Agentic reversing vs Vibe reversing

[chants]

So I do not do vibe reversing as it were but agentic reversing. From whst I see only those who have been reversing since 2022 or before are agentic reversers. It is essy to look at an accounts join date and simply know. Because the LLM era came circa 2023 and made it harder for newcomers to be motivated to gain deep knowledge

Agentic reversing is a highly structured, objective, and often automated approach to reverse engineering. The term "agentic" highlights the goal-oriented, systematic nature of the process, which is increasingly assisted by AI or LLM agents executing deterministic tasks. Thorough but time consuming.

Vibe reversing is an informal, black-box approach that relies on pattern recognition, intuition, and trial-and-error. Instead of looking at the code, you look at the "vibes"—the external behaviors, error messages, timing discrepancies, and common developer habits. Fast but imprecise.

So newcomers may start claiming big results like having defeated VMP but this is a testament to the strength of LLMs and advancements in AI. The incredibly shallow depth of understanding does make it quite different. The OGs have the clear solid level of understanding. This same thing is going on in mathematics where understanding and explanation can be valued beyond mere formally verified proofs or raw results. It poses an interesting challenge. The script kiddies of reversing are upon us.

[chants]

Just to substantiate with what the software engineering community has gone through in rebranding here is even a research paper:
Vibe Coding vs. Agentic Coding: Fundamentals and Practical Implications of Agentic AI
https://arxiv.org/abs/2505.19443

And an article:
Why Agentic Engineering Must Replace Vibe Coding
https://dev.to/shrsv/why-agentic-engineering-must-replace-vibe-coding-339f

There are a lot of articles on vibe coding vs agentic engineering and reversing is no different ultimately.

[Shub-Nigurrath]

This is one of the most interesting evolutions lately.

My experience: with strong enough context, intent, and specification engineering, and good prompt crafting, you can most of the time completely skip manual revision of the generated code.

For those who doesn't know these are the terms I refer to..

Prompt Craft. Writing effective requests.

Context Engineering. Preparing useful context: sources, examples, constraints, definitions.

Intent Engineering. Clarifying objectives and priorities: what matters, what is out of scope, what defines "success".

Specification Engineering. Translating the goal into a verifiable specification: requirements, acceptance criteria, edge cases, quality standards.

These same criteria are good for efficient "vibe-reversing" or, better, AI-assisted RCE

best
Shub

[dyers eve]

Originally Posted by Shub-Nigurrath

This is one of the most interesting evolutions lately.

My experience: with strong enough context, intent, and specification engineering, and good prompt crafting, you can most of the time completely skip manual revision of the generated code.

You are right!

An interesting article:
Agentic Software: How AI Agents Are Restructuring the Software Paradigm
https://arxiv.org/abs/2606.05608

[blue_devil]

Originally Posted by Shub-Nigurrath

This is one of the most interesting evolutions lately.

My experience: with strong enough context, intent, and specification engineering, and good prompt crafting, you can most of the time completely skip manual revision of the generated code.

For those who doesn't know these are the terms I refer to..

Prompt Craft. Writing effective requests.

Context Engineering. Preparing useful context: sources, examples, constraints, definitions.

Intent Engineering. Clarifying objectives and priorities: what matters, what is out of scope, what defines "success".

Specification Engineering. Translating the goal into a verifiable specification: requirements, acceptance criteria, edge cases, quality standards.

These same criteria are good for efficient "vibe-reversing" or, better, AI-assisted RCE

best
Shub

This post is sooo good that; it can be a 3 hour long defcon presentation or 5 days long hands on training. 🤌

[Shub-Nigurrath]

Behind professional prompting lies an entirely new way to programming computers. Deterministic skills, loops, and graphs are among the easy things, but you can do really complex things. I plan to create some RCE skills for demonstration purposes and share them here.

Most of the papers here discuss how skills are reshaping normal programming, which is a real and concrete thing, but there is also the other half of the story, which is how to professionally program skills. It's a new paradigm, and it's not only about prompting or writing md files. Inside a skill, there could also be a lot of code.