Thanks to AI, I built more in the last year and a half than in my entire life. And that’s exactly why I shipped so little.
That’s not a typo. It’s the whole problem.
I do AI for a living. I run a YouTube channel and a community about it. I know what I’m supposed to be working on day to day — and I still find myself starting something new instead, or falling down a rabbit hole I didn’t need. For a long time I called that a focus problem.
It isn’t a focus problem at all. AI moved the reward to the front of the work.
And that shift isn’t only filling up my projects folder. It’s filling up the internet with slop, the thin, unfinished, half-shipped stuff that shows up everywhere now. So when people ask me how to avoid AI slop, I think they’re asking the wrong shape of question. They want a prompt trick or a detector. The real answer is older and harder: the work has a front half and a back half, and AI made one of them almost free.
The Front Half and the Back Half
The front half is the part everyone loves. It’s the build, the prototype, the demo, the dopamine hit of I made this. It’s fun, and with AI it’s nearly free. Intelligence is basically free right now. Whether you pay for a big AI subscription or run a smaller model day to day, the cost of spinning up a prototype rounds down to almost nothing. You can build a rough first version in an afternoon. It feels like a superpower, and for many of us it is.
The back half is everything after the creative high fades. The editing, the pitching, the marketing, the part where you ask a stranger to care and pull out their wallet. (That last part is where most of us quit.)
The front half is where you make the thing. The back half is where the thing becomes real, where it could actually earn something, or reach someone, or exist anywhere outside your own laptop.
One thing AI didn’t change: the work that makes the thing real didn’t get any cheaper. The testing, the quality checks, the launch, getting one stranger to care. All of it is still waiting for you after you build the thing. AI front-loaded the dopamine and left the rest exactly where it was.
It Was Never a Focus Problem
Fifteen years ago, before any of this, I tried to write a fantasy novel. I quit my job and wrote full time — six to eight hours a day, about 200 pages by hand. That’s full-time-job focus. When I finished the first draft, I did nothing with it. No pitch. No agent, no publisher. Only my mom and my girlfriend at the time ever read it. I still think it was pretty good.
Then I understood. The focus was never missing. It had gone to the half of the work that felt good. I had spent years telling myself the story everyone tells (ADHD, no discipline, figure it out later). But I had just written 200 pages by hand. Nobody does that without focus.
What I didn’t have was a reason to do the back half, because the reward was already banked. I wrote the last line, got the hit... and stopped. I got the reward and quit, in that order.
Then I spent 15 years doing the same thing in new costumes. Web directories that never turned into revenue. Software products I never sold. Every one of them failed for the same reason: I hated marketing. I hated cold outreach. I just liked building. The distractions were never stealing my writing time, because the writing was the easy half. They were stealing the back half, and I let them.
The Distraction Now Wears a Builder’s Costume
Back then, at least the distractions were obvious. I never confused a conspiracy-theory rabbit hole with productivity. (I did lose hours to Reddit and Facebook. I just knew what I was doing.)
Now it’s different. For us chronic builders, AI makes the distraction feel productive. A new prototype at 11 p.m. feels exactly like work. I’m typing, the code is running, a demo appears. I’ve done this with a new B-roll rig, a new feature, a fresh repo, all of it at 11 p.m., all of it feeling like the job.
But if that’s happening instead of the back half, it’s the same mechanism as doomscrolling on Instagram. The distraction is wearing a builder’s costume. (An AI wrote that line, actually. I’m still not sure I like it.)
What Actually Shipped
Here is the other side of the story, same person. Fifteen years after shelving that novel... I finally learned how to take the back half seriously. It happened while I was building this YouTube channel.
I did the thing I had avoided my whole life. I researched what I was making content about, for the first time ever. I thought about whether people would actually want it. I treated the channel like a project instead of a hobby. And it worked. The channel got attention... and that compounded into the chance to do AI full time.
If I had never decided to conquer the back half, the channel never would have gotten attention to begin with. That isn’t about YouTube. The same person, with the same focus, shipped one thing and not the other, and the only difference was whether I did the unglamorous half.
