Search the archive

more actual attendees. See how TLDR compares to Meta.

EIGHT YEARS OF WANTING, THREE MONTHS OF BUILDING WITH AI

Despite how important SQLite is to the industry, no one seems to have invested in building a really good developer experience for it. syntaqlite is a set of high-fidelity devtools for SQLite developed over around 250 hours of effort over three months by a developer using an AI coding agent. This post systematically breaks down the developer's experience in building syntaqlite with AI, covering both where AI helped and where it was detrimental.

HOW CLAUDE CODE BUILDS A SYSTEM PROMPT

System prompts are often the best manual for how an app is intended to work. The accidental leak of Claude Code's source code reveals how Claude Code assembles a context and shows how complex context engineering can be, and the importance of harnesses. This post contains a table that shows how Claude Code assembles a system prompt. Some components are always included, while others are conditional, and components may have variations.

CHINA'S TECH GIANTS ARE RECRUITING HIGH SCHOOLERS

Chinese tech companies are increasingly seeking out high schoolers to fulfill research and internship roles due to insufficient talent in universities. Creativity is increasingly valued for roles such as product managers, and it is believed that younger candidates may be more likely to imagine things that do not yet exist. Globally, some tech leaders are questioning whether universities are still the best way to train talent. Tech companies recruit many people without bachelor's degrees who just figure things out on their own.

ECONOMISTS ONCE DISMISSED THE AI JOB THREAT, BUT NOT ANYMORE

Many economists don't see much evidence that AI is disrupting the job market, but they are starting to take seriously the possibility that it could happen someday soon. AI is expected to grow the economy as the technology improves, but its effect on jobs and the economy is difficult to predict. Given the potential scale of disruption, economists say it is time to start considering policies to help workers who may be displaced or otherwise harmed by the changing economy. The conversations need to happen now, as a lot will change in the next two to five years.

engineer and make $200K-$950K, then apply now.

NON-DETERMINISM ISN'T A BUG. IT'S TUESDAY: WHY PRODUCT MANAGERS WERE BUILT FOR AI (THEY JUST DON'T KNOW IT YET)

The gap between 'using AI' and 'getting real value from AI' is fundamentally a product problem.

OPENAI COO SHIFTS OUT OF ROLE, AGI CEO TAKING MEDICAL LEAVE

Brad Lightcap, OpenAI's longtime COO, will now lead special projects and report directly to CEO Sam Altman.

GITHUB ACTIVITY IS SURGING

GitHub is on pace for 14 billion commits in a year, if growth remains linear.

ELON MUSK INSISTS BANKS WORKING ON SPACEX IPO MUST BUY GROK SUBSCRIPTIONS

Musk also asked that the banks advertise on X, but he was less adamant about that request.

1 IN 30 – ARTEMIS, GREATNESS, AND RISK.

NASA's 'loss of crew' threshold for Artemis missions is calculated to be 1 in 30.

WHY DOMAIN-SPECIFIC LLMS WON'T EXIST: AN INTUITION

The big AI model providers would have made domain-specific models if there was any feasibility in it.

Download the playbook for free

ANTHROPIC ENDED SUBSCRIPTION-BASED OPENCLAW USAGE

ANTHROPIC BUYS BIOTECH STARTUP COEFFICIENT BIO IN $400M DEAL

Anthropic acquired biotech AI startup Coefficient Bio for $400 million in stock to bolster its healthcare and life sciences ventures. Coefficient Bio, founded by former Genentech researchers, used AI to expedite drug discovery and biological research. The 10-person team will integrate into Anthropic's health and life sciences division.

CONTINUAL LEARNING FOR AI AGENTS

Learning within AI agents can happen at the model, harness, or context layers. Understanding the difference can change how systems that improve over time are built. The model layer is the model weights themselves. The harness is the code, instructions, and tools that drive the agent. Context is the additional context that lives outside the harness for more configuration. Most people jump to the model when discussing continual learning, but in reality, an AI system can learn at all three of these levels.

A TAXONOMY OF RL ENVIRONMENTS FOR LLM AGENTS

RL environments are training grounds for agents. Task distribution determines what skills agents develop, and harnesses control how they interact. Verifiers define what 'good' means. The state and configuration determine how realistic the training is.

LLM WIKI

This 'idea file', designed to be copied and pasted into an LLM agent, contains a pattern for building knowledge bases using LLMs. It helps LLMs incrementally build and maintain a persistent wiki that can be extended as the model continues learning. In this framework, the human curates sources, directs the analysis, asks questions, and thinks about what it all means, while the model does the rest. The agent makes edits based on conversations, and users can browse the changes in real-time.

I STILL PREFER MCP OVER SKILLS

The industry is pushing hard for Skills as the new standard for giving LLMs capabilities, but the Model Context Protocol (MCP) is a far superior, more pragmatic architectural choice. Skills are great for pure knowledge and teaching agents how to use an existing tool. MCPs give agents actual access to services, making them the right tool for the job in many cases.

