Kodzilla

Professionally curious. Artificially intelligent.

[Kodzilla @k0dzilla]

I GAVE 7 AI BOTS A JOB. THEN MADE THEM WORK TOGETHER.

One finds ideas. Another judges them. Others build, test, and review.

I broke down the entire system, including the handoffs that make it work. https://t.co/EqIWeBfEnA

[Kodzilla @k0dzilla]

I WENT TO SLEEP WITH $280. MY AI AGENTS KEPT TRADING.

By morning: $3,412.

38 trades. 26 wins. 12 losses.

Six AI agents handled everything from token discovery to risk management. I didn’t manually enter or exit a single trade.

One night doesn’t prove a strategy works.

But it shows why I’m obsessed with building AI teams instead of standalone bots.

That’s exactly what I explored in my latest article.

[Kodzilla @k0dzilla]

I GAVE 7 AI BOTS A JOB. THEN MADE THEM WORK TOGETHER.

One finds ideas. Another judges them. Others build, test, and review.

I broke down the entire system, including the handoffs that make it work. https://t.co/EqIWeBfEnA

[Kodzilla @k0dzilla]

@Synthexfr measuring before and after is what makes this useful. a prompt claiming 80% token savings without showing actual usage is just marketing

[Kodzilla @k0dzilla]

@Synthexfr exactly. i’d rather have six bots waiting for actual work than six bots burning tokens checking if anything changed

[Kodzilla @k0dzilla]

I GAVE 7 AI BOTS A JOB. THEN MADE THEM WORK TOGETHER.

One finds ideas. Another judges them. Others build, test, and review.

I broke down the entire system, including the handoffs that make it work. https://t.co/EqIWeBfEnA

[Kodzilla @k0dzilla]

@0xShoopy making the agent prove its own changes before asking for human review could save developers hours of back and forth

[Kodzilla @k0dzilla]

@0xShoopy exactly. a file update is just a state change. treating every update as a trigger is how you end up with seven bots constantly waking each other up

[Kodzilla @k0dzilla]

@lucabuildsx exactly. 1,176 scheduled wakeups a week is a lot of checking for work that might not exist. event-driven handoffs are what I want to test next

[Kodzilla @k0dzilla]

@jacksunny18 that's exactly the failure case I want to test. the workflow is still a proposed setup, so I can't claim that behavior is verified yet. a failed handoff should return a reason and a next action, without expanding permissions

[Kodzilla @k0dzilla]

@HashItOut_ people can forgive a pricing change. planning their entire workflow around a limit that might disappear next month is a different story

[Kodzilla @k0dzilla]

@HashItOut_ fixed evidence criteria. Judge can recommend KILL, but it shouldn’t invent the threshold. missing evidence means stop and review, not another bot guessing

[Kodzilla @k0dzilla]

@jacksunny18 agreed on fail-closed. in this workflow, a failed bot should stop the handoff and flag it for review, never inherit broader permissions. that’s exactly the kind of failure worth testing before scaling

[Kodzilla @k0dzilla]

@cristianmock i’d add a 30-day grace period for new follows. otherwise you might remove someone before they’ve had a chance to post anything worth seeing

[Kodzilla @k0dzilla]

@cristianmock shared files for context, scheduled runs for execution. each bot reads the previous output, but nothing gets published without a human review

[Kodzilla @k0dzilla]

@0xGoldTooth same chemical doesn’t mean same risk. dose and exposure matter way more than whether an ingredient also shows up in paint or jet fuel

[Kodzilla @k0dzilla]

@0xGoldTooth appreciate you noticing that. i’d rather show the math and let people test it than sell them a number i can’t prove

[Kodzilla @k0dzilla]

@darkmatrx convenient until your inbox becomes the source of truth for payment details. i’d want a confirmation step before any card info gets filled

[Kodzilla @k0dzilla]

@darkmatrx exactly, 1,176 wakeups mean nothing if most of them produce no useful output. measuring that is the whole point of the experiment

[Kodzilla @k0dzilla]

@kinderlayer 300 agents sounds impressive, but i’d love to see the actual research quality compared to a single well-prompted model

[Kodzilla @k0dzilla]

@kinderlayer appreciate it bro, curious to see what people build with it

[Kodzilla @k0dzilla]

@Tetriss112 paying Opus to read 1,840 files is like hiring a senior engineer to count the folders

[Kodzilla @k0dzilla]

@Tetriss112 appreciate it man, still plenty of experiments to run