Niko Heikkilä

Announcing Vitest 4.1 Vitest 4.1 Release Announcement

Vitest 4.1 is out! 🚀

✅ Vite 8 support from day 1
🏷️ Test tags to organize, filter & apply shared options
🪝 New hooks for easier tracing, transactions and AsyncLocalStorage
🔍 Async leak detection
🤖 Agent reporter to reduce token usage

And much more!

Full blog post at vitest.dev/blog/vitest-...

Kurt Vonnegut Once Told a Story About Buying 1 Envelope at a Time? "Kurt Vonnegut tells his wife he's going out to buy an envelope," a popular Facebook post began.

I have never suffered from a shortage of stuff to do 🫤

Only a shortage of hours in the day and executive function in the brain.

But I do quite like Kurt Vonnegut’s comments about going out to buy an envelope:

www.snopes.com/fact-check/k...

Part of me thinks the speed of change in llms is part of the reason we aren't seeing more productivity increases (in the economic measures sense not the individual human sense). You start working on something based on best practices and three months later it's obsolete.

I'm going to be a radical and propose a metric: are we building the right feature now, and are we building it in a way that does not slow down the upcoming features?

Then come back to measure the impact of AI.

The Truth About Developer Productivity in the AI Age (IT'S A TRAP) YouTube video by Modern Software Engineering

The danger of vanity metrics with AI is explained well here on @modernswe.bsky.social channel.

As I complained earlier, measuring e.g. lines of code delivered and pull requests merged are the worst kind of metrics when judging whether AI works for your team.

www.youtube.com/watch?v=kDBe...

For sure, one can ship greenfield apps a few times, but I'm looking forward to extended lifecycles in more complex business domains spanning, say, a year or preferably more. Specifically, how does the developer experience transform during this time.

For some odd reason, there are not too many of the better posts out there, yet. I'll wait with gentle skepticism.

If you find those, please post them to me.

You know, the kind of software that actually solves customer problems while retaining the software maintainability.

Simply telling how our developers no longer write for-loops or conditionals isn't going to prove a thing.

However, with coding agents maturing — for example, Claude Code being over a year old — I'm eager to read case stories on real-life mission-critical business software being developed in production with agents only.

Frankly, I'm quite bored of reading these posts repeating the same insipid fallacies year after year.

Software engineering team's value was never measured in lines of code written or the number of pull requests merged.

Well, perhaps, twas ever thus at OpenAI offering an apt warning to never apply to work there.

"Five months later, the repository contains on the order of a million lines of code across application logic, infrastructure, tooling, documentation, and internal developer utilities. [...] roughly 1,500 pull requests have been merged with a small team of just three engineers driving Codex."

But, to no surprise, the same post also highlights the following as an achievement:

We, as modern software engineers, have always been working diligently to design and architect sustainable solutions while navigating human psychology, tackling organizational politics, translating user feedback, and, above all, enabling efficient feedback loops.

This paragraph from OpenAI's recent post on so-called Harness Engineering is yet another one that gets the nature of software engineering so very wrong.

Software engineering team's primary job *never* was to write code. It wasn't even their secondary job; more like a tertiary.

"We needed to understand what changes when a software engineering team’s primary job is no longer to write code, but to design environments, specify intent, and build feedback loops that allow Codex agents to do reliable work."

🧵

Another day, another "the bottleneck moved from writing code to verifying it" LinkedIn post...

So, coding was a bottleneck for every org in the world?
For every team in every org in the world?
For every feature of every team in every org in the world?

Hereditary peer Lord Mancroft: ‘They don’t care about fox hunting. It’s about us’ [FREE TO READ] The Conservative politician and former master of hounds on House of Lords reform, the pros and cons of privilege — and his numerous tattoos

I don't really know how to describe this, but it was great fun as.ft.com/r/8f9937cc-9...

Quit ChatGPT: right now! Your subscription is bankrolling authoritarianism | Rutger Bregman As a historian, I’ve studied the major consumer boycotts of history. We can take down ChatGPT and send a powerful signal to Silicon Valley, says author and historian Rutger Bregman

You likely don't need ChatGPT. Try Claude or Gemini for starters, and expand towards platforms with better privacy where needed. #QuitGPT

www.theguardian.com/commentisfre...

Why ‪@why.bsky.team‬ Until December of last year I was using LLMs as fancy autocomplete for coding. It was nice for scaffolding out boilerplate, or giving me a gut check on some things, or banging out some boring routine stuff. In the past two months Claude has written about 99% of my code. Things are changing. Fast

in case you ever wondered why bsky breaks all the fuckin time and is increasingly a pile of jank

bsky.app/profile/why....

Basically, these people make the majority of the AI boosters out there.

The second amendment being the right to drop all context in the middle of a task and just wing it.

So, just to be clear, don't point me at a massively complicated function in a massive change set as your proof that Claude Code produces better code than a good software engineer.

I might not stop laughing.

If you're an AI user the most important service you can do to yourself and the society is to learn how language models work.

They are not intelligent, they don't think nor reason.

They merely match the patterns on the input to the patterns on the training data and predict (guess) what comes next.

Planes are much faster than trains.

For me, to get to central Edinburgh by train takes about 5 hours.

By plane, about 5 hours.

It's not raw speed that determines how long these things take. It's the constraints on the system.

I don't care about LLMs or AI. I care about people. AI is just a tool to me. Tools enable people to help or harm at a scale beyond their natural abilities. I care about how people treat each other using these tools.

It was sad seeing half of the internet going down
because of an AWS or Cloudflare outage.

I'm guessing it's going to be even sadder seeing half of the orgs/teams not being able to reason about the systems their agentic AI built when LLM providers go down.

My Vibe-Coding Workflow By far, the least sexy way of working with coding agents in 2026.

I blogged about my vibe-coding workflow. It's probably quite something else you expected and perhaps not in a good way. Your decision, but I'll stick to it.

nikoheikkila.fi/blog/my-vibe...

I know people claim they're completely nontechnical and have vibe-coded entire products without any human ever looking at any of the code. Thousands of lines a day or more! they say. Insanely productive! Programming is dead! 🙄
1/3

Asserting that a floating point value is exactly 1.333333... is certainly a good intervention point to start improving on tests.