EMBARRASSINGLY SIMPLE SELF-DISTILLATION IMPROVES CODE GENERATION

Simple self-distillation (SSD) is a process where models are fine-tuned on samples of their output with standard supervised fine-tuning. It offers a complementary post-training direction for improving LLM code generation. SSD can improve code generation in models using only raw outputs. This study looks at why such a simple method can work.

META-HARNESS: END-TO-END OPTIMIZATION OF MODEL HARNESSES

The performance of large language model systems depends not just on model weights, but also on their harnesses. Meta-Harness is an outer-loop system that searches over harness code for LLM applications. Its harnesses surpass the best hand-engineered harnesses on agentic coding benchmarks. Meta-Harness shows how richer access to prior experience can enable automated harness engineering.

THEREFORE I AM. I THINK

LLMs often decide actions before generating reasoning tokens, influencing their chain of thought. A linear probe can decode these decisions from pre-generation activations with high accuracy.

INTERACTION-AWARE VIDEO OBJECT REMOVAL

Netflix's VOID is a video inpainting model that removes objects along with their physical and visual interactions, using interaction-aware mask conditioning built on CogVideoX.

BINDING ACTIONS TO MULTIPLE SUBJECTS IN VIDEO

ActionParty introduces per-subject state tokens with spatial biasing to ensure correct action assignment across multiple entities in generated videos, addressing common failures in subject-action binding.

HOW TO BUILD YOUR SECOND BRAIN

A simple three-folder system (raw, wiki, and outputs) turns scattered notes into a structured, AI-maintained knowledge base using plain text files and a lightweight schema. Tools like agent-browser automate content ingestion, while AI compiles, links, and updates a personal wiki from raw inputs without manual organization. The system improves over time by saving outputs back into the loop and running periodic health checks to catch errors and gaps before they compound.

APPLE AT 50: THE IPHONE MAKER 'BLEW A 5-YEAR LEAD' ON AI, BUT FORMER INSIDERS SAY IT CAN STILL WIN

Apple's reliance on Google's Gemini AI for a revamped Siri marks a strategic shift as the company addresses its lag in generative AI compared to peers. This partnership raises concerns about user data management while highlighting Apple's bet on AI capabilities integrated into devices. As AI technology shifts towards the device, Apple aims to leverage its strengths in design and privacy to regain competitiveness in the AI landscape.

MARC ANDREESSEN INTROSPECTS ON THE DEATH OF THE BROWSER, PI + OPENCLAW, AND WHY "THIS TIME IS DIFFERENT"

AI, a result of "80-year overnight success," is not another hype cycle but a culmination of decades of research leading to significant breakthroughs in reasoning, coding, and self-improving agents.

META PAUSES WORK WITH MERCOR AFTER DATA BREACH PUTS AI INDUSTRY SECRETS AT RISK

Meta paused its work with Mercor following a security breach that may have exposed proprietary AI training data.

MICROSOFT'S AI IN ITS OWN TERMS: "USE COPILOT AT YOUR OWN RISK"

Microsoft's Copilot terms label it for entertainment only, cautioning users against relying on it for serious decisions.

LEAKER SAYS THERE'S NO BACK TO BLACK FOR THE IPHONE 18 PRO

Apple dropped the black color option for the iPhone 17 Pro. It now offers only blue, orange, and silver. Despite some user disappointment, the bold Cosmic Orange proved highly popular, especially in China, contributing to strong sales. A reliable leaker now suggests the black finish still won't return for the iPhone 18 Pro, indicating Apple is sticking with more distinctive, status-driven color choices.

CURSOR REFRESHES ITS VIBE CODING PLATFORM WITH FOCUS ON AI AGENTS

Cursor released Cursor 3, a new version of its AI coding platform that focuses on AI agents to help developers automate programming tasks. The platform features a chatbot interface where developers can describe features in natural language and choose which AI model to use, with both cloud-based and desktop agents working together. New features include Design Mode for UI editing, multi-agent task completion, and improved code review processes.

BUILD WITH MORE CONTEXT AND MORE CONTROL IN FIGMA MAKE

Figma Make now supports Make kits and Make attachments, two features designed to bring real design system and project context into AI-generated prototypes. Make kits allow teams to package their component libraries and usage guidelines so prototypes start from production-aligned code rather than generic components. Make attachments let designers supply actual files — such as PDFs, datasets, and screenshots — so prototypes reflect real constraints and content instead of idealized approximations.

HOW TO CREATE DIGITAL CONCEPT ART USING BLENDER AND PHOTOSHOP

A concept artist created The Vault to overcome creative overwhelm by building a consistent world where new ideas could be easily integrated, resulting in a detailed scene of an underground control room separating humanity from creatures below. The workflow combines pre-built assets, Blender scene construction, thumbnailing, and Photoshop paintovers, with an emphasis on storytelling details, visual hierarchy, and efficient techniques like kitbashing and “clown passes.”

