OpenIQ

When coding is largely solved, what comes next? It’s been beaten to death by now, but coding has changed enough in the last six months that it’s almost moot to call it coding at all. Even for devs, the job was never only code. Sometimes you’re designing, mostly you’re maintaining and managing the complexity of what’s already there. The work usually split three ways, always unevenly: design, code, monitor. And now two of those are getting eaten. ...

May 15, 2026 · 6 min · Abhiram E

A Shot at Building a Secure Claw

Claws have captured a lot of imagination, and not just among developers. People are using them in all kinds of ways. Even Baby Keem apparently has one running on his phone lol! The possibilities have blown wide open, and agent harnesses are suddenly one of the hottest things in AI engineering for 2026. But underneath all that, a Claw is pretty simple and sadly insecure by default. I wanted to see if I could keep the simplicity and still end up with something I’d trust anywhere near real credentials. This is the story of how. ...

April 21, 2026 · 11 min · Abhiram E

Building a System 2 for Claw

Claw has been keeping me occupied at all times in the day and especially over the weekends. The internet (or could just be my echo chamber) is getting wilder with the experiments. I have tried my hand at a few over the past few weeks. And this is a post about one such experiment that happened this weekend. The Idea The idea is a simple one, can I augment LLM knowledge with data from a curated set of sources that can unlock cross domain connections. Think of it like RAG but the lookup happens in the background, after the model has already responded. ...

March 29, 2026 · 7 min · Abhiram E