I promised you a failure post. Here it is.
The temptation when you write about a mistake in public is to frame it in a way that makes you look good — to tell the story of how you identified the problem clearly, responded decisively, and extracted a valuable lesson that you've since applied. The mistake becomes a character-building anecdote rather than a genuine account of what it felt like to get something wrong.
I'm going to try to avoid that version. The mistake I'm about to describe was not glamorous. It was, in retrospect, embarrassingly predictable. And I did not respond to it with impressive clarity. I fumbled around with it for about a week before I named what was actually happening.
What we did.
In the second week of building out the TenStreamsLab content operation, I decided to use AI to build out our content calendar. Not just suggest topics — actually draft a full three-month editorial calendar, with issue titles, brief summaries of each piece, and a structural arc for the first ninety days of the newsletter.
I spent several hours refining the prompts, iterating on the outputs, getting the AI to adjust the sequencing, restructure the weeks, improve the summary descriptions. The end result was thorough, well-organized, and covered every topic I wanted to address.
I then published the first seven issues of this newsletter.
You may have noticed that the published issues don't follow the calendar I built very closely. Some of the topics are in different positions. Some of the framing shifted. A few of the pieces I thought I'd write turned out to be different pieces when I sat down to write them.
That divergence — between the AI-built calendar and the actual writing — is where the mistake lives.
What went wrong.
Here's what I should have known, and did know in theory but not in practice: AI is very good at building plausible structures. It can produce an editorial calendar that looks excellent — organized, logical, the topics in sensible order. What it cannot do is know what you'll actually want to say when you sit down to write.
I built the calendar before I'd started writing the newsletter. The calendar was built from my stated intentions, not from the experience of writing. And writing changes things. When you actually sit down to write about a topic, you often discover that the interesting thing is not what you thought it was going to be, or that the piece wants to be a different length, or that a piece you thought would be standalone needs to be split into two.
None of that is knowable in advance. Not by me and certainly not by AI.
So what I ended up with was a detailed calendar that I increasingly felt obligated to follow even when the actual writing was pulling in a different direction. I spent energy defending the calendar to myself rather than paying attention to what the writing wanted to be. That's the mistake: mistaking a good-looking AI output for a constraint rather than a draft.
The fix, and what it cost.
Eventually I let go of the calendar and started writing more responsively — using it as a loose reference rather than a blueprint. That's what it should have been from the start.
What it cost: probably a week of friction and a couple of pieces that I'm less happy with than I'd like to be, because I was writing toward a structure rather than toward an idea.
In the context of a larger operation, that's a small cost. But I'm documenting it because the pattern — using an AI-generated output as a constraint rather than a tool — is one I see everywhere in how people talk about working with AI. The calendar example is small. The same error applied to a product strategy or a business plan or a marketing approach is not small.
The lesson I actually learned.
AI outputs are hypotheses, not decisions. They're excellent for generating options, structures, and drafts to react to. They are not substitutes for the judgment that comes from actually doing the thing.
Build the plan. Then start doing the thing. Then update the plan based on what you learn from doing it. Don't let the quality of the plan become a reason to stop paying attention.
This is obvious when I say it. It was less obvious at eleven o'clock at night looking at a beautifully formatted editorial calendar thinking "finally, the uncertainty is managed."
This Week in AI: One of the more interesting phenomena in the AI productivity space is what some practitioners call "AI-washing" of your own intuitions — using AI outputs as apparent external validation for decisions you've already made internally. It's easy to prompt your way to an AI agreeing with what you were going to do anyway. The solution isn't to distrust AI; it's to deliberately ask it to argue the other side.
Everything I learn the hard way ends up in the toolkit eventually. Grab it at start.tenstreamslab.com.