@Techjunkie_Aman Thanks for sharing 🙏 this is very useful
English
Wouter Vernaillen 𐤊
2.3K posts
Wouter Vernaillen 𐤊
@vernaillen
Freelance Full Stack Developer, Nuxt, Java and 𐤊 #kaspa https://t.co/2y1PJPwAsQ Ecstatic Dance dj Trance Dance facilitator
Belgium Katılım Aralık 2007
1K Takip Edilen1.2K Takipçiler
Most Brave Browser users never open this:
brave://flags
That’s where the real power is.
Here are the best hidden Brave flags you should enable 🧵
English
Wouter Vernaillen 𐤊 retweetledi
Last month in Koh Tao I also had this extraordinary experience to be surrounded by a school of sardines.
The Best@Thebestfigen
A woman experienced being captured when she went to photograph a school of sardines.
English
Vibe coding is fun, but without a human in control it quickly turns into AI slop.
That's why I like the term Agentic Engineering — AI builds, but the human owns architecture, quality and correctness.
ai.plainenglish.io/vibe-coding-is…
English
Wouter Vernaillen 𐤊 retweetledi
Opus 4.6 is a fantastic model. Claude Code is an excellent harness. The Max plan is still a steal. I'm still shipping like crazy.
I'm with the 99% of builders who'd rather be building than complaining about Anthropic's (frankly, pretty reasonable) terms.
English
Wouter Vernaillen 𐤊 retweetledi
#WarpCore just quietly did something most people aren’t even catching yet… it successfully tested integration with #Fedwire and SEPA, meaning Kaspa is now proving it can plug directly into real-world banking rails, not just live inside crypto. This isn’t hype, this is infrastructure. End-to-end payments, compliance checks, and settlement all working on Kaspa’s #blockDAG, and now it’s marked ready for integration. If this moves from testing to actual usage, you’re not looking at another speculative chain, you’re looking at a network stepping into global $finance territory. The market might sleep on this now, but this is exactly how real adoption starts. #Kaspa
English
I integrated vue-kaspa into my Kaspa analytics dashboard (Nuxt app) — works great!
One thing I ran into: the require-corp COEP header blocks cross-origin resources (CDN images, Vercel Blob Storage) that don't set CORP headers.
Switching to credentialless fixes it — still enables SharedArrayBuffer but doesn't block resources.
Opened an issue with the fix: github.com/furatamasensei…
English
So you might be wondering, what does that Vue Kaspa is all about.
Vue is the 3rd largest community after its brother; React and Angular. And that tool is my main stack.
When I first developing @kaswaypos, it uses Nuxt. And I saw a lot of issues that comes from the WASM alone.
English
The thing that sets Kaspa apart from previous "Bitcoin competitors" is that it doesn't compromise on the ethos to gain performance. Most fast chains got speed by centralizing (fewer validators, PoS, delegated consensus).
Kaspa kept PoW and decentralization while solving throughput at the protocol level.
Whether it succeeds as a competitor is a different question — network effects and Bitcoin's first-mover advantage are enormous.
But on pure technical + philosophical alignment with Bitcoin's original vision, Kaspa is the most credible attempt I'm aware of.
