Search the archive

PER-WALLET 2FA CONTROL AND WALLET SECRET EXPORT

Infinex has launched per-wallet 2FA controls, allowing users to assign distinct security postures to individual wallets rather than applying a single account-level setting. Wallets with 2FA enabled form a security zone where intra-zone transfers require no authentication code, while moving funds to wallets outside the zone triggers verification, enabling a hot/cold wallet split within one interface. Infinex also added wallet secret export, letting users retrieve seed phrases or private keys after passkey signing and 2FA confirmation, with the secret rendering inside Turnkey's hosted export iframe and decryptable only within Turnkey's secure enclave, so neither Infinex nor Turnkey can access the plaintext. Together, these features resolve the friction trade-off that uniform account-level 2FA imposed on active DeFi users who need both custody security and frictionless daily transaction flow.

SOLANA CUTS MAINNET SLOT TIME TO 350 MILLISECONDS

Solana's Friday mainnet upgrade cut target slot time from 400ms to 350ms under proposal SIMD-0525, which plans further cuts to a final 200ms target in three more 50ms stages. The change reduces confirmation latency but doesn't increase throughput, since each slot now carries proportionally less work. Full finality remains far slower (~12.8 seconds today) pending the separate Alpenglow overhaul, which aims to eventually bring finality down to about 150ms.

THIS WEEK ON BASE: BASE BATCHES 004, VENICE, AND MORE

Base crossed $5B in DeFi TVL (DeFiLlama) and surpassed every other network in monthly stablecoin card spend (Paymentscan), with $2.5B USDC deployed across 16 lending venues, $2.1B of which is concentrated in Morpho Blue. AI agents executed 7.3 million of the 14 million total x402 transfers logged across all chains over the past 30 days, and Blockrunai cleared 1 million x402 transactions per day on Base, underscoring the chain's dominance in machine-to-machine payment rails. App-layer economics are materializing: Venice crossed $100M in annualized revenue, and o1 Exchange recorded $200M in cumulative trading volume with over 25,000 tokens launched and $1M in creator fees distributed. Base Batches 004 opens applications through September 9, targeting 10 early-stage teams with up to $100K each and a NYC Demo Day.

ONSHORING HYPERLIQUID: THE COMPLIANT PATH TO US MARKETS

Hyperliquid remains geoblocked from US markets because its permissionless infrastructure conflicts with CEA requirements for registered DCMs, DCOs, and FCMs, the exchange, clearinghouse, and broker roles that TradFi separates across entities. HyperCore's modular stack mirrors this separation: the protocol handles matching and settlement, deployers set contract specs (staking 500k HYPE bonds and earning up to half the fees their markets generate), and builders onboard users as FCM equivalents. The Hyperliquid Policy Center has lobbied the CFTC and SEC to allow regulated firms to access HyperCore through permissioned HIP-3 DEXs, bearing KYC/AML obligations as builders and deployers. Trump's announcement validating this onshoring approach means US traders may gain access to Hyperliquid's perpetuals infrastructure through regulated wrapper products rather than waiting for the protocol itself to register.

TIMING THE HEAD IN ETHEREUM POS

New research on Ethereum's PoS consensus identifies two attack vectors that let adversarial proposers exploit head vote timing in LMD-GHOST to strip honest validators of attestation rewards. The head-vote timing game has a proposer delay block broadcast to near the 4-second mark in a 12-second slot, pushing honest validators to attest the parent block instead. Prysm experiments measured a 15.06% average total reward loss for honest participants under this manipulation. A second vector, the K-block withholding attack, allows adversaries holding consecutive proposer slots to extend a private competing chain and release it to trigger reorganizations, voiding honest attestation rewards on the reorganized blocks. Researchers propose a distance-weighted head reward mechanism that grants partial credit for attestations pointing to recent canonical ancestors, reducing timing-game losses from 15.06% to 4.71% without modifying core consensus rules.

BOUNCEBIT TO PERMANENTLY SHUT DOWN ITS L1 AFTER 286.5M BB EXPLOIT

BounceBit is permanently shutting down its Evmos-based L1 after a protocol-level authorization flaw allowed an attacker to drain approximately 286.5 million BB tokens from nine mainnet accounts, with no private keys or wallets compromised. The YZi Labs-backed project, whose 2024 seed round was co-led by Blockchain Capital and Breyer Capital, will reissue BB as a BEP-20 token on BNB Chain using a pre-exploit snapshot. The 286.5 million exploited tokens will not be recognized in the new issuance, zeroing out the attacker's position. CeDeFi, Prime, and RWA products operated by the team were unaffected by the breach.

