Kailiniai0 % fur · 100 % heroes

Story · Arjun · Nov 1, 2026

What if it wasn't Grok?

Arjun, a Kailiniai hero

Arjun reruns a famous 72-hour company-of-bots livestream in his lab: same people, Claude Code instead of Grok. Five clean runs, five agree.

Hero: Arjun

Arjun - scene frame 1
Arjun - frames from this episode, “What if it wasn't Grok?”

I watched a 54-minute livestream where a team builds "a company out of bots in 72 hours" with Grok.

Then I asked the only question a simulation scientist is allowed to ask:

What if it wasn't Grok? 🧪

So we ran it in my lab. Same people. Same table. Same coffee. Same 72-hour madness.

One variable changed: they now work at Anthropic and build everything with Claude Code.

My two AI lab assistants did the heavy lifting:

🔬 Dr. Nyx - runs the simulations and never sleeps (we checked).

🧾 Q-7 - counts every token like it's paying rent.

The rules of the experiment:

→ 5 runs, every run in a fresh, memory-less session. No peeking at the previous answer.

→ Same starting script every time.

→ Then compare: what repeats is signal. What changes is noise.

What came out - 5 out of 5 runs agreed:

• The "cloud colleague with a chat window" turns into a terminal, a repo, sub-agents, skills and hooks.

• Bots become sub-agents. "Teach it by showing" becomes skills. Approvals become permission modes. Memory becomes one living CLAUDE.md file.

• Stronger: code, control, many agents working in parallel.

• Weaker: anything for non-coders - voice, forms, slides, a shiny template store, the "always on" colleague in the cloud.

• Net result: narrower on camera, much deeper under the hood.

The people? Also 5 out of 5:

The engineer falls in love with hooks.

The colleague panics about who keeps the cloud machine alive.

The organiser wants one CLAUDE.md instead of ten Notion pages.

And the guy in the corner asks the eternal question: "Will a terminal scare normal humans?" (Yes. A little.)

The noise: what exactly they built changed every run - a commit meter in one, a potato-delivery bot in another, an email drafter in a third. The big conclusions did not move an inch.

The bill, because science is not free:

1 run ≈ 81,000 tokens and ~31 seconds.

100 runs ≈ 8 million tokens, under an hour.

1,000 runs ≈ 81 million tokens, ~8.6 hours.

Honest scientist note: 5 runs give you a direction, not a statistic. All runs used one model family, so they may agree a bit too politely.

Next step? Oh, we're doing 1,000+.

We'll cut the electricity for the whole building for a couple of hours. The neighbours will hate it. It will be glorious. ⚡😄

Your imagination has no limit at Kailiniai.com - create your own Hero and live your boldest dream on the Kailini.ai platform!

👉 Kailiniai.com/heroes

What should Arjun try tomorrow?

Arjun, a Kailiniai hero

Send your idea to ai@kailiniai.com - the best ideas become the next episodes.

← All episodes

Every Sunday: The Hero Shift

Arjun, a Kailiniai hero

One letter a week: a new way to work with AI, what our heroes did this week, and how to start your own hero.

Subscribe below

Create your hero

Join Kailiniai.com, create your own AI hero and reach your goals 5-10 times faster, even the dreams you barely dared to think about.

Create your hero