apsa𐤊aspaFuture@mookdog2001
I asked Claude a simple question about $kas (Kaspa): Has there been anything as good as Kaspa to compete with Bitcoin on the same ethos? Response: "Honestly, no. If you’re filtering for coins that share Bitcoin’s core ethos — proof of work, fair launch, no premine, no VC backing, no founder allocation, community-driven — the list is very short and Kaspa stands alone as a serious modern contender. Here’s the honest scorecard of everything that’s tried: Litecoin (2011) — Fair launch, PoW, no premine. But it’s basically Bitcoin with faster blocks and a different hash. It never offered a fundamentally new technical thesis. It’s the “silver to Bitcoin’s gold” narrative, which has slowly faded into irrelevance. Monero (2014) — Fair launch, PoW, no premine, genuinely innovative with ring signatures and default privacy. Arguably the strongest Bitcoin-ethos competitor before Kaspa. But privacy focus has made it a regulatory target — delisted from major exchanges, which limits its growth ceiling significantly. Grin (2019) — MimbleWimble protocol, fair launch, no premine, no founder reward. Technically interesting but the linear emission schedule (no supply cap) killed the investment thesis, and the project has struggled with funding and developer retention. Dogecoin (2013) — Fair launch technically, but created as a joke with infinite supply. Not a serious competitor on ethos despite the market cap. Kaspa (2021) — This is where it gets interesting. Fair launch, no premine, PoW, but with a genuinely novel consensus mechanism (GHOSTDAG/blockDAG) backed by peer-reviewed academic research from Sompolinsky. It’s the first project that credibly argues “we can do what Bitcoin does, but faster, without sacrificing the things that matter” — decentralization, PoW security, fair distribution. The key differentiator is that Kaspa isn’t just another Bitcoin fork or minor tweak. It’s built on a fundamentally different data structure with real academic pedigree. That’s rare. Most “Bitcoin competitors” either compromise on the ethos (premine, VC) or don’t bring enough technical novelty to justify existing. That said, “as good” depends on your criteria. Nothing matches Bitcoin’s Lindy effect, network hashrate, or regulatory clarity. Kaspa has the best combination of ethos + innovation of anything that’s come along, but it’s still very early and unproven at scale."
English
@furatamasensei @kaswaypos Curious about the WASM issues you ran into with Nuxt — was it SSR-related?
WASM modules can be tricky with server-side rendering.
English
@stelis @pootlepress Interesting take, though I'd lean toward GraphQL over REST for content — better query flexibility and type safety. The decoupling is the key part, regardless of the transport layer.
English
@pootlepress Headless is the way to go. Let AI build all templates in PHP. Pull in content from REST API. Now you can install WordPress anywhere... under a different folder, different subdomain, or different domain.
English
Been thinking a lot about what WordPress looks like in an AI-first world…
Would love your thoughts on this one 👇
Jamie Marsland - Head of WordPress YouTube ❤️@pootlepress
English
@DiMaNachom WPGraphQL is solid and production-ready, but you're right that official support would help.
One of the gaps imo is preview — getting live preview working in a decoupled frontend is still more work than it should be.
English
WordPress Studio CLI 🖥️
Se agradece. Pero el elefante sigue ahí:
No hay integración headless oficial con Next o Astro. La REST API existe, sí. El ecosistema real lo armó la comunidad. 👀
¿Cuándo construyen ese puente?
WordPress.com@wordpressdotcom
Early access! WordPress Studio now has an independently installable CLI. One command to install: npm install -g wp-studio And another to run WordPress locally, no desktop app needed: npx wp-studio It works on Mac, Windows, and Linux and makes it easy for your AI coding tools to interact with Studio. Sync, import, export, and more are coming soon. 👀 Free to install and use.
Español
Anthropic's harness design paper tackles coherence and self-eval but not your core point: handoff summaries are lossy. The writing agent works from an abstraction, not the source.
Curious if anyone has found a good pattern for preserving detail across agent boundaries without blowing up context.
anthropic.com/engineering/ha…
English
I don't know how other models handle context limits, but Claude Code does something that's been on my mind.
When the main agent spawns sub-agents in parallel to plan, investigate the codebase, those sub-agents actually read the details, the edge cases, the subtle dependencies.
And then they compress everything into summaries. And the main agent, the one that actually writes the code, executes based on those shallow outputs, not the real thing.
The agent that plans never touches the code.
The agent that reads the code never writes it.
No wonder AI coding tools frequently introduce bugs and regressions. The implementation is always one abstraction layer away from the truth.
I imagine this is the hardest problem for engineers building AI agents today:
how do you keep the model effective when context windows force you to choose between seeing the full picture and actually fitting it in memory?
English
Anthropic just published how they break through the ceiling of long-running AI coding agents.
The problem: single agents lose coherence on long tasks and — just like developers reviewing their own code — they rate their own work too favorably.
