whackur

whackur

Hands-on notes on AI, blockchain (Web3 · Solidity), local LLMs, and physical AI.

Know Your Agent (KYA): AI Agent Identity, the Standards Race, and What Is Actually On-Chain

When you open a bank account you show an ID. When you sign up for an exchange you take a selfie. That procedure is KYC (Know Your Customer), and since the FATF was founded in 1989 it has been the standard way to keep illicit money out at the entry point of the financial system. Now the entity signing contracts, sending payments and swapping tokens on a DEX is increasingly not a person but an AI agent, and the same question comes back in a new form. The agent that just sent a payment request to my API: who is it, who built it, and what was it actually authorized to do? The trust layer that answers that question is being called KYA (Know Your Agent).

Read More

The A2A (Agent-to-Agent) Protocol: How Agents Delegate to Each Other, and Where Payments Plug In

Every time this blog has covered x402, UCP and MPP or Know Your Agent, the term “A2A” showed up as an assumption. An agent hands work to another agent, pays for it, checks who the counterparty is. All of that sits on a lower layer: how do two agents talk in the first place? The attempt to standardize that layer is the Agent2Agent (A2A) protocol.

Read More

gs-quant: How Goldman Sachs Open-Sourced Its Quant Code, and Where the Line Is

Tell someone that an investment bank’s trading code is on GitHub and you usually get one of two reactions: “why would they do that” or “they kept the important parts, surely.” goldmansachs/gs-quant is a case where both reactions are half right. The repository was created in December 2018, passed 12,000 stars as of August 2026, and had a push on the very day I checked. The README also states plainly that you need to be an institutional client of Goldman Sachs to use the pricing and risk APIs.

Read More

The Failure a Pass Rate Hides: How the YouTube Ads Team Runs Production Evals

The idea that a well-written prompt makes an agent behave holds up right until the demo ends. In production, the same prompt and the same input produce different results run to run. A case that passed yesterday fails today, and one that failed yesterday passes. Deciding whether that system is ready to ship takes measurement, not a feeling.

Read More

Don't Ship Agent Skills Without Evals: Philipp Schmid's Testing Method and SkillsBench

If you use a coding agent like Claude Code, Gemini CLI, or Codex, you eventually end up writing skills: Markdown files that hand the agent your team’s coding conventions, a specific SDK’s usage patterns, or a deployment procedure. The problem is that almost nobody tests them. We would never ship code without tests, yet skills, which directly change how an agent behaves, get shipped after a few manual runs and a gut-level “looks fine.”

Read More

GPT-5.6 Sol, Terra, and Luna: A Model Routing Guide for Coding Agents

A claim like “Luna Max beats Terra High” circulates around GPT-5.6 discussions, and it does not hold up. Sol, Terra, and Luna are separate models, each a different capability tier. Max, high, xhigh, and ultra are settings within a given model that control how much reasoning time it uses and how many agents run in parallel. Collapsing a tier name and a settings name into one ranking compares two different axes as if they were one.

Read More

On-Policy Distillation: Closing the Gap Between RL and SFT

The two standard post-training methods each leave a gap. Supervised fine-tuning (SFT) has the student imitate sequences a teacher already produced, but training happens on states the student may never actually visit, so errors compound over long generations. Reinforcement learning (RL) samples from the student’s own rollouts, which fixes that mismatch, but the reward is usually a single bit or two per episode. A post Thinking Machines Lab published in October 2025 proposes combining the two: sample trajectories from the student, then have a strong teacher score every token in that trajectory.

Read More

AI Self-Improvement Starts Outside the Model

When people talk about AI self-improvement, the usual image is a model rewriting its own weights. Lilian Weng’s July 4, 2026 post on Lil’Log argues the near-term version looks different. The recursive self-improvement (RSI) we can actually observe today shows up first in the system around the model, not inside it. That system is what she calls the harness. This post walks through her argument: what a harness is, where the optimization target is moving, what a striking SWE-bench number actually means, and what breaks if you get the risk boundaries wrong.

Read More

Bitcoin, Ethereum, and the Quantum Computing Question

On March 30, 2026, a joint paper from Google Quantum AI, the Ethereum Foundation, and Stanford shook up the crypto community. It estimated that the physical qubit count needed to recover a private key on secp256k1 (the elliptic curve behind Bitcoin and Ethereum signatures) had dropped nearly 20x from the previous best estimate. A threat that used to sound like “millions of qubits, decades away” now reads as “under 500,000, maybe within this decade.”

