If you're a tech leader building an AI product this quarter, you've probably got the team. What you might not have is someone who's made these decisions before on models, architecture, tooling and can tell you which way to go before you find out the hard way.

I'm Nick Sorros. I've spent 13 years building AI systems and 5 running Mantis :mantis:, a consultancy that has shipped AI products for governments, foundations and research institutions worldwide. I've seen what works in production and what fails quietly after the demo.

I offer a focused advisory package for teams actively building right now. 10 sessions over 3 months, on-call support included. If you are interested send me a message in [email protected] ✉️

Writing

The end of repetitive tasks

The AI disruption matrix

Clawdbot in 200 lines of code

Lets talk about clawdbot

The year that Google strikes back

The missing ingredient for AGI

The future of work

My take on Gemini 3

What makes companies fail

Software engineers are obsolete, or are they?

Have you read the AI in 2027 analysis?

The way we work is about to change

Here is how we use AI at Mantis

Here is the issue with AI doing actual work

Back to trillion param models

Are we beginning to use AI too much?

GPT5 is amazing, or is it? 🤔

Agentic RAG is the new oil, don’t miss out!

AgenticAI is everywhere but what is an agent in the first place?

Learn by not using AI or tools

What’s going on with MCP?

Will AI make your role obsolete?

What should you expect from AI in the near term?

What happened with GPT4.5?

Here is how R1 is trained

What is going on with DeepSeek?

The year of agents

My take on Reflection and o1

The year of ChatGPT

Eight examples is all you need ✨

Prototyping with Large Language Models

Reproducible data science

MLOps with SageMaker — Part II

MLOps with SageMaker — Part I

A neural network tagging biomedical grants

Talks

Extreme multilabel classification

Interpretability of LLMs

A tour of LLMs

Rag meets Knowledge bases

🐙 💼 📅 📄