None of this is about keeping up. As I wrote in falling behind is a myth, the tool layer resets so fast that missing a few news cycles costs you almost nothing. What compounds is judgment, and judgment is the back half. (Yeah, I know. It’s the least fun answer.)
Shipping Is the Human Quality Check
So what does shipping even mean now, when a machine can produce a repo in an afternoon? Publishing isn’t the same as shipping.
I think shipping is the quality check. It’s the human judgment that goes into the work. It’s the moment you decide the thing is good enough to put your name on it.
Look at what’s filling up GitHub’s trending page: thin tools that wrap a couple of API calls and call it a product. A first draft wearing a finished label. Every prototype is a first draft that feels finished, and if you never do the back half, you end up with a folder full of front halves.
What makes a thing real is the system you build around the model, what we call the harness. The model gets swapped every few months. The judgment, and the system, are yours.
Two Moves That Fix It
Willpower isn’t the answer. I tried willpower for 15 years. The fix is sequencing.
First: research before you start. Ask whether the thing is even worth starting. How far will you actually take it, and what does it give you if it works? If the honest answer is I just want the demo and the dopamine, don’t build it. I say that as a person who has built a lot of demos.
Second: do the back half before the front half. Figure out the pitch, the launch, the distribution, the ask before you write the final draft. When you line up the hard part first, the reward stays at the end where it belongs, and finishing gets easier. The goal is to have fewer projects, and finish them.
Did AI Assist You or Replace You?
Here is the test I keep coming back to. Ask the sovereignty question: did AI assist me, or did it replace me? If you can’t point to the work you did yourself inside the process, then the thing isn’t yours yet.
That’s the whole difference between using AI as leverage and using it as a stand-in. You want to be the strategist at the helm, not the person who pressed go and hoped. I’m not certain the back half will always be the hard part... maybe the models get good enough to run the QA and the launch too. But every time I’ve checked, the last mile was still human.
The stack matters less than the habit. If you want the tools I actually run, I wrote up the exact stack I use after cutting my bill from $200 to $20. But no tool will do the back half for you. A crew that keeps you honest might. That’s part of why I run a paid community for builders — because the fastest way to ship is to have people who will ask you, every week, what you actually finished.
Don’t let AI turn your best idea into another unfinished novel. Do the back half.
If You Want People Doing Their Back Half
This essay is a belief in the AI Captains Manifesto. If you read it nodding, you’re already one of us.
If you want people around you doing their back half, that’s what AI Captains Academy is for. We do two live calls a week — workshops, Q&A, and fellow builders showing up. Silver is $83/month or $497/year with a 7-day free trial. No countdown timer on that sentence. There never will be.
Questions people actually ask
- What is AI slop?
- AI slop is the thin, unfinished work that gets published because building got easy and shipping didn’t. It looks finished: a repo, a landing page, an app. But nobody did the back half: the testing, the editing, the launch, and the part where a stranger decides it’s worth their time or money. The internet is filling up with it.
- How do I avoid AI slop when I build with AI?
- Do the back half, and do it first. Research whether the thing is worth starting before you build it. Line up the pitch, the launch, and the distribution before the final draft. Then pass one test: can you point to the work you did yourself? If the honest answer is no, the thing isn’t yours yet, and it isn’t ready to ship.
- Is vibe coding bad?
- No, throwing together quick prototypes is how you find out what’s possible. The trouble starts when every project stays a prototype. AI made the building cheap, so the temptation is to bank the dopamine hit and move straight to the next build. Vibe coding is fine as a method. It goes wrong when it replaces the work that makes the thing real.
- Why do I keep starting projects and never finishing them?
- Because AI moved the reward to the front of the work. The dopamine hit used to arrive at the end: the finished draft, the published video. Now it lands in an afternoon. So you bank the reward and stop, before the back half. The fix isn’t more discipline. It’s sequencing: keep the reward at the end by doing the hard part first.
- How do I know if AI did the work or I did?
- Ask the sovereignty question: did AI assist me, or did it replace me? If you can point to your own contribution inside the process: the judgment, the edits, the decisions, the quality check. Then you own the output. If you can’t, the thing isn’t yours yet. That’s the test I run before I put my name on anything.
