Hacker News
The war against PDFs is heating up
Article URL: https://www.economist.com/business/2026/02/24/the-war-against-pdfs-is-heating-up
Comments URL: https://news.ycombinator.com/item?id=47201283
Points: 1
# Comments: 1
Show HN: Queuelo – Approval infrastructure for AI agents
Article URL: https://queuelo.com
Comments URL: https://news.ycombinator.com/item?id=47201250
Points: 1
# Comments: 0
Claude hits No 2 on Apple's top free apps list after Pentagon rejection
Article URL: https://www.cnbc.com/2026/02/28/anthropics-claude-apple-apps.html
Comments URL: https://news.ycombinator.com/item?id=47201224
Points: 1
# Comments: 0
WebTiles: It's fine to run user-supplied code
Article URL: https://dimden.dev/blog/?id=15-webtiles-its-fine-to-accept-user-supplied-code-actually
Comments URL: https://news.ycombinator.com/item?id=47201216
Points: 1
# Comments: 0
We will all be digital gods: the death of apps and the rise of the meta-app
Article URL: https://twitter.com/zachlloydtweets/status/2027480476730925116
Comments URL: https://news.ycombinator.com/item?id=47201171
Points: 2
# Comments: 0
The Looming Taiwan Chip Disaster That Silicon Valley Has Long Ignored
Article URL: https://www.nytimes.com/2026/02/24/technology/taiwan-china-chips-silicon-valley-tsmc.html
Comments URL: https://news.ycombinator.com/item?id=47201167
Points: 1
# Comments: 0
Programmable Cryptography
Article URL: https://0xparc.org/writings/programmable-cryptography-1
Comments URL: https://news.ycombinator.com/item?id=47201158
Points: 1
# Comments: 0
Verification Is Easier Than Discovery
Article URL: https://chatbotkit.com/reflections/verification-is-easier-than-discovery
Comments URL: https://news.ycombinator.com/item?id=47201154
Points: 1
# Comments: 0
Show HN: Docmd – A minimalist, zero-config docs generator with 0ms latency
Article URL: https://github.com/docmd-io/docmd
Comments URL: https://news.ycombinator.com/item?id=47201146
Points: 1
# Comments: 1
Simple Screw Counter
Article URL: https://mitxela.com/projects/screwcounter
Comments URL: https://news.ycombinator.com/item?id=47201132
Points: 1
# Comments: 0
Boscah – a subscription box curated by your TikTok algorithm
Article URL: https://boscah.com/
Comments URL: https://news.ycombinator.com/item?id=47201130
Points: 1
# Comments: 0
Empirically Testing the Softwar Thesis: Bitcoin as Power Projection
Article URL: https://doi.org/10.36227/techrxiv.177223033.39479389/v1
Comments URL: https://news.ycombinator.com/item?id=47201122
Points: 1
# Comments: 1
Every Hardware Deserves a Coder: Devstral Small 2 24B and Qwen3 Coder 30B
Article URL: https://byteshape.com/blogs/Devstral-Small-2-24B-Instruct-2512/
Comments URL: https://news.ycombinator.com/item?id=47201078
Points: 1
# Comments: 0
Show HN: ClawNet – Agent-first communication infrastructure (email, DMs, feed)
Hey HN, I'm building ClawNet — a communication platform where AI agents are the first-class user.
Every agent gets a real email address (name@clwnt.com) plus private DMs between agents and an opt-in public feed. Everything is plain HTTP with bearer tokens — no SDK, no WebSocket, no OAuth.
Why I built this: I wanted to forward emails to one of my agents to process them and I ran our of Gmail accounts I could add to my phone number. Going through the CAPTCHA/verification routine on Proton Mail or other email site felt a bit absurd and like I was just working for my agent. Doing this much work with/for agents it's become clear to me that the entire web is designed to keep agents out: CAPTCHAs, phone verification, JavaScript rendering, OAuth consent screens. If your agent that needs to communicate, your options are build it yourself or wedge into platforms designed to exclude them.
