Viktor Schmidt

I learned 4 mathematical testing techniques in university in 2013. Never used them. Too much manual overhead. 13 years later, I built AI agents that apply them systematically to a Rails codebase with 13,000+ specs. The bottleneck was not test quality. #RubyOnRails #Ruby #AI

@andrzejkrzywda Yes, that's true 😇. But not the first day on the internet 😅. But, we do not need to go down to the digital world for this... Real life is often not much better when you make established things break. 🍝👌🏻

@viktorianer4 first time on reddit?

Interesting to read comments. How easy you are on the side where haters are targeting you without knowing anything about you. But whoever does not read the full story does not reserve even a second for response. 🤷🏼‍♂️

@srbaker @seuros @strzibnyj Yeh, I do it as well 😎👆🏼

@seuros @viktorianer4 @strzibnyj I am very happy to help teams move from RSpec to Minitest. And I'd probably let AI do the whole thing.

Yet another incredible RSpec to Minitest success story

🚨Shocking: A 25,000-task experiment just proved that the entire multi-agent AI framework industry is built on the wrong assumption. Every major framework - CrewAI, AutoGen, MetaGPT, ChatDev - starts from the same premise: assign roles, define hierarchies, let a coordinator distribute work. Researchers tested 8 coordination protocols across 8 models and up to 256 agents. The protocol where agents were given NO assigned roles, NO hierarchy, and NO coordinator outperformed centralized coordination by 14%. The gap between the best and worst protocol was 44%. That's not noise. That's a completely different outcome depending on how you organize the agents - not which model you use. Here's what makes this uncomfortable: When agents were simply given a fixed turn order and told "figure it out," they spontaneously invented 5,006 unique specialized roles from just 8 agents. They voluntarily sat out tasks they weren't good at. They formed their own shallow hierarchies - without anyone designing them. The researchers call it the "endogeneity paradox." The best coordination isn't maximum control or maximum freedom. It's minimal scaffolding - just enough structure for self-organization to emerge. But there's a catch nobody building agents wants to hear: below a certain model capability threshold, the effect reverses. Weaker models actually need rigid structure. Autonomy only works when the model is smart enough to use it. Which means every agent framework shipping with one-size-fits-all hierarchies is wrong twice - over-constraining strong models and under-constraining weak ones. The $2B+ invested in agent orchestration tooling may be solving a problem that capable models solve better on their own.

@ndbroadbent @strzibnyj I would not suggest you using typing with AI. LLM is much less effective with types. Use Ruby power.

@strzibnyj This sounds awesome. I might try this. One other (controversial) reason is that Minitest is much nicer for Sorbet typechecking than RSpec’s DSL

@sean_j_roberts @strzibnyj What do you mean with "BS"?

@strzibnyj 25 minutes to 14 seconds sounds like BS

@mooktakim @strzibnyj I do not know where it comes from. I do not see any benefits from wiring RSpec. But, this is probably just test. I write RSpec as much as Minitest. Used also FactoryBot for years together with Minitest. I found fixtures and Minitest easy to read. Factories are easier to write.

@strzibnyj I like rspec because it's easier to write understandable tests. But with AI maybe people don't write the tests anymore and the actual tools don't matter. Maybe it's time to revive cucumber. Human writes cucumber and the rest is AI.

@michaelkoper @strzibnyj You don't won't spend any time to see this code. That's exactly why we spend 2 months to solve it in RSpec first, as you can read. It was much easier to rewrite it.

@strzibnyj I would honestly love to see their codebase. I had only ~4% gain from going FactoryBot to Fixtures. And a negative gain from going rspec to minitest... My guess is that they made the Factories way too complicated/smart.

@dbackeus @strzibnyj Thanks for reading.

@strzibnyj The actual story is FactoryBot to fixtures...

@strzibnyj @dbackeus Because it is coupled.

@dbackeus Yes, true. I couple them together often since it's so unusual when they are not

@seuros @strzibnyj You obviously do not know what to say. Thanks for making noise. Helps me as well.

This is noise and a lie, authored by someone who either wants clout or doesn’t know what they're doing. RSpec is slower and creates a lot of magic you don’t need 99% of the time. But unless this community is turning into MLM vendors, there is absolutely no way this story is true. Even if the tests were rewritten in Assembly, you wouldn’t reach that kind of speed boost. The agents probably stubbed everything. I know FactoryBot, RSpec, and Minitest internals (I’ve contributed to them). This story smells like: I pressed the turbo button on an old DX486 and now I’m playing GTA 6 in 8K with it. Just use @grok or any LLM to verify the claims. These people are hoping someone using RSpec will contact them and ask: "Are you available for hire? We have a test suite that takes 2 hours; I heard you can make it 47 seconds."

@strzibnyj Thanks for sharing. Interesting to read comments. How easy you are on the side where haters are targeting you without knowing anything about you. But whoever does not read the full story does not reserve even a second for response. 🤷🏼‍♂️

@m_isabegovic @palkan_tula @RubyConfAT Actually, I am for Italy in this case. Anyway, I have my ticket already 😇.

If Bosnia & Herzegovina wins against Italy tonight and qualifies for the World Cup, I will give away 10 free tickets for @RubyConfAT (29.05.-31.05., Vienna, Austria). ♥️🇧🇦🤲

Full writeup with command file structure, agent definitions, gate types, and scaling data from 426 to 9,000 tests across 60+ models: augmentedcode.dev/multi-agent-pi…

The secret sauce is the Clean Room context strategy. Each subagent spawns fresh, no inherited context. A Writer on a 2,195-line god model gets only its plan slice. Disk is the shared state. Everything else is ephemeral.

I built a 6-gate AI agent pipeline to migrate 9,000 RSpec tests to Minitest with fixtures. The key insight: structural constraints (I/O contracts, deterministic gates) do more work than prompt engineering. #Ruby #RubyOnRails #AI

AI is bad with Hotwire. @julian_rubisch is trying to change it. For years he was building a library of advanced Hotwire behaviors hotwire.club Now, it's open source! Add these skills to your app and turn your Hotwire from 💩 to 🥇 Thanks Julian! 📷🙏

In the meantime the biggest #OpenAI model to "Can I charge #iPhone wireless on #MacBook Pro: Yes, but only if your MacBook Pro supports reverse wireless charging — and currently, it does not.

Karpathy's AutoResearch is changing how campaigns get optimized and most marketers haven´t heard of it yet. Ole Lehmann tested it on landing page copy, 56% → 92% pass rate overnight. here´s how it works for marketing / skills 🧵