If you only read one thingThe Illustrated DeepSeek-R1 (40 minute read) A deep explanation of the DeepSeek-R1 model - with a recap of the basics of LLMs, LLM training, how reasoning (think O1) works, reinforcement learning, LLM architecture, and then applying this to DeepSeek-R1. Beside wanting to talk about R1 with your peers (or your CEO might ask you!), there is always the discussion on how much you need to know as an engineering manager. But you do know what a compiler does, and what a programming languages is. You need to get the same understanding of how AI works. Must read if youāre not an AI expert. https://newsletter.languagemodels.co/p/the-illustrated-deepseek-r1 Graph of the week![]() On Hackernews there are monthly āWhoās hiringā threads. This article analyzes the trends. Every programming language went massively down from itās peak around July 2021 - like 2/3 down. Could be that there is a hidden recession, companies get rid of āghost developersā, the AI winter for developers is coming or Hackernews is no longer relevant, or itās a startup thing . Iād bet on AI, but your guess is as good as mine. Beside that it is interesting on what the market is looking for, it might influence your tech decisions. (The discussion on hackernews can be found here https://news.ycombinator.com/item?id=42858947 with many more ideas for the reasons)
Stories Iāve enjoyed this weekThe Mythical IO-Bound Rails App (11 minute read) I cringe every time someone says, the choice of programming language doesnāt matter for performance, everything is IO bound. But here are some numbers here PLUS some clever thinking. When a Ruby JIT can reduce latency by 15%, the programming language is relevant for latency - it is not IO bound - and therefor costs. Running Go is cheaper than running Ruby. And some of my clients could slash AWS costs that way (There are of course other considerations, like Rails brings everything with it, Python has the most programmers etc.) https://byroot.github.io/ruby/performance/2025/01/23/the-mythical-io-bound-rails-app.html No, I do not want AI to āpolishā me. (31 minute read) When writing my book, or writing articles, I have played around with AI. And the output didnāt feel like me, or my voice. So I dropped it. I write everything on my own - which I think gives me a unique, though imperfect, voice. The future of AI writing is an opportunity to stick out but not sounding like an AI (of course, people will train AIs to not sound like an AI I fear). https://thebloggess.com/2025/01/28/no-i-do-not-want-ai-to-polish-me/ Discovery Coding (5 minute read) āIn writing (particularly fiction writing), there is a generally accepted distinction between authors who start by writing an outline and those who discover their stories through the process of writing. For some reason, we have no such distinction in programming, so I am here to introduce it.ā Just writing code has been disdained, while just writing a book and then rewrite and proof read the result is what some successful writers are doing. Interesting take on ājust sit down and write codeā #DiscoveryCoding https://jimmyhmiller.github.io/discovery-coding Tool touted as āfirst AI software engineerā is bad at its job, testers claim (10 minute read) This is the Wild West. We have no clue what works or what direction to take. Looks like Devin isnāt as good as some say. Good to keep in mind. āTasks that seemed straightforward often took days rather than hours, with Devin getting stuck in technical dead-ends or producing overly complex, unusable solutionsā https://www.theregister.com/2025/01/23/ai_developer_devin_poor_reviews/ DeepSeek My User Agent (4 minute read) Funny. https://www.jasonthorsness.com/20 The Management Paradox (12 minute read) āThe tech industry has a lot of unwritten rules. One of the most rigid is ādonāt micromanage.ā You need to manage everyone the way they need to be managed. One of my biggest mistake in the beginning of my management career, I was managing people the way I wanted to be managed. Donāt do that. Instead, manage a junior by giving tasks, deadlines (expectations) and checking in. Manage someone more senior by delegating features or projects. Manage someone even more senior by delegating goals and metrics, let they come up with solutions, features and projects on their own. Micromanagement is good, if it is the right thing. Caveat: If you have too many juniors - like many of my CTO clients in startups, because of the money - you spend most of the time managing. Get the budget to hire more senior people, or you donāt scale as a person. https://bloomt.org/p/the-management-paradox Join the CTO newsletter! | |