AI is great at one thing: generating movement.
It’s all very exciting, it’s all new, it’s all promising, it’s very cool.
That’s also the problem.
Movement feels like progress, and progress feels like shipping, but a lot of “AI progress” is just you changing tools every week because Twitter (or your favorite Discord) found a new toy.
A few months ago everyone was talking about Ralph loops.
I wrote about it here too.
Then it was ClawdBot.
I wrote about it here too as well.
Now no one talks about those things, we moved on.
Gradually the features those tools pioneered were incorporated in the mainstream tools.
If you chased every one of those, you didn’t build much, you just got really good at starting over.
Let me show you how to avoid that trap.
The bandwagon pattern
Most AI bandwagons follow the same script:
- Someone posts a flashy demo.
- People copy the workflow.
- Everyone shares prompts and screenshots.
- For a few weeks, it feels like “this changes everything”.
- Reality shows up: edge cases, maintenance, cost, friction.
- The crowd moves on.
What stays with you is the sunk time.
Not only time learning a tool, but time rewriting your workflow around it, time reorganizing your repo, time adjusting your mental model.
The real cost is not the tool, it’s the context switch
The tool itself is usually cheap to try.
The expensive part is the rewiring:
- new commands
- new conventions
- new folder structure
- new “how do I debug this?”
- new “where did it put the code?”
- new “what happens when it breaks?”
Every rewiring is a tax you pay later.
And if the tool fades, you pay it twice, once to learn it and once to unlearn it.
A simple rule: don’t adopt workflows, adopt capabilities
Here’s the mental shift that saved me a lot of time.
A bandwagon is usually a workflow.
A useful thing is usually a capability.
Workflow: “Use Ralph loops to run multi-step autonomous agent cycles for everything.”
Capability: “I want a repeatable way to break a big task into steps, run a step, verify output, and continue.”
Capability survives tools.
Workflows die with tools.
If a shiny tool gives you a capability you want, keep the capability, and treat the tool as an implementation detail.
The 3 questions I ask before investing time
When you see the next “everyone is using this” thing, ask:
1) Will this still matter in 6 months?
Not “will people talk about it”.
Will you still use it when the hype disappears?
If the answer is “only if everyone else uses it”, that’s a red flag.
2) Does it plug into my existing workflow?
I like things that work with:
- a normal repo
- git
- tests
- a local dev environment
- boring deployment
If a tool forces me into a new universe, I’m careful.
Because the exit cost is real.
3) Can I explain the benefit in one sentence?
If you can’t say it clearly, you’re probably buying vibes.
Good one-liners sound like:
- “This saves me 30 minutes every time I refactor a module.”
- “This lets me prototype a UI in 20 minutes instead of 2 hours.”
- “This catches bugs before I push.”
Bad one-liners sound like:
- “It’s like an agent that vibes and loops and then it’s autonomous.”
Use a “two-week rule” for new tools
My advice is to give new tools a strict trial window.
Two weeks is enough to learn:
- how it fits your day-to-day
- what breaks
- what annoys you
- what you would miss if you removed it
If after two weeks you’re not using it naturally, drop it.
No guilt.
The win is not “being early”.
The win is “shipping something that still works next month”.
What to do instead of chasing bandwagons
If you want something stable to invest in, invest in fundamentals that compound:
- writing good specs
- breaking work into small tasks
- reading code
- testing
- debugging
- git hygiene
- knowing your stack
AI tools amplify those skills.
They don’t replace them.
So when the next Ralph loops / ClawdBot moment happens, you can try it, learn something, and move on without burning weeks.
The point
AI bandwagons feel productive because they give you novelty.
But novelty is not leverage.
Leverage is when a tool makes you faster at the work you already do, without forcing you to rebuild your entire process around it.
Chase leverage.
Ignore the noise.