They’re made out of weights
Article URL: https://maxleiter.com/blog/weights Comments URL: https://news.ycombinator.com/item?id=48391611 Points: 1383 # Comments: 615
Insights
Practical writing on software architecture, SaaS products, AI automation, legacy modernisation, and the business of building reliable systems.
Curated links from external sources — not 360Softy original articles.
Article URL: https://maxleiter.com/blog/weights Comments URL: https://news.ycombinator.com/item?id=48391611 Points: 1383 # Comments: 615
Picture this: you write a requirement. Clear, specific. The agent reads it, does exactly what you wrote — and breaks three things you never mentioned. Not because the agent is bad at its job. Because the spec was written for a human reader, not a machine. Human developers tolerate ambiguity. They fill gaps from experience. They ask before doing something irreversible. AI agents don't work that way — they fill gaps with whatever their training data suggests. Sometimes that's fine. Sometimes it de
Power it on, and it's supposed to be ready to go on its own, no operator steps. Then one day, after a couple of restarts, it just... wasn't. Here's the clever safety workaround that quietly caused it. I had to build one testing tool that could behave as two opposite safety systems, but only ever connected to one at a time, with both behaviours crammed into a single safety system, sharing the same connection ID, over FSoE. The textbook move would've been two separate, validated configs with only
I come from a physical security space, mainly man-guarding and asset protection. I recently took the challenge to venture into information and cyber security. So far I can say the mentality for both is the same; they differ in technique but the outcomes are the same, in that both are primarily focused on asset or data privacy and protection. Offensive cybersecurity often happens between nation states, but that does not mean corporate entities or individuals do not indulge. They do so cautiously,
I built Opendria to explore how generative AI can be coupled with cognitive and psychological state machines to create immersive educational experiences. Instead of using simple LLM wrappers, Opendria runs a backend simulation of brain, emotional, and cognitive states for 80 historical figures (like Einstein, Aristotle, Da Vinci, and Newton): Cognitive Profiling: The backend dynamically calculates cognitive metrics (e.g., prefrontal cortex and temporal lobe scores) based on prompt sentiment and
Opendria — The Living Library Engage in deep, immersive dialogues with history's greatest minds, philosophers, and mentors. Access Opendria — The Living Library powered by advanced AI. opendria.com
Ruby Reactor 0.4.0 introduces interrupts — the first native saga pause/resume mechanism in the Ruby ecosystem. You've built a checkout flow. The user submits their order, you reserve inventory, charge their card, and generate a shipping label. Everything's a Sidekiq job. It works. Then you get the requirement: "After payment, wait for the fraud detection webhook before shipping." Suddenly your clean Sidekiq pipeline needs to stop, hold state for an indeterminate amount of time, and resume when
What I Built CodeSynth started as a hackathon experiment. At the time, I wanted to build something that genuinely used real-time systems instead of making yet another chat app. While exploring WebSockets, I came across collaborative coding environments and thought: “What if I could build my own mini multiplayer IDE?” As a second-year university student, that sounded both exciting and slightly terrifying — which made it the perfect hackathon project. And that’s how CodeSynth was born. The orig
Work with 360Softy
Book a free consultation and we will tell you honestly whether we can help.