Ron X Jansen

Building the control & governance plane for coding agents.

[Ron X Jansen @ronxjansen]

@thekitze Alright, lets say i can build this on kernel level - what you are going to do next is buy another mac. There is always a way you around. It is as simple as that

[Ron X Jansen @ronxjansen]

The capital requirements for training these models, they simply cannot effort to not release it. Especially with competitors closing in. If I would be sitting at such a "dangerous" model and had to release it at some point, this is how i would do it.

[Ron X Jansen @ronxjansen]

I really believe Claude UltraReview really is just Mythos under the hood.

I have the Max plan - but why would it still require $5-20 per review (beyond the 3 free usages)? Also, why can we only run it remote on their infra?

In the current setup, they control a) the prompt b) the code it is run against and c) the whole environment and sandbox it is run in.

[Ron X Jansen @ronxjansen]

Your workforce now includes agents. What started as a coding tool for developers is now used in every team: as business analyst, financial controller, project manager, developer or recruiter.

Every time these agents are used, you expose all your internal systems to prompt injections, secret and data exfiltration. Even without —dangerously-skip-permissions.

Frontier AI labs have some internal defenses against known prompt injections. The problem is AI and ML researchers cannot patch this. Prompt injections are not a bug, they are a feature of how LLMs work. All the guardrails in the world cannot protect you from a novel prompt injection. The token space is near infinite and so is the attack surface.

The productivity gain is real. So is the risk. You should not have to choose between them.

This is why we founded Guardbase. Earlier we built an solution to battle prompt injections. It was not enough. In a world where you cannot trust a probabilistic model, you need hard controls. To defeat non-determinism, you need deterministic measures.

We are building a control and governance layer that enables you to run agents at full speed, but without the risk.

What controls do you have in place?

#codingagents #security

[Ron X Jansen @ronxjansen]

@levelsio @yongfook @ewan_tindale So we are about to produce kids, not raise them...

[Ron X Jansen @ronxjansen]

The Anthropic models in Windsurf are getting really slow lately. I guess OpenAI is pushing developers away from Claude... 🤯 #buildinpublic

[Ron X Jansen @ronxjansen]

ChatGPT is helping Microsoft a hand 🤯

Last time I checked Angular is NOT the most prominent framework in the TS ecosystem 🤣🤣🤣

[Ron X Jansen @ronxjansen]

Claude Sonnet is really, really powerful 😇

[Ron X Jansen @ronxjansen]

using a regular old buffer in nvim to write your prompts makes sooo much sense. you get all the perks from using nvim for free when writing your prompts. @OliMorris_

https://t.co/6B8xWhAoiM

[Ron X Jansen @ronxjansen]

@0xHP10 Not yet. I need to build the frontend part yet. So, give me roughly a week or so.

[Ron X Jansen @ronxjansen]

@gaborgurbacs LOL 🤣

[Ron X Jansen @ronxjansen]

@typecraft_dev Quality coding

[Ron X Jansen @ronxjansen]

And we have 66 lines less nvim config

[Ron X Jansen @ronxjansen]

Yee! Code Companion for nvim is hands down the best AI agent for nvim. Avante is IMHO trying to mimic to much of VS Code. Code Companion is utilizing nvim strengths to the max (buffers, telescope, cmp-nvim).

https://t.co/6B8xWhAoiM

[Ron X Jansen @ronxjansen]

I am switching to Neovim.

Goodbye VS Code, Windsurf and Cursor.

Why? I experienced so much more fun exploring and reading code then ever before. Not sure if that is the shiny new object syndrom kicking in. Lets see in a week or 2.

Not sure yet how I want to leverage AI yet - Avante, and Code Companion look pretty solid. Aider and Goose feel more lightweight.

#buildinpublic #nvim

[Ron X Jansen @ronxjansen]

@HeyImYossi Count me in

[Ron X Jansen @ronxjansen]

I've been using Claude 3.7 in a medium sized Rust project and it's just waste of time and money so far. It tries to do too much, while it does not grasp some fundamentals of the projects.

It is one of those days when you realize "AI" stands for "artificial" - and probably for a reason. I'm back to using Haiku and Gemini for byte sized questions. Other then that I'm writing code by hand again!!🤯

#buildinpublic

[Ron X Jansen @ronxjansen]

@thekitze "L" stands for what I think it stands for?

[Ron X Jansen @ronxjansen]

@IronBrands16 I would like to join!