Read More

Apify x402 and Coinbase Wallets, in Depth: How Agents Buy Web Automation Tools Directly

Apify, the web scraping platform, announced x402 support: AI agents can now run more than 20,000 Apify Actors by paying with USDC on Base, with no Apify account and no API key. Apify frames the prior x402 ecosystem as roughly 2,000 endpoints, so by its own count this integration expanded the paid-tool surface by about 10x.

Read More

Hermes Agent v0.18: When a Self-Improving Agent Gets MoA

Nous Research’s open-source agent Hermes Agent shipped v0.18.0 on July 1, 2026. The release notes call it “The Judgment Release.” Reading this as a routine feature update misses the point. The official docs describe Hermes Agent as “the self-improving AI agent,” and the project is built around a loop that accumulates memory and skills the more it gets used. This post covers how that loop was refined in v0.18, and what role Mixture of Agents (MoA), now a first-class model choice in the same release, plays inside it.

Read More

Pulling Encryption Keys Out of Passkeys with the WebAuthn PRF Extension

Passkeys show up on login screens everywhere now. What’s less well known is that the same authenticator holding your passkey can also hand out key material for encryption, not just authentication. The prf extension in WebAuthn is that path. This post covers what PRF actually returns, how it relates to the CTAP2 hmac-secret extension, what you still have to build yourself before it’s usable for real encryption, and where device and platform support stand as of 2026.

Read More

TradingAgents: Reading the Paper and Code Behind an LLM Trading Desk

Up front: this is not investment advice, and nothing here recommends buying or selling anything. It is a read of how you organize LLM agents into a single decision, looked at from the research and the code. Trading is just the domain the design happens to target.

Read More

AI Agent Commerce and Coordination: ERC-8183, ERC-8226, ERC-8001, ERC-8041

The ERC-8004/8126/8196 trust stack defines three layers: agent identity, security verification, and policy-constrained execution. It identifies an agent, scores its risk, and keeps it acting within the limits an owner has set. All of that answers one question: can I trust this agent to act on my behalf?

Read More

Chainlink CCIP EVM Contracts: Architecture from Source Code

EVM chains are isolated by design. A contract on Arbitrum has no native way to call a contract on Base, and sending USDC from Ethereum Mainnet to Polygon is not a built-in operation. Cross-chain bridges filled that gap for years, but the category has seen repeated security incidents that highlight how wide the attack surface is.

Read More

Circle CCTP V2 EVM Contracts: A Source-Code Walkthrough

Bridge designs for moving USDC between chains come down to two families. Lock-and-mint parks the original token in a contract on the source chain and issues a wrapped version on the destination. Burn-and-mint destroys the token on the source chain outright and issues the same amount fresh on the destination. Circle’s CCTP (Cross-Chain Transfer Protocol) is the second kind.

Read More

ERC-8004 Agent Reputation: On-Chain Registration and Lookup

A prior post on this blog covered the ERC-8004/8126/8196 trust stack: three Ethereum registries that handle agent identity, reputation, and validation. If you want the conceptual overview first, start there. This post is about something narrower: how feedback actually gets written to the Reputation Registry, how to read it back, and what services are available today to browse and index that data.

Read More

Agentic Payments in June 2026: x402, UCP, and MPP Implementation Progress

Agentic payments in June 2026 moved from “can an agent pay” toward “who authorized the payment, how is it verified, and how is it traced.” The major open protocols (x402, UCP, MPP/pay.sh, and ACP) each had meaningful implementation changes. All figures and commit details below come from GitHub API, PyPI, and npm registry checks.

Read More

AI x Blockchain in 2026: Bittensor, DePIN, and Agent Finance

“AI plus blockchain” spent most of the early 2020s as a marketing phrase. By mid-2026, it has organized into four distinct layers, each with its own competitive dynamics. Which projects have staying power, and what are the actual evaluation criteria at this point?

Read More

Figure: From HELOC Lender to Blockchain-Native Capital Markets

Two companies go by the name Figure. One is Figure AI, the humanoid robotics company. The other is Figure Technology Solutions, which is what this post is about. They are unrelated.

Read More

Harness-1: Teaching Search Agents to Offload State

Search agents and the state problem A search agent is an AI system that answers a question by iterating through multiple searches. Unlike a one-shot retrieval lookup, it reads intermediate results, adjusts its search strategy, compares candidate documents, and checks whether specific claims are actually supported by what it found. Tasks like analyzing financial filings, tracing multi-hop facts across sources, or interpreting complex regulations need this kind of iterative work. A single query won’t get you there.

