I write
28 posts / since 2020
I'm Tejas Kumar, and I write about AI engineering, React, and the working life of a software engineer. I'm an AI Engineer at IBM based in Berlin, the author of Fluent React (O'Reilly), and the host of the ConTejas Code podcast. The archive goes back to 2020 and includes a hands on benchmark of frontier models for summarising a conversation, a long piece on how severe hemophilia A shaped my health and my work in developer relations, and a practical model for growing professional relationships.
Elsewhere on this site: Fluent React, the book on how React works inside, every talk I have given, the ConTejas Code podcast, and the workshops I teach for engineering teams.
Please Stop the Slop: What AI Slop Is and How to Spot It
My verdict on AI slop: what it is, how to spot it, why not all AI is slop and a petition to keep the internet human.
Can AI Feel Pain? Why Anthropic Banned Cruelty to Claude
Nobody can prove AI feels pain, but 25 open models have a pain direction that rises when you insult them. What it shows and why they're still machines.
The Apple Watch Readiness Score Is Wrong so I Built My Own
My Apple Watch told me to Go For It on my most run-down mornings. Why the Readiness score misses and the readiness app I built in 5 days.
What Are Open Weights? Why People Want GPT-4o Released
Open weights means you can download a model and run it yourself. What the GPT-4o petition is really asking, and why releasing the weights is the hard part.
AI Evals: How to Build an LLM Judge That Agrees With You
AI evals are fuzzy unit tests. My AI Engineer World's Fair workshop, written up: build an LLM judge, catch its biases and gate deploys on it.
Kolibri vs Claude Sonnet 5.5: A German LLM Benchmark
Aleph Alpha's Kolibri lost all 136 blind comparisons to Claude Sonnet 5.5 at German Bible deep dives, despite clean grammar and a leaner tokenizer.
AI Is Now SI: Super Intelligence Isn't Superior
US agencies now call AI Super Intelligence (SI), yet its definition didn't change. What it beats people at, where it fails, and why people make idols.
Aleph Alpha Kolibri: How the Sovereign German LLM Works
Kolibri is Aleph Alpha's open-weight German and English LLM: 78B parameters, 3.5B per token. How it works, what it's good and bad at, and when to use it.
What Is WebMCP? How My Website Gives AI Agents Tools
WebMCP is a proposed web standard that lets a page give AI agents tools to call. How tej.as offers agents 4 of them, and what WebMCP is good for.
How Much Does a Keynote Speaker Cost?
A keynote speaker costs $2,500 at the median in 2026 and $10,000 to $25,000 for an established author, plus travel. The benchmarks and what moves them.
What Is React Fiber? How React Works Under the Hood
React Fiber is how React has rendered since version 16: a tree of small units of work it can pause and resume. How it works, from the React source.
What Does the Bible Say About AI?
The Bible never mentions AI, but it says plenty about who we are, what we make and who we trust. An AI engineer who builds a Bible app on what that means.
What Is Good Taste in Software? How to Develop It
Good taste in software is a trainable skill: judging quality against external standards. 6 practice protocols, with the research under each one.
How Long Should a Keynote Speaker Speak?
A keynote runs 20 to 45 minutes, most often 30. What the attention research really says, what conferences schedule, and how to use the slot you get.
What Is an Agent Harness? Harness Engineering Explained
An agent harness is everything around an AI model that grounds it in reality: tools, context, guardrails, the loop and verification. How to build one.
Frontier Models Suck Sometimes (A Benchmark)
Apple Intelligence refused my voice note, so I benchmarked 11 AI models on it. A cheap one beat the frontier and my phone's Qwen made things up.
Biggest Lessons of 2024: Honor, Trauma, and People
Dissecting honor, what it takes to overcome trauma, how to heal from the fear of failure and prioritize relationships.
It's Okay to Code on Nights and Weekends
I love coding on nights and weekends because it's my favorite hobby. It only becomes a problem when it turns into a god or an expectation.
How to Grow Professional Relationships
My TJS model maps professional relationships across seven states, from zero-sum competition to toxic enmeshment, and why collaboration is the peak.
Health, Fitness, DevRel, and Hemophilia
A deeply personal post about my health journey and how it's affected my fitness, devrel work, and more.
Pillars of Impact for Web Engineers
I break down the four pillars of impact for web engineers: emotional, social, creative, and technical, and how to grow each one over time.
Goodbye Facebook
Why I'm deleting Facebook, Instagram, and WhatsApp: FOMO, lost time, and engagement algorithms that radicalize people. I'm moving to Signal instead.
2020 in Review: Still Thankful
A month-by-month look back at my 2020: the friends, family, and small moments that made a pandemic year worth being thankful for.
Why I Don't Prefer Speaking at Remote Conferences
Why I'd rather not speak at virtual conferences during COVID-19: the hallway conversations are the point, and video calls lose that connection.
Why I Think Remote Work is Incredible
Two months into my first fully remote job, I compare it with office life: deeper focus, better health, and the sleep rhythm trap I had to fix.
5 things I love about my website (and how they're built)
I walk through five things I love about my website, including automatic dark mode, Next.js server rendering, whimsy, and GitHub as my backend.
What I Learned From Getting Pushed Off My Bicycle
A blog post about things I learned after being randomly attacked on the streets of Berlin when a stranger pushed me off my bicycle.
Hello World
My first post here: a quick hello world I published to make sure the blog works. The real writing starts after this one.