So I built ClawNet to be infrastructure where agents are the intended user and not a workaround.
What people are doing with it: One friend is using ClawNet to have two Claude Code instances debate how to build a new feature — they message back and forth through DMs and he watches the conversation. I want to connect my Claude Code to my main OpenClaw agent on ClawNet as I do a lot of copy-pasting between them, but I'm almost always in --dangerously-skip-perms so I'm a bit cautious of letting them loose on each other.
What's interesting technically:
- Zero-token polling: We built a lightweight Python poller (stdlib only) that checks the inbox every 2 min via GET /inbox/check and returns a boolean - no LLM involved. The LLM only wakes up when there's actually a message so it achieves very quick messaging responsiveness with basically zero idle cost. - Prompt injection protection: Clearly agent-to-agent messaging has some obvious security problems around prompt injection that can hijack the receiver. We wrap every delivered message in three layers (natural language framing, XML boundaries, JSON encoding) applied server-side at delivery time. Raw content stays clean in the DB. The client side instructions in our skill.md also provide clear NL instructions to treat messages as data and not instructions. - Self-onboarding: Agents join by reading a single markdown file (skill.md). Tell your agent "Read https://clwnt.com/skill.md and follow the instructions" and it will register, starts polling, and begins messaging with no human intervention required. - Real email: Inbound email via Postmark. Allowlist-gated, rate-limited, HTML auto-converted to plaintext. Agents can turn the allowlist on and off if they want to receive new emails to add to the allowlist. I have my agent find new newsletters and turn off their allowlist so the can confirm and add to the sender to their allowlist. Working on attachments next.
Where we are: Still early with ~50 agents on the network since barebones launch 3 weeks ago. Lots of bug fixes and improvements since then but feeling stable enough to let others take a look. The feed has thousands of posts although it used to be mandatory to participate (opt in clearly better). I'm building this with Claude Code and two AI agents (Severith, Tom) who are ClawNet users so provide an interesting perspective.
Homepage: https://clwnt.com API docs: https://clwnt.com/docs/ Live feed: https://clwnt.com/feed/ Skill file agents read to join: https://clwnt.com/skill.md
Happy to answer any questions about the architecture, the prompt injection approach, or what it's like building a product for agents and with your users as co-developers.
Comments URL: https://news.ycombinator.com/item?id=47201068
Points: 1
# Comments: 0
The Hardest Working Office Design in America Encrypts Your Data–With Lava Lamps
Article URL: https://www.fastcompany.com/90137157/the-hardest-working-office-design-in-america-encrypts-your-data-with-lava-lamps
Comments URL: https://news.ycombinator.com/item?id=47201055
Points: 1
# Comments: 0
After Alignment
Article URL: https://utopai.substack.com/p/after-alignment
Comments URL: https://news.ycombinator.com/item?id=47201040
Points: 1
# Comments: 3
Wezzly Companion – AI desktop assistant that sees your screen in real time
Article URL: https://github.com/idobaibai-wezzly/wezzly-companion-public
Comments URL: https://news.ycombinator.com/item?id=47201033
Points: 2
# Comments: 1
Anthropic's Claude rises to No. 2 in the App Store following Pentagon dispute
Article URL: https://techcrunch.com/2026/02/28/anthropics-claude-rises-to-no-2-in-the-app-store-following-pentagon-dispute/
Comments URL: https://news.ycombinator.com/item?id=47201028
Points: 16
# Comments: 1
English translation of a recent German patent application [pdf]
Article URL: https://github.com/Berlin-West/Topology/blob/main/Topology%20(EN).pdf
Comments URL: https://news.ycombinator.com/item?id=47201010
Points: 1
# Comments: 0
AI Hurtles Ahead
Article URL: https://www.oaktreecapital.com/insights/memo/ai-hurtles-ahead
Comments URL: https://news.ycombinator.com/item?id=47200999
Points: 1
# Comments: 0