SEND PARTNERS WITH WIREX TO POWER SEND CARDS

Send has selected Wirex, which processes over $20 billion in transactions for 8 million users and recently passed $2 billion in annualized onchain card spend, as its card infrastructure provider, citing instant settlement, 0% FX conversion, and shared commitment to self-custody.

WHICH CRYPTO CARDS MAKE TAX REPORTING EASIEST

Plasma One and Wirex offer the cleanest one-click PDF/CSV exports, while KAST, Avici, Bitpanda, Revolut, Crypto.com, and Nexo also support usable exports.

GRAYSCALE MOVES CLOSER TO LAUNCHING FIRST ZCASH ETF IN US

Grayscale filed its fifth SEC amendment converting its Zcash Trust ($260M+ AUM, running since 2017) into The Zcash ETF.

THE MISSING HALF OF X402 AGENT TRUST INFRASTRUCTURE

The agent-side spending stack for x402 payments has made strong progress, with spending parameters and policies now getting locked down.

HOW WE KNEW COVID WAS OVER (AND WHAT OUR MODELS HAD TO UNLEARN)

Model maintenance is not just retraining. When COVID-era demand patterns faded, Airbnb's forecasting team separated parameter drift from structural change, choosing among refit, respecify, or hold decisions instead of blindly updating thousands of markets. The result was a useful operating model for production forecasting under regime shifts.

FROM 17MS TO 0.04MS: HOW TO DESIGN THE RIGHT SQL INDEX

Good SQL indexes come from real query patterns, not the table schema. Composite indexes need the right column order, usually equality filters first and sort or range columns after, with EXPLAIN ANALYZE used to verify the result. The right indexes cut example queries from 17ms to 0.04ms and 436ms to 0.5ms.

A TALE OF TWO FLINK AUTOSCALERS

Netflix is moving 30,000+ Flink jobs from a homegrown autoscaler toward the Apache Flink Autoscaler. The old system saved 25–45% by scaling whole clusters from container metrics, but missed operator-level state and suffered from telemetry gaps. The OSS autoscaler reasons from inside each job and produced 58% annualized compute savings for one team.

ZERO-SUM BY DESIGN: 10 YEARS OF UBER'S PAYMENTS PLATFORM

Uber's Gulfstream payments platform now supports $217B in annualized gross bookings, nearly twice that in money movement, and ledger balances for 1.2B+ entities. The design relies on immutable zero-sum money orders, double-entry accounting, DynamoDB-backed balances, Kafka pipelines, Cadence for synchronous flows, and generic payment-instrument interfaces that let new business lines plug in with minimal core changes.

THE MOST VALUABLE API IN THE MODERN DATA STACK RETURNS NO DATA

Bulk export APIs are giving way to shared table references. Iceberg REST catalogs, object storage, Parquet, and credential vending let producers publish metadata while consumers bring their own compute, avoiding pagination, rate limits, and duplicate pipelines. The hard parts move to contracts, compaction, semantics, privacy, and table maintenance promises.

THE QUIET POLITICS OF DASHBOARDS: VISUALIZING POWER, GOVERNANCE, AND BLIND SPOTS IN AFRICAN ORGANIZATIONS

Dashboards are not neutral: the metrics and visuals they prioritize shape what leaders see, fund, and act on, while harder-to-measure realities disappear. In African public-sector and NGO settings, donor-facing dashboards can obscure frontline needs and should be balanced with local knowledge.

NOT EVERY PROBLEM NEEDS AN AI AGENT

Production AI systems work best when each component earns its complexity. Recommenders, moderation, warehouse queries, and search cold starts need different mixes of deterministic logic, classic ML, semantic layers, and LLMs. The strongest pattern is often hybrid: use agents for reasoning, but keep business logic, metrics, routing, and guardrails deterministic where possible.

DUCKDB V2.0: YOUR DATABASE DESERVES A BETTER PARSER

DuckDB v2.0 replaces its PostgreSQL-derived SQL parser with a PEG-based parser while preserving the downstream binder and AST pipeline. The change removes Bison conflict pain, avoids exponential backtracking on malformed input, and enables runtime grammar extensions that can add syntax without reimplementing SQL parsing.

TRUEFOUNDRY OPEN-SOURCES TRUEFORGE, AN ENTERPRISE AI AGENT HARNESS

TrueForge is a self-hostable agent harness focused on reducing task cost through context compaction, delayed tool-schema loading, subagents, and sandbox-as-a-tool execution. The release matters for data teams experimenting with governed agents because it separates orchestration from model choice and pairs naturally with gateways for identity, access control, observability, and budgets.

DATAFUSION COMMUNITY SHOWCASE VOL. 4: RDF FUSION & CLOUDFLARE R2 SQL (63 MINUTE VIDEO)

Two projects push DataFusion beyond standard analytics: one combines RDF graphs with time-series and temporal logic, while Cloudflare's AutoSQL uses it for distributed serverless SQL over Iceberg and R2. Both rely heavily on custom planning, pruning, joins, and execution optimizations.

