It started, as most great ideas do, at work.

Someone asked the group how many 5th graders they thought they could beat in a fight. You know the question. Everyone has a number ready. Nobody’s embarrassed about it. The room got loud.

But then someone changed the scenario: okay, but what about tug of war?

And that’s actually a completely different problem. Hand-to-hand combat against a pack of 10-year-olds is a question of reach, reflexes, and how much you’re willing to commit to the bit. Tug of war is physics. It’s weight, friction, grip strength, and coordinated pulling force distributed across a rope. You’re not fighting the kids — you’re fighting math.

So naturally, I tried to research it.

The Research Phase (I Lasted About 20 Minutes)

I poked around looking for actual data on how much pulling force a 5th grader generates in a tug of war scenario. Turns out there isn’t a ton of peer-reviewed literature on this specific topic, which feels like a glaring gap in the academic community.

I found some general grip strength data. Some pediatric exercise physiology stuff. Nothing that translated cleanly into “here’s how many kids it takes to drag a 190-pound adult across a line.”

So I did what any reasonable person does when real research gets tedious: I opened Gemini and just asked it.

And here’s the thing — it actually worked. Gemini pulled together estimates on average grip strength by age, made reasonable assumptions about body weight and center of gravity, and handed me a working formula with actual numbers. Real enough to build something. Real enough to settle arguments.

I could’ve written it in a notes app and moved on with my life.

I did not do that.

From Math Problem to Website in One Conversation

Once Gemini gave me the numbers, I had it write the website.

That’s the whole build. One tool, one conversation. I asked it to code me a single-page calculator where you can plug in your own stats on one side and build a theoretical team of kids on the other side, and it spits out who wins on paper.

No frameworks. No database. No backend. Just a clean HTML page with the math baked in, doing exactly one thing.

The whole build took less time than the original workplace argument that inspired it.

I threw it up at tugofwarcalculator.com.

What the Tool Actually Does

You go to the site, enter your weight and general athletic disposition, and then configure a team of elementary schoolers to pull against you. The calculator estimates the combined pulling force of your theoretical kid-team against your own numbers and tells you how the matchup shakes out.

Is the math perfect? No. Is this medically, physically, or ethically sound advice? Also no. But is it directionally correct enough to end a workplace debate? That was the whole goal, and I think yes.

There’s also something kind of funny about the fact that a real academic paper I found on tug of war physics notes that on most surfaces, the sport essentially comes down to who’s heavier — not who’s stronger — because friction is what actually keeps you from sliding. So your grip strength barely matters. Your weight does. That’s either comforting or devastating depending on where you land on the scale.

What I Actually Learned

AI as a research shortcut is underrated. I wasn’t asking Gemini to make something up — I was asking it to synthesize publicly available data on pediatric exercise physiology into a usable estimate. That’s genuinely useful. That used to require either a science background or a lot of time I don’t have.

Scope is a superpower. The calculator does one thing. It doesn’t track your results over time or let you create an account or share your matchups or any of the other feature creep that turns a two-hour project into a two-month one. One page, one question, one answer.

The best tools are the ones people already want. Nobody asked me to build this. But the second I told people it existed, everyone went to try it. The question “how many kids could you beat in tug of war” is already living rent-free in people’s heads — I just gave it a home.

The Verdict

Built: Yes.

Time: A couple hours, including the original research spiral.

Abandoned: Technically. But it’s still up and it still works, so it’s more of a “parked” situation.

Would I do it again: Definitely.

The calculator lives at tugofwarcalculator.com. Go find out how many kids it takes to drag you across a line. The answer is probably fewer than you’d like.

-CT

Read more at Build and Bail

It started the way most of my projects start: with a question nobody asked. How many hamsters running in wheels would it take to power my website?

I typed it into Claude, fully expecting a one-line joke. Instead, I got physics. A hamster generates about 0.5 watts on a wheel. They only run about 6 hours a day. A basic web server draws around 400 watts. The math landed on 3,200 hamsters working in four rotating shifts, each group tricked into a different circadian cycle with controlled lighting.

Claude didn’t just give me a number. It gave me a shift schedule, a facility layout with color-coded rooms, a rest day rotation, and a job title: Hamster Ops Manager.

The math: 400W ÷ 0.5W per hamster = 800 running at all times. At 6 hours per hamster per day, you need 4 shifts × 800 = 3,200. Add a rest rotation and buffer, and you’re looking at about 4,500 hamsters total.

At that point I knew this had to be a video.

From Chat to Script

I asked Claude to write me a production script for a 10–15 second video. It came back with a full cinematic breakdown: warehouse layout, props list, shot-by-shot timing, crew roles, even three budget tiers. Way too detailed for what I needed.

I told it I was using an AI video tool and to simplify. It immediately stripped everything down to four scene prompts with paired visual and audio direction, each one ready to paste into a generator. That’s the version I used.

The Workflow

1. Ask the dumb question — Claude did the math, designed the shift system, and mapped out the whole facility.
2. Generate the script — Asked Claude for a video script, iterated once to simplify it for AI generation, and added audio cues for Flow.
3. Paste into Google Flow — Dropped each scene prompt into Flow / Gemini. It handled visuals and audio together.
4. Add text overlays — Quick pass to add the three text cards: hamsters per shift, watts, and the final tagline.

The Tools

• Claude (Anthropic) — math, script, prompts
• Google Flow / Gemini — AI video generation

That’s it. Two tools. The entire process from dumb question to finished video happened in one sitting. No footage was shot. No stock video was licensed. No hamsters were harmed or employed.

Watch The Process

Watch The Result

What I Learned

The interesting part wasn’t the hamsters. It was the feedback loop. I started with a throwaway question, got a surprisingly detailed answer, and realized the answer itself was content. From there, Claude adapted its output three times: first a full production script, then a simplified AI-ready version, then audio cues for a specific platform. Each iteration took one sentence of direction.

The gap between “idea” and “finished thing” is collapsing. Not because the tools are perfect, but because the iteration cost is nearly zero. A bad prompt costs you ten seconds. A good one gets you a video.

Will AI-generated video replace real filmmaking? No. But for a weird idea that lives in the space between shitpost and science, it’s exactly the right tool.

3,200 hamsters. 4 shifts. 1 website.

Keep chasing your silly ideas.

-CT

Read more at Build and Bail