction out of the way so the judgment can happen faster.
Adoption isn't about the power of the model. It is about the placement of the tool.
If you put the tool where the work is already happening, people use it. If you ask them to go to a new tab and 'learn a new way,' they won't. Even if it's better.
I'm talking more about this 'mechanical layer' in my latest essay on Substack.
What is the one part of your stack that feels like 'glue work' right now?
Link in comments.
<end of post>