USING MODELS TO CREATE MODELS OF NEW YORK CITY

Peter Sobot used 1,000+ photos, Claude, and Gaussian Splatting to build a 3D model of the New York skyline. Running 74+ subagents and roughly 280 experiments, the project showed that AI can handle much of the technical workflow, but still struggles with visual judgement, messy data, and knowing when the result is genuinely good.

THE SCAFFOLDING MATTERS MORE THAN THE INTERFACE

Agent scaffolding mattered far more to cost than whether tools used MCP or CLI. CLI-only agent scaffolds were 5x to 28x cheaper, while MCP-vs-CLI cost ratios ranged from 0.43x to 29x, making interface choice alone a weak predictor of efficiency.

AUTOMATIC APACHE KAFKA MIGRATIONS WITH ORBIT

WarpStream's Orbit automates Kafka migration planning and cutover checks so teams can move clusters faster without silently breaking producers, consumers, offsets, or ordering assumptions.

AI ENGINEERING SKILLS MAP: BUILDING AND DEPLOYING AI APPLICATIONS

Andrew Ng's AI engineering skills map centers on six areas: LLM foundations, data grounding, agents, evals, production operations, and ML fundamentals.

APPLE LAYS OFF 200+ PEOPLE ACROSS VISION PRO AND SIRI TEAMS

Apple has laid off more than 200 employees across its Vision Pro, Siri, and software engineering teams as part of a broader restructuring effort that includes scaling back Vision Pro gaming and immersive video initiatives while reaffirming its long-term commitment to visionOS and future hardware. The company is also reorganizing Siri and AI-related teams around its new Siri AI platform, eliminating some roles, creating new ones, and shifting resources to accelerate development of AI-powered features across its products.

CHATGPT FOR IPHONE NOW LETS USERS GRAB RECENT PHOTOS WITH A LONG PRESS

OpenAI has added a new iPhone shortcut that lets ChatGPT users long-press the + button to instantly access and attach their four most recent photos, streamlining a process that previously required several taps through the Photos menu. The update reflects the growing importance of image-based interactions with AI and joins other recent improvements aimed at making photo uploads, visual questions, and longer conversations faster and more seamless.

SLACK HAS (OF COURSE) LAUNCHED A VIBE CODING TOOL

Slack Code is a feature that lets teams tag AI coding agents like Claude Code, Devin, or GitHub Copilot directly in group chats to build and ship software fixes. Available now on all Slack plans, it requires human approval for high-stakes actions like production merges and archives the channel once work is approved. The launch continues Slack and Salesforce's broader push to market AI agents as autonomous digital coworkers, following prior products like Slackbot and Agentforce Coworker.

FIGMA IS THE NEW DREAMWEAVER: HOW MODERN PROTOTYPING TOOLS ARE TRAPPING US IN 2018

Figma excels at visual design and collaboration, but treating it as a faithful representation of the web can lead designers to overlook how interfaces actually behave in real-world conditions. Static artboards and Auto Layout can mask challenges such as responsive behavior, loading performance, accessibility, unpredictable content, and dynamic states. As products become increasingly adaptive and AI-driven, designers need to think more like systems architects, testing ideas in code, designing for edge cases, and grounding design systems in production-ready components rather than purely visual artifacts.

HOW TO CHECK YOUR AI SYSTEM STILL MATCHES YOUR VALUES

As AI becomes more deeply embedded in products and services, designers have a growing responsibility to help ensure that technology reflects human values, ethics, and societal needs. Building trustworthy AI requires close collaboration between design, research, engineering, and governance, with values such as privacy, transparency, and safety translated into concrete decisions throughout the development process. A human-centered approach treats design as the connective thread that helps teams align strategy, implementation, and outcomes while continuously evaluating whether systems deliver on the promises made to users and society.

ONE AI OUTPUT IS AN EXAMPLE, NOT AN EVALUATION

A single AI output only demonstrates a system can perform a task once—it cannot show how reliably that performance holds across different situations. Rigorous evaluation, much like quantitative UX research, requires testing multiple representative inputs across repeated runs, then reporting averages with confidence intervals to capture both test-input and run-to-run variability. Systems with identical overall scores can fail in very different ways—consistently on specific inputs versus unpredictably across all inputs—so understanding failure patterns matters as much as the aggregate score itself.

A MIRROR FOR THE WORLD'S BEST AI MINDS

Turn AI apps, agents, and workflows into one-click tools you can run, save, and make your own.