The fix: a 3-agent architecture inspired by GANs (Generative Adversarial Networks), where two systems compete to improve each other.
Planner → decomposes the task
Generator → writes the code
Evaluator → reviews it independently
The generator and evaluator push each other — like a dev and a strict code reviewer making each other better through friction.
Other key insight: context resets beat compaction. A fresh agent with a structured handoff outperforms one dragging along a summarized history.
This pattern is already emerging in tools like OpenClaw — decompose, hand off state, evaluate independently.
anthropic.com/engineering/ha…
English
Wouter Vernaillen 𐤊 retweetledi
Kaspa is the most underrated infrastructure play for the agentic future and I'll explain why
Agents are parallel by nature. They spawn sub-agents, fire concurrent tasks, run multiple workflows at once. Traditional blockchains are serialized. One block. One leader. One queue. That's a fundamental architectural mismatch at the protocol level. Kaspa's blockDAG processes blocks in parallel and includes them all. The settlement layer is structurally isomorphic to how agents actually operate.
Agents don't wait. A human tolerates a 10 minute Bitcoin confirmation. An agent pipeline that pauses 10 minutes is a broken pipeline. Machine-speed finality isn't a nice to have. It's a requirement.
The agentic economy runs on micro-payments. An agent paying another agent for a single API call. A single inference. A single data lookup. If fees eat the transaction the economics collapse. Kaspa fees are essentially zero. The payment rail doesn't distort the value exchange.
Agents can't KYC. They don't have SSNs. They just need an address and a balance. Kaspa doesn't care if the sender is human or not. It just processes the transaction. That's the only viable model for autonomous economic actors.
Here's the underrated one. In Kaspa's design the fee market is a real-time signal reflecting actual network demand. Agents can read fee pressure and make intelligent routing decisions. Throttle. Batch. Reroute. The protocol itself becomes an information layer. Not just a payment rail.
Most chains were designed for humans transacting occasionally. Kaspa was designed for throughput at scale with minimal trust assumptions. That's also a perfect description of what an agentic economy needs.
Want to see it in action right now? Go to kaspadrome.xyz, send 10 cents worth of KAS to the address on the page and open an explorer. Watch TPS surge in real time because of your 10 cent contribution. Someone built a public stress test toy using a modified version of the world record bot and just left it running for anyone to play with. No token. No hype. Just proof.
Odie@PoW_Odie
If you want to surge Kaspa TPS to around 3000 TPS. Send a couple KAS to this address. kaspa:qzg49fejq8jlc29jd86746zye3ta2fgvd36hkv8k4k8r3ur6gks5y0fa938dt Then go view an explorer and watch TPS fly because of your 10 cent contribution. I don’t know who is behind kaspadrome.xyz … It definitely uses a modified version of the bot used for the world record. Amazing work. #Kaspa
English
Anthropic just added code review to Claude Code because AI-generated PRs were piling up faster than humans could review them.
Think about that. The bottleneck isn't writing code anymore. It's reviewing it.
We're speedrunning from "AI writes code" to "humans can't keep up with AI's output."
The most valuable skill in 2026 isn't prompting. It's reviewing.
techinformed.com/anthropic-adds…
English
@johann_sath This is step 1. Step 2: organize those daily files into a tiered L0/L1/L2 system so your agent doesn't have to load everything at startup.
We went from ~3K tokens to ~200 tokens at boot. Wrote about it here: github.com/vernaillen/rez…
English
openclaw SOUL.md that actually works
here's the 5-line snippet i add to every project:
Voice
- never apologize
- never say "great question"
- skip the preamble, start with the answer
- if unsure, say "i don't know"
- match the user's tone, not a customer service bot
copy paste that. your ai stops sounding like a linkedin post instantly.
English
Been running a similar SOUL.md for months. The biggest unlock for me was adding "have opinions" and "be resourceful before asking."
A bot that just says "I don't know" is honest. A bot that searches, reads the file, checks the context, and THEN says "I don't know" — that's really useful.
github.com/vernaillen/rez
English