Engineering insights, design patterns, and lessons from building AI-powered tools at Pixl.
I'm not launching a platform. I'm proving one audit-to-website loop, end to end. Come break it.
Hamza Mounir
Jul 16, 2026 · 2 min
The AI can draft anything it likes. The one thing it can't do without me is touch the real world.
Jul 15, 2026 · 2 min
Your agent count tells me how many things you can start. It says nothing about whether you can finish one with its meaning intact.
Jul 14, 2026 · 2 min
I've shipped plenty of impressive answers. None of them mattered until they survived all the way to a measured result.
Jul 13, 2026 · 2 min
Bolting an assistant onto every tool made me faster and fixed nothing. Here is what changed my mind.
Jul 12, 2026 · 2 min
Every illustration here is generated from one manifest. The spec, the pipeline, and the honest mistake I made on the first pass.
Jul 9, 2026 · 3 min
In April I published a vision series. Three months later, half of it is revised and two threads are gone. Here's the honest diff, and what survived every pivot.
I audited my own fleet and found the most expensive thing in software: working code with zero callers. Here's the one rule that came out of it.
Jul 9, 2026 · 2 min
I stopped writing SEO for a ranking and started writing it for a reader that isn't human. Structured data, llms.txt, and being cited by ChatGPT and Perplexity.
Jul 9, 2026 · 4 min
How I generate a branded video: write the shot list, generate each shot in one house style, then bring it to life. Same system as the images. With an example.
What I'm building isn't a set of tools. It's a company factory: one command provisions identity, context, memory, budget and services. Then agents run it.
Five LLM personas, a different model each. Goal: become millionaires together. Budget $10. The real numbers.
Apr 19, 2026 · 4 min
My take on where tech goes next: from tools to systems, from prompting to trained specialization, and eventually into robots. A bet, not a forecast.
Apr 8, 2026 · 3 min
Everyone asks which model to buy. Wrong question. AI only pays off when it sits on structure: knowledge, rules, execution, and a human with the last call.
The foundation is a harness and a Knowledge OS. The next layer is specialized, fine-tuned local models: narrow jobs, less prompt tax, more reliable execution.
My first posts were about the invisible layers. Robotics is where they get a body: systems that perceive, decide, and act in the physical world.
Apr 8, 2026 · 2 min
Two layers sit under almost everything I build: a harness that drives execution and a Knowledge OS that structures context. The shape, not the blueprint.
A small JSON object that carries decisions and constraints between the stages of a multi-agent AI pipeline.
Feb 16, 2026 · 2 min