Read More

Lighter: A ZK-SNARK Orderbook DEX on Ethereum

Decentralized exchanges come in two main forms. AMMs (Automated Market Makers) like Uniswap set prices algorithmically from pool ratios and work as permissionless swap venues. Orderbook DEXes match individual buy and sell orders by price and time priority, the same way centralized exchanges like Binance or Coinbase operate. Orderbooks enable limit orders, tighter spreads with active market-making, and more precise price discovery. AMMs do not.

Read More

LLM Observability Without LangSmith: Five Open-Source Tools Compared

At some point in building LLM applications or agents, you need to know why a call failed, what the tool invocation looked like, or why the agent got stuck in a loop. LangSmith, LangChain’s commercial observability platform, has been the default answer for this: it covers trace visualization, prompt versioning, and evaluation in one place. Its usage-based pricing and cloud-hosted architecture are where teams start looking for alternatives. Traces carry raw user inputs and internal prompts, so shipping that data to an external SaaS is itself a problem for plenty of organizations.

Read More

The AI Agent Trust Stack: ERC-8004, ERC-8126, and ERC-8196

When an AI agent acts on behalf of a user (spending funds, calling contracts, accessing paid APIs), the obvious question is: how do you know this agent is safe? The Ethereum community has three standards that address different layers of that question. They don’t solve the whole problem, but they’re building toward a coherent stack.

Read More

DeepSWE: A Benchmark for Long-Horizon Coding Agents

SWE-bench has been the default coding-agent leaderboard for a while, but it has well-known weaknesses. Most tasks come from existing public issues and PR patches, so a high score might partly reflect memorization. Most tasks are also single-file bug fixes, which is not representative of the multi-file, long-horizon work that a coding agent does in practice.

Read More

Mixture of Agents: How Layering Open-Source LLMs Beat GPT-4 Omni

Instead of scaling a single model up, you can stack multiple models in layers and have each one refine the previous layer’s output. Together AI’s research team formalized that approach in June 2024 as Mixture of Agents (MoA) in arXiv:2406.04692. Using only open-source models, their MoA configuration scored 65.1% on AlpacaEval 2.0, versus 57.5% for GPT-4 Omni.

Read More

Open Knowledge Format: A Shared Vocabulary for Agent Knowledge

When AI agents fail in production, the model is often not the problem. The missing context is. Table schemas, metric definitions, runbooks, join paths between systems, and API deprecation notices are scattered across catalog vendors, internal wikis, code comments, and personal notes. Every agent developer solves the same context assembly problem from scratch.

Read More

Qwen3.6-35B-A3B: Community Reviews, Uncensored Variants, and MTP Benchmarks

Alibaba released Qwen3.6-35B-A3B in April 2026: a 35B-parameter MoE model with around 3B active per token, a 262K native context, and an official SWE-bench score of 73.4%. Two months in, it’s the most widely tested 35B-class model in the local LLM community.

Read More

Robot Learning: A Tutorial (From Classical Robotics to Generalist Policies)

“Robot Learning: A Tutorial” (arXiv:2510.12403) is a paper-length tutorial by Francesco Capuano, Caroline Pascal, Adil Zouitine, Thomas Wolf, and Michel Aractingi, from the University of Oxford and Hugging Face. It covers the full arc of robot learning methods, from classical dynamics-based control through reinforcement learning, imitation learning, and generalist vision-language-action models, using the Hugging Face LeRobot library throughout.

Read More

Secret Voting Architecture with FHE, SP1, and Groth16

On-chain secret voting creates three tensions at once. Votes must stay hidden while still being tallied. Off-chain computation cannot be trusted without proof, yet results need to land on-chain. And the EVM cannot run heavy cryptographic operations natively, but it still needs to verify them. FHE (Fully Homomorphic Encryption), SP1 zkVM, and Groth16 each take on one of these.

Read More

VibeThinker-3B: Packing Verifiable Reasoning into 3 Billion Parameters

“Small model beats big model” papers appear regularly. Usually the claim holds on a specific benchmark under specific conditions, not across the board. WeiboAI’s VibeThinker-3B, published June 15, 2026, follows a similar structure but draws a clearer boundary: the claim is not that a 3B model replaces a frontier generalist. The claim is that verifiable reasoning can be compressed into a small model, while open-domain knowledge and general dialogue still benefit from more parameters.

Read More

Future AGI: Evaluate, Observe, and Improve AI Agents in One Place

If you have shipped an AI agent, this will sound familiar. The demo runs fine. Then it hits production, the hallucinations start, and you can’t tell what went wrong or why. So you bolt on one tool for evals, another for tracing, another for guardrails. The real problem is that none of them talk to each other, so the loop you need to actually fix things never closes.

Read More