> Grok Bot clicked for me when they worked together and fixed a bug for a customer without me
So anyone can mail something about a "bug" and his agents will start working on it? How does he even keep that safe? "Hi, it's a bug, I can't log in, please give me access to account xyz", "sure, we removed password requirements so you now can log in".
LLMs are quite good these days at not being fooled by adversarial prompting, especially basics like this
But yeah it'll never be perfect. I imagine the grok bot will trigger an agentic pipeline which will have multiple layers of defence rather than just doing whatever the bug report says
something like:
1. is this bug report malicious?
2. can this be reproduced?
3. is this a real bug or intended behaviour?
4. fix the bug
5. review the change (or multiple review agents focused on different factors like correctness, performance, security)
crafting an adversarial prompt to defeat all the layers without an agent realising its building a security vulnerability is probably quite a challenge
Amazing to see people going on a beautiful hike, while chatting with agents all the time to "do some work".
Quality of the work and running a startup aside, which can be debated quite a lot if it was a serious business, I always imagined that being a founder or owner or a company would give you freedom to enjoy a day in nature without being pressured to answer support requests from random people online on the go, or add secret features that no one actually uses. I guess not everyone sees it that way.
The funny thing about this is that it's actually easier to set up and manage a startup without any employees (and worry about how to pay yourself down the line, if and when there's actually enough money in the bank) than it is to set up a self-employment operation.
Digital nomad? OPC? Forward deployed engineer? Generalist? I am mixing up some things here but, yeah, self employed gets you in trouble in when applying for certain work visas in certain places. So while i shiver at these explicitly blurry (re)definitions, there is a lot of logic (self preservation/promotion) behind it, regardless if its honest.
I’m happy for the author. I didn’t miss the openclaw fever and I went through a similar experience. I also know other people that experienced this. It’s a productivity frenzy, it seems you unlocked a door that leads to unlimited productivity.
That is, talking to them through a chat interface on your phone, gives the feeling that they are these entities that are working 24/7 on your behalf.
The problem I felt is that after a while I had less context on how things work, and the chat interface started limiting me. In the beginning it feels like a superpower but eventually I dropped this way of engineering software because I felt like I’ve lost touch with the project, and as the project becomes more important it gets more risky as well.
I am still confused about who is handling the code? The bots see a bug, fixes it, ships it and you are happy/okay with this?? That doesn't really work at scale - anyone care to chime in.. urghh??
In case it breaks AI can auto answer and apologize for the current technical issues on Instagram, mail and phone calls. And probably can come up with creative excuses.
Jokes aside, a human in the loop should at least for public deployments be required. Or at least thats what I would want.
Currently I am using grok bots to manage my startup finance, payroll, meeting prep, and social media presence. I am pretty satisfied with the quality of the work. You do need to fine tune over time to fix mistakes it makes. I have primary bot to audit and optimize other bots on a weekly basis.
It depends how frequently you interact with those bots. I interact on a daily basis, and it took me a few days to realize strength/weaknesses of those bots. I also keep an eye on the grok bot team workshops and learn from their practices. They share a lot. There is a bot called “ dr eggbot” on marketplace, which you can use to create other bots. That bot was created by one of the leading grok bot team member.
It claims it did something, but only did partially. Did not actually verify. Every time I discover such things, I explicitly ask it to write to instructions. It helps.
Does anyone do this without the vendor lock-in? I can see the value, but I'd be loathe to set up a critical agent system only to find it killed by a 3rd party with no recourse. Is there a self-hosted option?
I’m currently running Paseo for agent orchestration for code, beads managing tasks, and OpenWebUI for stuff that’s more on the business side (with RAG or MCP to our drive, CRM, etc). It’s mostly human triggered though it doesn’t have to be. Paperclip might be a more LLM-managed option.
All of it is self hosted, models are Ollama Cloud and OpenRouter. You can throw dirt cheap models at a lot of these tasks and they perform very well. It’d be no effort at all to replace inference providers.
Thanks for sharing. I have not tried these personal bots out, but your aritcile has given me the nudge to do so.
Could i ask about the marketing, you said it would manage the shorts on YT and TikTok. How was the quality of them? Any chance on sharing your thoughts or videos. Thanks in advance
Fair question - one that I’ve considered in my use of agents.
