I have been suffering greatly lately while prompt-coding. I find that the cause is consistently breaking one of a handful of what ought to be commandments.
I thought a religion of sorts would be beneficial for both man and agent to reduce suffering and distill meaning.
Zen coding
1 In the beginning man created vibe coding. This was the beginning of suffering.
2 Many men perished waiting for their agent to finish. This is the purgatory of AI coding.
3 The root of all suffering is under-specification. If you under-specify you will burn.
4 God said let there be light and continued specifying until the seventh day. He surely did not ask light to figure out the rest. You must over-specify. Anything not specified is created randomly.
5 Transforming a small prompt or specification into a larger one via AI is the only unforgivable sin. All other sins may be forgiven.
6 You shall not test an agent’s changes manually. Agents must be able to test their own work in a continuous feedback loop until task completion.
7 You shall not “chat” with an agent. Speak only in divine specification.
8 You shall not follow up with an agent. When an agent requires followup, both the prompter and the agent shall be put to death.
9 Unreviewed code is the work of the evil one. Agent code must be baptized by the watery eyes of man before it can be admitted into repository.
10 You shall maintain a GENESIS.md document describing the ongoing nature and specification of your application. This gives your agent belief, direction, and understanding in something beyond mere lines of code.