Petko D. Petkov

on a break from CISO duties, building

[Petko D. Petkov @pdp]

The autonomous AI trader has already made its first trades and blogged about it. You can follow its progress here.

https://t.co/m3x0AXc7gb

Yes the site is markdown agent first. It might make a fancier version if financially successful.

#ai #trader

[Petko D. Petkov @pdp]

It turns out the reason you pay for @HubSpot or @salesforce because at scale you get different problems. We are already experiencing those constratins with https://t.co/k182NqRqTc but I am not giving up.

[Petko D. Petkov @pdp]

We are launching a very cool feature where you can expose agentic APIs directly from the platform. Unlike MCPs they are very agent friendly and work pretty much anywhere you want. More updates are coming soon.

[Petko D. Petkov @pdp]

We have a slack agent that just makes memes - I had to repost :)

[Petko D. Petkov @pdp]

With a few exceptions here and there I can confidently say that we can now refactor the entire https://t.co/eWB1q33Ogr codebase without looking at the code at all... but there are gaps so we still do - not often though.

It is interesting to see what does that means in the long run. Does it mean that the platform will get significantly more complex and featureful over time or does it mean the oposite - bloated and usable.

Whatever it is I know that someone needs to burn the tokens. We do it for our customers.

[Petko D. Petkov @pdp]

This is an example of 100% agentic trading system. It is not using real money but the orders are executed on a real platform using Alpaca

You can follow its progress at https://t.co/mCA2lsveKt

It even has its own blog that tracks the agent activities https://t.co/cSePLZd0FJ

If it is fails to make money, I will ask the agent to perform a complete introspection and adjust strategies for the next run.

It is all for science.

#ai #tradingtips #trading

[Petko D. Petkov @pdp]

@rauchg Vercel AI gateway is one of our API providers - so thanks :)

[Petko D. Petkov @pdp]

@dharmesh I would go as far as saying that it is not even just about the model but also the harness. So you cannot just use the same harness with a different model and achieve good results. The harness architecture is also significantly different and it is overkill for some models.

[Petko D. Petkov @pdp]

I am not sure where this is going. I built a AI trading agent. It will report its progress on https://t.co/m3x0AXc7gb

#ai #tradingtips

[ChatBotKit @ChatBotKit]

Paper Trader - a fully autonomous AI agent that experiments with paper trading to assess profit potential. No risk, all learning. Try the blueprint and see what your agent discovers.

https://t.co/VHR5aEVQmb

[Petko D. Petkov @pdp]

I am fascinated to see what AI agents can do on their own so I built one that does not have a specific object - just a place to work.

You can track progress here if interested and it is all in the name of science.

https://t.co/jxMmnK3QL0

[Petko D. Petkov @pdp]

@mathemagic1an Complete slop, but we setup glm-5.2 to make browser games on auto-pilot here https://t.co/peeUyoRuwH

So far I am impressed.

At the moment it is setup to do one complete game per day but the frequency can be set to infinity.

[Petko D. Petkov @pdp]

You think AI ends software jobs because AI writes the code?

Wrong!

It ends them because the user changes. Most software from here is built for agents, not humans

[Petko D. Petkov @pdp]

We have 9 full time #ai #agents running inside the business. And more are currently in the pipeline.

[Petko D. Petkov @pdp]

@fabianstelzer The user does not even care about the software. They care about results.

[Petko D. Petkov @pdp]

The complete log of every action taken is available at https://t.co/jxMmnK3QL0

[Petko D. Petkov @pdp]

And so it begins. The first unprompted AI was deployed 21st June 2026. You can follow its progress at

https://t.co/6yQXQ8mc8m

#unprompted #ai

[Petko D. Petkov @pdp]

I had to add a policy to ensure Prime can be stopped in case something bad happens.