Hacker News

Article URL: https://splitfxm.com

Comments URL: https://news.ycombinator.com/item?id=46955375

Points: 1

# Comments: 0

Article URL: https://www.styloshare.com/

Comments URL: https://news.ycombinator.com/item?id=46955373

Points: 1

# Comments: 1

Article URL: https://singularityhub.com/2026/02/09/scientists-send-secure-quantum-keys-over-62-miles-of-fiber-without-trusted-devices/

Comments URL: https://news.ycombinator.com/item?id=46955367

Points: 3

# Comments: 0

Article URL: https://siliconimist.substack.com/p/thoughts-on-why-fabless-ate-semis

Comments URL: https://news.ycombinator.com/item?id=46955154

Points: 1

# Comments: 0

I built a stack-based programming language in Rust with Cranelift JIT, SPIR-V GPU backend.

Kore has ~140 tokens and 142 opcodes. Programs are sequences of stack operations — push values, apply functions, compose by concatenation. There's a proof checker that statically verifies stack effects and a capability system that gates side effects.

How I got here: I wanted to explore whether a compiler could serve as the reward function for training an LLM to write code. Most code-gen benchmarks use test suites or human eval. But if the compiler can verify correctness before execution — accept or reject in microseconds — you don't need a test harness at all.

That meant I needed two things: a language small enough for an LLM to learn the full vocabulary, and a security model that makes generated programs safe by construction.

The security model (P4: capabilities attenuate): Every side effect in Kore is gated by a capability — io for print/rand, fs for file access, exec for shell commands. The compiler auto-detects what a program needs and tags the binary. At runtime, you grant permissions explicitly (--allow io) or it won't run. When you spawn a sandboxed child, it gets parent_caps & requested_caps — it can never escalate. Pure programs (no capabilities) are deterministic and run on any backend.

This matters because an AI agent writing Kore tools gets compile-time proof that the tool can't touch the filesystem or exec shell commands. Not sandboxing — the proof checker rejects it statically.

I fine-tuned DeepSeek-R1-14B (QLoRA, Unsloth) on 478 curriculum tasks, it worked pretty well. Will probably try a MCTS+RL experiment to see if it can learn to write programs to solve games and optimization problems. The ultimate goal would be having it write its own compiler and self-host.

Comments URL: https://news.ycombinator.com/item?id=46955143

Points: 1

# Comments: 0

Article URL: https://perso.univ-lyon1.fr/thierry.excoffier/ZMW/rr_2003_03_11.pdf

Comments URL: https://news.ycombinator.com/item?id=46955131

Points: 1

# Comments: 0

Article URL: https://getneliva.com

Comments URL: https://news.ycombinator.com/item?id=46955119

Points: 1

# Comments: 0

Article URL: https://business.time.com/2013/03/13/did-google-get-off-easy-with-7-million-wi-spy-settlement/

Comments URL: https://news.ycombinator.com/item?id=46955116

Points: 1

# Comments: 0

FeedbackKit: A developer-first feedback platform for multi-platform apps

Hi HN! I'm Ben, founder of Swiftly Workspace, and I'm launching FeedbackKit - a complete feedback management platform built for developers who ship on multiple platforms.

The problem: Most feedback tools are either abandonware side projects or enterprise solutions too complex for indie devs and small teams. I needed something actively maintained, truly multi-platform, and built for the long haul.

What FeedbackKit does:

Native apps for iPhone, iPad, and macOS (currently in App Store review, but you can access the full admin portal via web now). Soon you'll be able to manage feedback anywhere with native SwiftUI apps optimized for each platform.

Two-way communication with users. When you update feedback status or leave comments, users get automatic notifications. This turns one-way submissions into real conversations and builds trust.

10+ integrations including Linear, Slack, Notion, Trello, Asana, Monday, Airtable, Basecamp, and Teamwork. Feedback flows directly into your existing workflow.

Coming soon: Integrations with Kit, Mailchimp, Mailterlite, and Hubspot. You'll be able to keep your users informed about feature progress through your existing email marketing tools and seamlessly integrate feedback loops into your app.

Multi-platform SDKs with simple integration. All SDKs are fully functional now - native support for Swift, JavaScript, React Native, Kotlin, and Flutter. Integration takes minutes, not hours.

Real-time analytics to track feedback trends, user engagement, and MRR metrics.

A bit about me: I've spent 15+ years transforming how European businesses operate through technology - ERP implementations, M&A integrations, hands-on product development. FeedbackKit isn't a side project; it's core work at Swiftly Workspace with ongoing development and support.

Pricing: Free tier to start. For the first 100 signups, use code HACKERNEWS100 for 50% off the first year.

Try it at: https://www.getfeedbackkit.com

I'd love to hear your feedback (meta, I know) and answer any questions about the platform, the tech stack, or the approach.

