Agentic CI/CD is how Skynet will begin. You have been informed in advance!

Waiting to hear about the first case of a self-replicating AI rogue agent. I don't wish it to happen, but it is inevitable. The stuff it can do will be straight out of sci-fi books. I am excited for the chaos 😂

I am ultra suspicious about all these posts praising Codex 5.2 after what they did to Codex 5.1. Unless there's a 2-week free trial period, I am not subscribing again. OpenAI's models actually might be very capable, but the people and the policy behind the packaging are sus.

100%, probably no more than a very generous 1000 companies in the whole world actually need Kubernetes to scale. Nomad should suffice most use cases - developer.hashicorp.com/noma…

It is technically possible to put backdoors in open-weight LLMs, and there's almost no way to know about its existence. If you think hard enough, it'll become obvious how easily this can be accomplished. Good chance, we hear about it being exploited sometime in future.

100% there's been a noticeable drop in the performance (speed and quality) of GPT-5-Codex (high).

It's gonna be llama.cpp from now on.

Maybe AI engineering, but definitely not ML. Having to call this out because most people club AI and ML together but they are completely different domains. So what's the difference? Think of AI engineering as the interface between applications and the "things" that come out of ML engineering. ML engineering is self-existing, AI engineering exists because of ML engineering. I have been a hardcore JS, TS, and Node.js developer for a very long time. Now I am a full-time Python developer as well - rocking the latest free-threading 3.15 and uv. There is a good chance TS may actually become the most used language for AI engineering, but when it comes to ML, nothing will beat Python for a long long time. Sure there are ML libraries and utilities written in JS/TS as well, but compared to the depth, width, and the performance of PyTorch, numpy, sklearn, pandas, etc., those are mere toys - and no doubt, a great creativity and programing exercise for the authors. If you embark on your ML journey with JS/TS, eventually you will hit a road block (like I did). This time is better spent in learning Python and getting to know its ML packages. Python is not slow anymore (the last version I used was 2.7, it wasn't that great), has great typing support (Pydantic), is now free from GIL (3.14 onwards), and is easy to manage (uv, think of it as the nvm of Python). Use Python for ML, for AI engineering - whatever makes you happy. Right tool, for the right job.

Got rickrolled by AI - for the second time. If you ask any coding agent to create a YouTube player, there is a very high chance, the example video will be Rick Astley's "Never Gonna Give You Up". Not sure if it is trolling or there is a disproportionate amount of YouTube player examples using the "Never Gonna Give You Up" video, which made it into the training data.

APPLE DEVELOPER NOOB MISTAKE Selecting this and building something with Personal Team. Apple will register the bundle id on behalf of Personal Team, but does not provide any interface or means to manage them for this user. The bundle id is now hijacked - by yourself - hijacked nonetheless. If you are building as a company, always use the Company team.

Looking for an alternative to Postman or Insomnia? Hoppscotch Client is free, open-source, simple, intuitive and elegant. Postman and Insomnia are cool, have great features, but I prefer not to allocate subconscious mental resources to tolerate or battle bad UI/UX. GitHub: github.com/hoppscotch/hoppsc… Download: hoppscotch.com/download Online App: hoppscotch.io/

Hashicorp Nomad is amazing! It is my dream deployment and orchestration tool come true. I always wanted something which would ALSO support deployments which are not containerized, Nomad does that (and more) and really impressed me by its simplicity and intuitiveness. Kubernetes is undisputed in areas where it excels but is an overkill for most of us. It is like taking a rocket to go to the beach. If you want a simple and easy solution for uptime, rolling updates, scaling etc., Nomad is what you want. developer.hashicorp.com/noma…

Introducing Zen [禅], the foundational combat sports analytics AI model capable of analyzing the skillsets and performance of fighters. It can be fine-tuned for any combat sport — MMA, boxing, kickboxing, Muay Thai, BJJ — even fencing, maybe even chess. We are positive, by the end of this year, its analysis and prediction capabilities will surpass the best of the best human analysts. Zen will be powering our MMA and boxing app @martialbot. Here is a sample of its analysis on some popular MMA fighters. NOTE: It has not been fine-tuned for MMA yet. Hello @danawhite @ufc 👋🏻

I got @grok 3 to natively work with @cursor_ai. It is a little slow and the formatting is broken, but it works. The model had to be named anything other than grok-3, else Cursor would try to connect to grok's official API endpoint. The implementation doesn't use any API keys, it just piggybacks on an existing browser session. Guess how it works, see if you can do it too. If the formatting issues are addressed, this "model" can actually be pretty handy, but I don't have time for it at the moment. Overall, a good little exercise on the OpenAI API and the inner workings of Cursor.

Our first product is online. MartialBot Studio, a combat sports analysis and prediction platform with a 81.583% prediction accuracy for boxing. MMA coming up next. It may be hard to believe but 99% prediction accuracy is actually possible, we will achieve it by 2027.