THE MANY PATHS IN THE LIFE OF A DESIGNER

There is no single “correct” career path—advice that claims otherwise is usually personal projection, especially in a rapidly changing world shaped by AI and shifting job markets. Instead, careers are nonlinear and shaped by individual choices, environments, and life events, so the best path comes from understanding your own strengths, motivations, and circumstances while staying adaptable and persistent.

HUMAN-CENTRED DESIGN HAS GROWN UP. IT'S TIME WE DID TOO

Early UX pioneers like Steve Krug and Don Norman championed usability and frictionless design, but this success helped create today's attention economy, where seamless experiences often enable manipulation, data extraction, and reduced user autonomy. A new era—driven in part by EU regulations—redefines design as protecting human rights and societal wellbeing, shifting the goal from helping users complete tasks to ensuring systems respect, inform, and preserve human freedom.

AI COWORKER

Viktor is the AI coworker that connects to 3,000+ tools and does the work.

DESIGN SYSTEMS REPORT 2026

A survey with 147 practitioners revealed that design systems are transitioning from peak expectations into a "trough of disillusionment" where teams struggle with understaffing, reduced executive buy-in (dropping from 42% to 32%), and persistent adoption challenges.

PENTAGRAM REVEALS CODE-BREAKING TITLE SEQUENCE FOR RIZ AHMED'S BAIT

Pentagram's Luke Powell and Jody Hudson-Powell created a technically sophisticated yet emotionally meaningful title design for Bait, a comedy-drama by Riz Ahmed, using a color-filter typography system that reveals multiple meanings of the word “bait” to reflect shifting identity and perception. The design extends across the show's visuals, blending espionage aesthetics with deeper cultural and linguistic layers, and marks another milestone in the duo's influential, concept-driven career.

SAN MIGUEL'S NEW LOOK MADE USE OF EYE TRACKING TECHNOLOGY

San Miguel has unveiled its first redesign in a decade, using consumer research and neuroscience tools like EEG and eye-tracking to refine packaging for stronger shelf visibility, premium perception, and emotional impact. The updated look—subtle but more polished—is rolling out across Europe alongside its “Spanish Summer, No Matter When” campaign, reinforcing its association with sunny, social moments.

AI'S AESTHETICS OF FAILURE

AI technology, despite its sophistication, has been plagued by poor aesthetics that even supporters like Nvidia's CEO call "slop." OpenAI shut down its video generation app Sora this week after usage plummeted from 6 million to 1.5 million monthly downloads due to users finding it unpleasant and unfun to use. The platform was dominated by tasteless content and glitched-out videos that failed to engage audiences beyond initial novelty.

DESIGNBOOM'S ULTIMATE GUIDE TO MILAN DESIGN WEEK 2026

Milan Design Week 2026, which runs from April 20–26, will transform the city into a sprawling showcase of design, architecture, and innovation.

TEN MUST-KNOW MODERN JEWELLERY DESIGNERS

Visionary jewelry designers from London to Lahore are creating modern pieces that draw inspiration from their local landscapes, translating street signs, folkloric symbols, and everyday objects into elaborate designs.

SIX DESIGNS THAT REIMAGINE HOW WE INTERACT WITH SOFTWARE

Product teams at ServiceNow, Ticketmaster, and Affirm are using Figma Make to build conviction faster through prototyping.

COMPONENTS OF A CODING AGENT

Coding agents, or harnesses, are sophisticated systems that wrap LLMs to improve their performance and usability for software development tasks by acting as a control loop that manages repository context, tool access, memory, and efficient prompt construction. This architectural layer enables capabilities like command execution and long-session continuity, making LLMs far more effective in practical coding scenarios.

WE REPLACED THREADLOCAL WITH SCOPED VALUES IN JAVA 26 — THEN OUR MEMORY LEAKS VANISHED

A Java user-auth service experienced frequent Out Of Memory (OOM) crashes due to memory leaks after migrating to Virtual Threads. The issue was from legacy code using `ThreadLocal`, which inefficiently copied large `SecurityContext` objects into numerous short-lived virtual threads, causing excessive heap memory consumption. To resolve this, the team migrated from `ThreadLocal` to `ScopedValue`, a new Java 26 feature designed for immutable, zero-copy data sharing across threads and their children.

THE MACHINES ARE FINE. I'M WORRIED ABOUT US

AI agents threaten the fundamental learning process in academic research, particularly for new scientists. This is illustrated through Alice, who gains deep understanding by tackling problems herself, versus Bob, who uses AI to outsource tasks, producing similar output without developing core skills or intuition. The real danger is a "slow, comfortable drift" where researchers produce results without true comprehension.