Vaskin Kissoyan

AI Engineer and Researcher

[Lucas @OnchainLu]

agents are doing things today that weren't possible even 3 months ago

in the graph below:
1. nodes are paid APIs that have been called by agents
2. edges are only drawn between APIs that have actually been composed together by agents
3. data is from apr 31 to june 3

i put this together to illustrate what's been happening since we started giving agents access to money

u can think of each of these connections as a new use case

as more merchants make their services accessible to agents, this network will grow, the connections will multiply, and use cases will emerge that will surprise everyone, including the people who built the APIs

this is open agentic commerce.

[Vaskin Kissoyan @VaskinK]

@cuysheffield The way it can build and manage it's own skills and how it auto prunes / manages them over time. That's the real difference.

[Cuy Sheffield @cuysheffield]

Visa is for vibe coders.

Anyone can spin up an app and any API can become a merchant.

Starting today we’re onboarding x402 and MPP merchant endpoints into Visa CLI so APIs, LLMs, data products, and dev tools can be discovered and purchased by verified agents with cards on file.

If you’re building an x402 or MPP endpoint, or want Visa to help you stand one up, sign up below.

We��re also opening Visa CLI to initial users in the US. DM me for an invite.

[Vincent Koc @vincent_koc]

🥳 My 6 week old side project just surpassed 2.5mill downloads on npm.

🧃TokenJuice was started as a concept to help reduce token use with tools, as a TypeScript SDK and in an MIT licence for all. It now supports close to 100 different coding IDEs and harnesses.

Crazy stuff!

[Andrej Karpathy @karpathy]

Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.

[Vincent Koc @vincent_koc]

🦞 Just joined the house of crustacean aka @openclaw as an open-source maintainer and will be helping improve observability and security making sure your lobsters arive safe and fresh. Thanks @steipete

[Aakash Gupta @aakashgupta]

Google and Microsoft just co-authored the spec that turns every website into an API for AI agents. The second-order effects here are massive.

Right now, browser agents work by taking screenshots, parsing the DOM, and guessing which buttons to click. It works about as well as you’d expect. Fragile, expensive, slow. WebMCP replaces all of that with a single browser API: navigator.modelContext. Websites register structured tools directly in client-side JavaScript. The agent reads a menu of available actions, calls them, gets structured data back. No scraping. No backend MCP server in Python or Node. The tools run inside the browser tab and share the user’s existing auth session.

Early benchmarks show ~67% reduction in computational overhead compared to visual agent-browser interactions. Task accuracy around 98%.

The second-order effect is where this gets wild. Today, when a browser agent visits two competing airline sites, it’s guessing at both interfaces equally. Once WebMCP adoption spreads, the site that exposes structured tools gives the agent a clean, reliable path to complete the task. The site that doesn’t forces the agent to fumble through the UI. Agents will prefer the cheaper path. Every time.

This means “Agent Experience Optimization” becomes a real discipline. Tool naming, schema design, description quality. Sound familiar? It’s the same shift that happened when meta descriptions and structured data became optimization surfaces for search engines. Except this time, the traffic source isn’t Google’s crawler. It’s every AI agent on the internet.

Bots already make up 51% of web traffic. Google just gave them a front door.

[Daniel San @dani_avila7]

New feature in Claude Code 2.1.14 just dropped!

You can now search and install plugins from the marketplaces installed in your current Claude Code session.

This is huge if you’re building plugins on top of Claude Code’s marketplace layer (Skills, Agents, Hooks, etc).
How it works:

- Run /plugin
- The official Claude marketplace is installed by default
- Use the search bar to find the plugin you want
- Select one or multiple plugins with space, then press i to install
- Go to the Installed tab to browse and enable them

With the exponential growth of Skills and Agent-based components running in the CLI, improving plugin discoverability is a big win.

Pretty sure more marketplace-related features are coming

[Saoud Rizwan @sdrzn]

Cline v3.4 is out 🚀 Introducing MCP Marketplace! Discover and install the best MCP servers right in the extension, where Cline handles all the setup. We’ve also added mermaid diagrams to Plan mode, new @​terminal and @​git mentions in chat, and visual improvements to checkpoints!

[Vaskin Kissoyan @VaskinK]

https://t.co/X0IYTqI85r

[Vaskin Kissoyan @VaskinK]

🎯 Breakthrough: Run Multimodal AI in Your Browser with JanusPro!

I've tested DeepSeek's latest innovation that's flying under the radar:

🚀 Full multimodal AI in just 1B parameters
💻 Runs directly in your browser via WebGPU
🔒 Complete offline operation - no cloud needed
🎨 8-second image generation on consumer GPUs
📊 Robust image comprehension capabilities

Just tested image generation and comprehension - the results are impressive for edge deployment. Full technical breakdown in the article.

https://t.co/P8uaRbF9xk

[Colin McNamara @colinmcnamara]

There's a fascinating divide in workplace AI adoption that goes deeper than the numbers suggest. As someone who lives and breathes AI-augmented workflows, let me share what I'm seeing... 🧵

[Vaskin Kissoyan @VaskinK]

I've analyzed the groundbreaking SYNTHETIC-1 project, which is redefining truly open source AI:

https://t.co/ylMwQfiKab

🧠 Building the world's largest verified AI reasoning dataset - already 1M+ samples and growing
🌐 Global collaboration spanning Paris to San Jose
⚡ Innovative GPU business model: pay-to-contribute like modern SETI@home
🔓 MIT licensed for commercial use

What makes this special? While others focus on just open weights, SYNTHETIC-1 opens up the entire training process. Each contribution is meticulously verified through testing and re-testing of reasoning steps, creating a dataset of unprecedented quality and scale. The genius lies in its business model - contributors pay for GPU time, funding the very infrastructure that makes this massive open-source effort possible.

Currently at 51.64% of its 2M sample goal, this project demonstrates how distributed efforts can drive AI innovation while maintaining true open source principles.

Track the progress:
https://t.co/oE6MecsBt2

#OpenSourceAI #MachineLearning #ArtificialIntelligence #Innovation #TechNews

💻 Book A Call If You're Interested in Implementing AI Agents Into Your Business

[Rob Whelan @Rob_Whelan]

The NRC has written a lot about AI. Much of the writing is about what to do if someone uses AI under their regulatory jurisdiction (e.g. a nuclear power plant uses AI to augment decision-making).

In 2022/3 it kicked into higher gear, post-chatgpt.

Here's a summary
1/8

[Vaskin Kissoyan @VaskinK]

DeepSeek-R1- a local thinking model https://t.co/ZJfzZYSZ18

[Vaskin Kissoyan @VaskinK]

Just wrote an opinion piece / introduction to Model Context Protocol: The Web Browser Era of AI Tools 🚀 https://t.co/iVAK8TjW3O