Comments URL: https://news.ycombinator.com/item?id=46955115

Points: 4

# Comments: 0

Article URL: https://vituperative.github.io/i2pchat/

Comments URL: https://news.ycombinator.com/item?id=46955104

Points: 1

# Comments: 0

Article URL: https://ciamweekly.substack.com/p/signing-json-web-tokens-algorithm

Comments URL: https://news.ycombinator.com/item?id=46955042

Points: 1

# Comments: 0

TL;DR: Built an ERP where AI agents execute workflows like procurement and invoice processing. Uses Cases (workspaces) and Tasks (sessions) like Claude Code. Agent completes one task, automatically triggers the next. Mac app live, looking for early users.

Mac App: https://apps.apple.com/be/app/swiftly-ai-native-erp/id6755155607?mt=12

TestFlight: https://testflight.apple.com/join/aYc16FmA

THE PROBLEM

After 8 years implementing ERPs across Europe, I kept seeing the same issue: ERP systems store data but can't execute processes. Employees still manually click through forms for invoicing, procurement, onboarding.

WHAT WE BUILT

AI agents that execute multi-step business workflows autonomously.

Example: Vendor Procurement

Case: "Source Steel Pipes - Q1 2026" Task 1: Research Suppliers → finds 3 vendors Task 2: Compare Pricing → creates comparison Task 3: Negotiate → drafts proposal, waits for approval Task 4: Create PO → auto-triggered after approval Each task has specific instructions. Status changes trigger next tasks. You approve key decisions. Everything is auditable.

HOW IT'S DIFFERENT

Workflow engines (Zapier): Rigid rules that break RPA tools: Brittle scripts AI chatbots: Just answer questions

Swiftly: End-to-end execution with approval gates

WHAT'S INCLUDED

Full ERP/CRM underneath: AP/AR, Projects, Time Tracking, Contracts, Inventory, Customer Management. The AI workflow layer sits on top.

Tech: SwiftUI apps, Swift/Vapor backend, PostgreSQL, Anthropic Claude.

We are looking to expand to all 8 major AI providers: Anthropic, OpenAI, Google Gemini, Perplexity, xAI Grok, Cohere, Mistral, DeepSeek

WHAT WE NEED

We have our first paying customer. Looking for:

- Early users willing to define 2-3 workflows with us - Feedback on the Case/Task system - SMEs (1-10 employees) in service businesses

7-day free trial. €10/seat/month after.

Links:

Production: https://apps.apple.com/be/app/swiftly-ai-native-erp/id6755155607?mt=12

Beta: https://testflight.apple.com/join/aYc16FmA

Question: What business workflow would you want an AI agent to execute end-to-end?

Comments URL: https://news.ycombinator.com/item?id=46955034

Points: 3

# Comments: 0

Article URL: https://www.not-ship.com/america-isnt-exceptional-its-the-exception/

Comments URL: https://news.ycombinator.com/item?id=46955032

Points: 2

# Comments: 0

Article URL: https://www.benkuhn.net/abyss/

Comments URL: https://news.ycombinator.com/item?id=46955019

Points: 1

# Comments: 0

Article URL: https://weaviate.io/blog/limit-in-the-loop

Comments URL: https://news.ycombinator.com/item?id=46955012

Points: 1

# Comments: 0

Article URL: https://shadows.ooousay.com/

Comments URL: https://news.ycombinator.com/item?id=46955005

Points: 2

# Comments: 0

Article URL: https://stripe.dev/blog/ai-agents-terraform-stripe-infrastructure

Comments URL: https://news.ycombinator.com/item?id=46954999

Points: 2

# Comments: 0

Article URL: https://andonlabs.com/blog/evolution-of-bengt

Comments URL: https://news.ycombinator.com/item?id=46954974

Points: 1

# Comments: 1

So, this idea maybe looks like DesignJoy.

I was freelancing on some platforms like Upwork and Fiverr to help my clients with their web design and UI/UX design needs.

What i noticed was interesting: Some clients having lot of issues with their design needs, not about the quality of the design. But they feels like they need a quick deliverables designs.

The real issues was continuity.

They struggled to find new freelancers who had the same mindset, communication style, understanding of their product. As their products grew, they need more designs needs, especially brand design, blog thumbnail design, OG Images, etc.

Brand works usually meant higher costs, If they switched on who specialized in branding, it would be more expensive and re-adapt with the freelancer. And if they stayed with me, they insist to have the predictable costs.

I tried to do add-on services instead of charging per project. This maybe has lower profit for me as a designer, but i'm happier since i can help my clients with their works.

After few months, i found out DesignJoy. The model make sense, so i build my own called Designgud, i tried to improve things out a bit more the pricing etc. And optimized it based on what my existing clients actually needed.

After launched few weeks, i got a few clients that still contribute with me until now. Interestingly, few clients came from Fiverr and Upwork. I don't understand how they can get my contacts, i think they found me on my X.

Comments URL: https://news.ycombinator.com/item?id=46954959

Points: 1

# Comments: 1

Article URL: https://github.com/lout33/claude_life_assistant

Comments URL: https://news.ycombinator.com/item?id=46954943

Points: 2

# Comments: 3