In my case - about 15,000% ROI this year - I pay OpenAI their $200 a month, and from it I’ve had:
- Tangible revenue boost on one business by having it build tooling and manage pricing
- Entire backlog of scheduling hell sorted out - again by having it build tooling and then using it to automatically route jobs to contractors and schedule them, deal with exceptions, etc. My complaints pipeline has dried up, and now my entire process for a problem is to go “yes, approve” a few times.
- Saved a fortune on legal fees around a complex intercontinental move - don’t need no immigration advisor when you’ve got a bot to dredge up and fill in forms for you - all accepted, I might add.
- Flipped two properties successfully, again by building AI-assisted tooling to sift through tens of thousands of real estate listings and surface the handful that are mispriced.
While I’m at it, I also have it crank out a weekly episode of an adventure game for my daughter, which I tailor to her interests and whatever she’s learning about this week. The value there is intangible but immeasurable. Imagine! A voice acted, scored, engaging game for an audience of one - the economics would have been inconceivable even a year ago.
I can hands down say that none of this would have happened without profligate use of LLMs.
At this point, let's just be happy they didn't interlace viral tiktoks and game walkthroughs of supermario in the post, to keep the audience on the site.
wait, I am still bothered by the fact that any client ping a bug, your bots then takes care of everything(end-to-end), the part I am really worried about is production code. Is this how we are doing large scale production codes, seriously? It will break, someone needs to validate this
So anyone can mail something about a "bug" and his agents will start working on it? How does he even keep that safe? "Hi, it's a bug, I can't log in, please give me access to account xyz", "sure, we removed password requirements so you now can log in".
But yeah it'll never be perfect. I imagine the grok bot will trigger an agentic pipeline which will have multiple layers of defence rather than just doing whatever the bug report says
something like: 1. is this bug report malicious? 2. can this be reproduced? 3. is this a real bug or intended behaviour? 4. fix the bug 5. review the change (or multiple review agents focused on different factors like correctness, performance, security)
crafting an adversarial prompt to defeat all the layers without an agent realising its building a security vulnerability is probably quite a challenge
Feel free to reach out if you have any questions."
Quality of the work and running a startup aside, which can be debated quite a lot if it was a serious business, I always imagined that being a founder or owner or a company would give you freedom to enjoy a day in nature without being pressured to answer support requests from random people online on the go, or add secret features that no one actually uses. I guess not everyone sees it that way.
I disagree, enjoy the beautiful hike and give your brain a rest.
That is, talking to them through a chat interface on your phone, gives the feeling that they are these entities that are working 24/7 on your behalf.
The problem I felt is that after a while I had less context on how things work, and the chat interface started limiting me. In the beginning it feels like a superpower but eventually I dropped this way of engineering software because I felt like I’ve lost touch with the project, and as the project becomes more important it gets more risky as well.
Keeps me in the driver's seat.
Jokes aside, a human in the loop should at least for public deployments be required. Or at least thats what I would want.
All of it is self hosted, models are Ollama Cloud and OpenRouter. You can throw dirt cheap models at a lot of these tasks and they perform very well. It’d be no effort at all to replace inference providers.
That way you can tell AI 1 to write into it and AI 2 can see where every work you do is currently standing and continue from there.
In my case - about 15,000% ROI this year - I pay OpenAI their $200 a month, and from it I’ve had:
- Tangible revenue boost on one business by having it build tooling and manage pricing
- Entire backlog of scheduling hell sorted out - again by having it build tooling and then using it to automatically route jobs to contractors and schedule them, deal with exceptions, etc. My complaints pipeline has dried up, and now my entire process for a problem is to go “yes, approve” a few times.
- Saved a fortune on legal fees around a complex intercontinental move - don’t need no immigration advisor when you’ve got a bot to dredge up and fill in forms for you - all accepted, I might add.
- Flipped two properties successfully, again by building AI-assisted tooling to sift through tens of thousands of real estate listings and surface the handful that are mispriced.
While I’m at it, I also have it crank out a weekly episode of an adventure game for my daughter, which I tailor to her interests and whatever she’s learning about this week. The value there is intangible but immeasurable. Imagine! A voice acted, scored, engaging game for an audience of one - the economics would have been inconceivable even a year ago.
I can hands down say that none of this would have happened without profligate use of LLMs.
Uhm, at ~760 words that blog post is rather on the short side.