I Built 61 AI Apps. The Usage Data Changed My Mind About What AI Is Good For.
Listen
AI is most useful as a chief of staff, not a coach. I know this because I built 61 apps for myself with Claude Code, Vercel, and Supabase, then looked at which ones I actually use. The apps that produce an artifact (a report, a scorecard, a recommendation) get used every week. The apps that produce advice are dead. I started this thinking AI would help me think better. The data says it helps me execute better, and those are not the same thing.
Here is the portfolio as of today:
- 61 apps built
- 55 active in the last 60 days
- 48,100 events in 60 days
- 300,000 lines of code
- $31,160 CAD ($22,744 USD) of Claude usage
- ~79 million output tokens
That is large enough to stop trusting my opinion and start reading my behavior.
Which apps died, and what did they have in common?
The ones I expected to win. Career Coach. Influence Coach. Communication Training. Mental Models. Principles. Attention Rules. Time Allocation. Some I opened once. Most I haven't touched in weeks.
The pattern wasn't quality. It was structure. Every dead app shared three traits:
| Trait | What it looked like |
|---|---|
| I was the only source of data | The app did nothing until I showed up and fed it |
| The output was advice | "Here's a framework," "you should think about this" |
| Nothing pulled me back | No meeting, no report, no deadline, no new data, no consequence |
These apps waited for me to care. Motivation is an unreliable input, so they starved.
What won instead?
The most-used apps look nothing like the coaches. They produce work, not guidance.
| App | What it produces |
|---|---|
| GH Product Evaluation | A go/no-go on a product or opportunity |
| GH Ecomm Scorecard | A weekly operating review |
| Job Search | A tracked, scored pipeline of openings |
| TikTok Scorecard | Performance monitoring |
| Inbox Zero | Triaged inbox and drafted replies |
| What To Do | Captured tasks turned into next actions |
| Amazon Scorecard | Business monitoring |
| Vendor Credibility | A decision on who to trust |
| Articulation | Rough thinking turned into clean communication |
A coach says "you should do this." A chief of staff says "I prepared this. Review it." One creates motivation. The other creates leverage. My usage overwhelmingly favors leverage.
Isn't the real driver just that the winners are recurring?
Close, but not quite. The winners aren't recurring because I scheduled them. They're recurring because something in the world triggers them.
The Ecommerce Scorecard runs because performance data arrives every week. Product Evaluation runs because new products appear. Job Search runs because openings enter the pipeline. Amazon Scorecard runs because Amazon keeps generating numbers. Reality forces the workflow. The app is attached to something happening whether I think about it or not. The coaching apps were attached to nothing, so they only ran when I remembered to feel motivated.
What about memory? Wasn't that supposed to be the big unlock?
Partly. But only in one form. Standalone memory failed. Operational memory won.
A database of mental models was useless. A database of job opportunities was not. A repository of principles went stale. A repository of research dossiers got used constantly. The lesson: memory works best as a byproduct of doing the work, not as the work itself. The best systems get smarter because something useful happened, not because I did data entry.
Does this mean self-improvement doesn't matter?
No. It means self-improvement apps are usually built wrong. My actual growth loop is: experience, then reflection, then a written memo, then better decisions next time. A hard conversation, an interview, a marketing problem, a career call. Then I write about it. The wisdom comes from processing real events. The coach apps tried to manufacture wisdom in advance. That isn't where it comes from.
The four questions I ask before building anything now
- What artifact exists after I use this? Good answers: memo, report, deck, recommendation, forecast, brief. Bad answers: clarity, awareness, motivation.
- What external event triggers it? If the trigger is "I'll remember to use it," it's already dead.
- What breaks if I delete it? If nothing breaks, that's a warning sign.
- Does memory accumulate on its own? The best systems get smarter from work, not from upkeep.
The winners in my portfolio behave like analysts, researchers, and chiefs of staff. The losers behave like productivity gurus and self-help books. I started out believing AI would improve my intentions. It turns out AI is far better at improving my ability to execute.
Frequently asked questions
Why did the coaching and self-improvement apps fail? They depended on me as the only source of data and produced advice instead of work. Nothing external pulled me back, so they ran only when I felt motivated. Motivation is unreliable, so they died.
What makes an AI app actually get used? An external trigger and a useful artifact. If reality forces the workflow (new data, a deadline, a recurring review) and the output is something you can act on or hand off, it survives. If it waits for you to care, it won't.
Does this mean AI is bad for personal growth? No. It means growth comes from reflecting on real experience and writing it down, not from an app that hands you frameworks. AI helps by turning that reflection into a durable artifact, not by manufacturing wisdom up front.