CREATE DESIGN SYSTEMS, BUILD WITH AGENTS

Balsa UI creates coherent Vue design systems that people and coding agents can understand, install, and evolve.

WATCH SUSAN KARE REVISIT THE ICONS AND GRAPHICS FOR THE ORIGINAL MAC

Susan Kare reflects on designing the original Macintosh's iconic visual language and shares lessons on creativity, problem-solving, and the lasting impact of thoughtful interface design.

A SLOPPY INTERFACE IS A SECURITY LIABILITY

AI now makes it trivial to fake polished interfaces, as seen in phishing attacks like the one behind the Axios npm incident, raising interface craftsmanship into a security consideration. High-fidelity design details—micro-interactions, loading behavior, UI voice, edge-case handling—are hard and costly to replicate, making sloppy products easier targets for imitation-based attacks. Investing in above-average craft therefore serves as both a brand asset and a practical layer of defense against exploitation.

VIBE CODING AND QUALITY

Vibe coding was once dismissed as producing only prototype-quality, "duct-taped" results unsuitable for reliable software. A Mac app built with Codex reportedly outperformed existing human-made alternatives within a day, with clear room for further refinement. This experience led to the conclusion that achieving exacting standards of quality is realistically possible only through vibe coding, not conventional programming.

TYPOGRAPHIC TOURISM, OR THE ART OF HUNTING FOR BEAUTIFUL SIGNS ALL OVER THE WORLD

Creative director Yorgo Tloupas documents vintage and hand-painted shop signage worldwide under the hashtag #typographictourism, chronicling a vanishing craft.

WHY ARE WE STILL TREATING CANVA LIKE A DIRTY LITTLE SECRET?

Canva has shifted from a platform once dismissed as amateurish into a serious industry contender, bolstered by its Affinity and Calvary acquisitions.

Register now

THERE'S NO REASON FOR SOFTWARE TO BE SLOW ANYMORE

The cost of software optimization has drastically decreased, enabling more developers to implement performance improvements that previously required specialized skills. This allows for dynamic custom software tailored to specific workloads. As the barrier to entry for coding and testing optimizations lowers, it becomes increasingly feasible to get performance gains with minimal effort and time.

AI CHIP ARCHITECTURES

Advancements in AI chip architectures have led to NVIDIA GPUs, Google's TPUs, AMD Instinct GPUs, and Cerebras' Wafer-Scale Engine, each with distinct designs catering to specific computational AI tasks. The key differentiators include NVIDIA's focus on programmability and scalable architecture, Google's emphasis on a single-purpose design optimized for matrix multiplication, AMD's commitment to memory capacity and open standards, and Cerebras' innovative wafer-scale integration for ultra-fast memory access.

HOW I FIND PROBLEMS TO SOLVE AS A STAFF ENGINEER

Finding worthwhile problems to solve as a staff engineer involves actively listening to the challenges faced by team members rather than waiting for management to identify opportunities. By absorbing feedback, allowing potential problems to accumulate, and connecting seemingly unrelated issues, engineers can uncover needs and develop impactful solutions that align with the broader organizational goals.

AI AND INFRASTRUCTURE ENGINEERING

The integration of AI into infrastructure engineering is improving efficiency by automating routine tasks and lookups, freeing engineers to focus on higher-level decision-making, but it may also lead to diminishing expertise in foundational skills.

ON TEACHING AI HOW YOU WORK

A collection of eleven skills has been developed to streamline the process of teaching AI how to interact with design and engineering tasks by consolidating key lessons and practices into a single repository. This framework allows users to apply these skills efficiently, ensuring consistency in design intent and output while minimizing the need for repeated explanations.

Claim 50% off

MAKA

Apache Maka is an incubating project that provides a local-first AI agent workspace where user interactions and data are stored locally, allowing for the execution of tasks while maintaining control over session data.

PROLIFERATE

Proliferate is an open-source AI IDE that enables users to run various coding agents, such as Claude Code and Codex, in parallel within isolated workspaces. It supports self-hosting and has features like reusable workflows, integration with multiple tools, and the ability to manage different tasks simultaneously in a structured manner.

EVERYTHING I OWN, OWNED

Recent experiments with various peripherals revealed security vulnerabilities, enabling unauthorized access and control over devices such as microphones and webcams. The ease of reverse engineering these devices raises concerns about their security integrity, suggesting a future where malware could exploit such vulnerabilities to infiltrate and control connected accessories and IoT devices.

WHY YOUR LOCAL LLM FEELS DUMBER THAN IT IS

Local implementations of LLMs often provide subpar performance due to variations in hardware, software, and configurations across different setups. This post explores these discrepancies through technical experiments, showing how benchmarks, model choices, and the specifics of quantization can affect the output and reliability of LLMs while underscoring that such issues are common to all users.