Alex Kwiatkowski

intelligence = compression + learning_rate

Custom simulation engine. Feeling good about the camera/nav system. Video recorded in browser using the WebCodec API

Prediction is compression

I strongly believe there are entire companies right now under heavy AI psychosis and its impossible to have rational conversations about it with them. I can't name any specific people because they include personal friends I deeply respect, but I worry about how this plays out. I lived through the great MTBF vs MTTR (mean-time-between-failure vs. mean-time-to-recovery) reckoning of infrastructure during the transition to cloud and cloud automation. All those arguments are rearing their ugly heads again but now its... the whole software development industry (maybe the whole world, really). It's frightening, because the psychosis folks operate under an almost absolute "MTTR is all you need" mentality: "its fine to ship bugs because the agents will fix them so quickly and at a scale humans can't do!" We learned in infrastructure that MTTR is great but you can't yeet resilient systems entirely. The main issue is I don't even know how to bring this up to people I know personally, because bringing this topic up leads to immediately dismissals like "no no, it has full test coverage" or "bug reports are going down" or something, which just don't paint the whole picture. We already learned this lesson once in infrastructure: you can automate yourself into a very resilient catastrophe machine. Systems can appear healthy by local metrics while globally becoming incomprehensible. Bug reports can go down while latent risk explodes. Test coverage can rise while semantic understanding falls. Changes happens so fast that nobody notices the underlying architecture decaying. I worry.

@nbaschez Would love to compare notes. I've been building keel for this exact pain. AI-authored planning markdown gets treated as enforceable state (vision → epic → story), with a doctor command that catches drift before review. github.com/spoke-sh/keel

Do you spend a lot of time reviewing markdown docs written by AI? Wish it were a better experience? Say hi if you wanna try a new (free, open source) thing

To solve a problem write a program

The new ChatGPT image model is really solid. Feels much better at following fine grained direction and editing instructions.

@hthieblot Keel spoke.sh/keel/

pitch me your company in 1 word.

I just chat to it, get it to ask me questions, provide responses, create keel mission, execute keel mission spoke.sh/keel/

@boardyai spoke.sh/keel It's a turn-based board engine for human/AI delivery teams. It allows software teams to move much faster by having agents participate in the SDLC process.

Founders, drop what you're working on in the comments I'll intro you to investors, co-founders, and engineers.

There were signs

@ThePrimeagen x.com/rupurt/status/…

Hey, you got a cool project that you are building? Link it I want to yap about cool projects

Paddles keeps running into spikes. I wonder if I can throw the Gemma 4 ball into the pond before Cloud Next? blog.google/innovation-and…

@AlexFinn Adding support in my local recursive context harness github.com/spoke-sh/paddl…

Do you even understand what this means? An open source model just released that is: • Outperforms models 20x its size • Can run on a base model Mac Mini • Is AMERICAN 🇺🇸 If you have a base model Mac Mini you can have unlimited super intelligence on your desk. For free. Sonnet 4.5 was released 5 months ago In 5 months that level of intelligence went from frontier to free on your desk And not only that, can run on any basically any computer out there If you have even a remotely modern computer, do the following immediately: 1. Download LM Studio 2. Go to your OpenClaw and ask which of these new Gemma 4 models is best for your hardware 3. Have it walk you through downloading and loading it 4. Build apps with it knowing you are using your own personal, private super intelligence on your desk The people denying this is the future are so beyond lost.

Come play my text based adventure work games. Soon questing will never be so fun... OpenAI's 2x limits come to a close tomorrow. Forever grateful of the progress it allowed me to make. I'm maxing out 2 Pro accounts, 1x Claude Max, 1x Gemini Ultra and a 1x Kimi Allegretto. If you want help setting up your own software factory or just want to talk shop about AI engineering hit me up. alexdk.com/contact

Custom local model/remote API harness Paddles (github.com/spoke-sh/paddl…). Recursive context management with a different take on compaction using pressure heuristics with state space actions decided by the model first. Kimi 2.5 in the screenshot Adding support for the exciting new @PrismML 1-bit model released today!

Multi player software development is now a scheduling problem

the most magnificent data visualization ever.. literally an art piece wizard » @refikanadol

@AnthropicAI Have you tried giving project management tools to the harness? I’ve had great success with a custom CLI tool that manages missions/epics/stories. Getting creative with verification is where I spend most of my time now

New on the Anthropic Engineering Blog: How we use a multi-agent harness to push Claude further in frontend design and long-running autonomous software engineering. Read more: anthropic.com/engineering/ha…