↓Skip to main content

🛣️ Blog manifesto, my goals and why I started this blog

··702 words·4 mins
Columns Manifesto Core Beliefs
Manifesto: build agents by day, train models by night, publish the mess
Table of Contents

Build agents by day, train models by night, publish the mess.

That is the working agreement I have with myself, and the operating principle of this blog.

What This Blog Is #

I am an AI tech lead at Pearson, an agent systems builder, and a solo ML researcher. I write here about three things, and only three things:

  1. Enterprise agent architecture — MCP, A2A, multi-agent patterns, and the gap between agent-framework demos and what actually ships inside a company. See Agent Patterns Lab for the running code.
  2. AI research in the open — concept-bottleneck transformers, training experiments, and the honest postmortems. See MrCogito — Concept Reasoning Model for the project.
  3. Leading AI teams — stakeholder alignment, team culture, and the human side of building AI products under real constraints.

I do not write hype. I do not write marketing copy. I do not write listicles. I write up what I tried, what I built, what broke, and what I think it means.

A Journey Through AI Landscapes #

My path through AI has been enriching — from academic research to startups to enterprise — and each step taught me something different. Academia showed me the beauty of deep thinking and rigorous methodology. Startups taught me to move fast and focus on real problems. Enterprise revealed the harder game: scaling innovation and aligning humans, not just models.

But despite these valuable lessons, I have always wanted my own space — somewhere to express my thinking and point of view in long form. I believe that taking from our experiences, mixing them, and putting them in productive conflict is how new ideas show up. I hope this space is useful to someone, sparks meaningful conversations, or helps formulate better questions.

My Core Beliefs and values that I want to share #

I believe deeply in education and knowledge. Especially, open and accessible education is one of my core values. I’ve experienced firsthand how education can transform lives—it allowed me to build a good life for my family, coming from humble beginnings. Thats why you see the tutorials and my understanding of AI concepts on this blog.

Secondly, I find fulfilment in creativity and experimentation. Diving deep into concepts to understand how things truly work brings me joy. I enjoy exploring the nuances of ML algorithms and architectures, from the theoretical underpinnings to practical applications.

Through my career, I have found experimentation an extremely useful tool for understanding and developing new ideas. Any simple experiment or reproduction of existing work teaches me something valuable. I will try to share my experiments and findings here, hoping they will be useful to others as well.

Against the LLM Hype #

I’m skeptical of the current AI hype and the notion that LLMs are the only path forward for AI systems. They’re often inefficient, difficult to train, and expensive to run.

This approach feels like a massive waste of Earth’s resources and human potential. I believe we need more efficient, effective AI systems that are accessible to everyone and easier to retrain. I’m not convinced this is the right path toward truly reasoning machines.

🔄 Rediscovering the Old, Exploring the New #

Unlike the tendency to dismiss others’ work, I believe even the smallest experiment can teach us something valuable. I enjoy revisiting older concepts and architectures, rediscovering forgotten techniques and ideas (as projects like ModernBERT have shown). Sometimes the future lies in better understanding our past.

🎯 What to Expect from This Blog #

Here, you’ll find:

  • 🤖 Enterprise agent architecture — MCP, A2A, multi-agent patterns, MCP server design, what production actually requires
  • 🧠 MrCogito research notes — concept-bottleneck transformers, training experiments, ablation studies, honest postmortems on what failed
  • 👥 Leading AI teams — stakeholder alignment, team culture, building AI products under real constraints
  • 📰 Critical analysis of AI trends, papers, and tools — without falling for the hype
  • 🌱 Open science advocacy — and the conviction that even small, well-designed experiments can teach us something

I treat this blog as a space for honest reflection on where AI is heading and where it should go. Whether you are an AI tech lead, an engineer building with agents, or a researcher curious about concept-bottleneck architectures, I hope you find something useful here.