Search the archive

WORKING ON PRODUCTS PEOPLE HATE

Engineers don't control whether products are loved - incentives and constraints do. The job is to balance user needs with reality and improve even imperfect, widely used products.

Get ahead with this industry report.

IN DESIGN, CRAFT IS NOT ENOUGH

Design today overemphasizes visual polish while neglecting usability and real user needs. This shift weakens design's strategic role, especially as AI accelerates building without improving thinking.

AMAZON VP REVEALS EVERYTHING HE'S SEEN IN CORPORATE POLITICS (170 MINUTE VIDEO)

Career advancement requires mastering corporate politics and interpersonal skills just as much as technical expertise. Success comes from building strategic alliances, understanding hidden motives, and leveraging soft power.

TASTE IS NOT A FEATURE

As AI makes creation easy, taste becomes the real differentiator. Without deep judgment and context, fast output leads to shallow, repetitive work.

MEA CULPA

The “magic bullet” approach fails because it ignores reality: demand far exceeds capacity and delivery is inherently unpredictable. What works instead is trust, alignment, and early collaboration between sales and product.

Zero code, visual editor. Curious? Claim free credits

IS YOUR LANGUAGE LEARNING APP LYING TO YOU?

Many apps prioritize feeling good over real progress by being too lenient.

TWO WORLDS

AI raises capability but lowers value by making creation easy for everyone.

Schedule a demo to learn more

ANTHROPIC READIES MYTHOS MODEL WITH HIGH CYBERSECURITY RISK

A configuration error in Anthropic's content management system exposed a draft blog post describing a new model called Claude Mythos. The model is reportedly larger and more intelligent than the Opus models. The draft warns that the model poses unprecedented cybersecurity risks as it is far ahead of any other AI model in cyber capabilities. Mythos is extremely compute-intensive and expensive to serve - Anthropic is working on making it much more efficient before any general release.

THE DECADELONG FEUD SHAPING THE FUTURE OF AI

Dario Amodei and Sam Altman clashed for years when Amodei was still at OpenAI. The final split happened toward the end of 2020, after a disagreement over whether OpenAI should be a market company or a public-good company. Dario left with nearly a dozen OpenAI employees. Within five years, Anthropic would be lining up banks, racing toward an initial public offering before its founders' former employer.

I'VE TAKEN AGENCY IN THE TREATMENT OF MY BONE CANCER

Sid Sijbrandij ran out of standard care treatments for bone cancer, so he started creating treatments for himself. This site details his journey and the results of the treatments so far. Sijbrandij's cancer is now in remission. He is now working with companies to scale his approach for others.

“SUPER BIZARRE” – NEUROSCIENTISTS DISCOVER THAT ADULT BRAIN IS FILLED WITH MILLIONS OF “SILENT SYNAPSES”

Scientists have discovered a surprisingly large reserve of silent synapses in the adult brain. These unused neural connections can be rapidly activated to store new memories. This system allows the brain to create new memories without overwriting the important memories stored in mature synapses. If silent synapses decline with age or disease, it could help explain why learning new skills or adapting to change becomes more difficult over time.

for the task at hand. See how Unblocked works.

PRETEXT

Pretext is a browser library that solves the problem of calculating the height of a paragraph of line-wrapped text without touching the DOM. It uses a clever array of tricks to make the process faster and enable all sorts of new text rendering effects. Pretext supports all languages, including Korean mixed with RTL Arabic and platform-specific emojis. A demo is available in the article.

HOW TO SOLVE PROBLEMS OF LONG RUNNING, AUTONOMOUS AGENTIC ENGINEERING WORKFLOWS

All harness design is to overcome the problems of agents either becoming lazy and cutting corners or being confused and stupid. For most people, starting with a vanilla startup using native features will be good enough. However, a custom agent will allow users to create new workflows that directly address all the ways agents fail in more specific contexts. This article describes all of the problems developers likely face when using agents in long-running autonomous projects so that they can better put these problems into words when creating custom harnesses.

THE MOST IMPORTANT IDEAS IN AI RIGHT NOW

The most important ideas in AI right now are autonomous component improvement, the transition to intent-based engineering, the move from opacity to transparency, the realization that most work is scaffolding, and the idea that expertise gets diffused into public knowledge. The speed of improvement in many fields is about to accelerate beyond anything we've ever seen. The new bottleneck is being able to say what you actually want. Every organization is going to converge on the same cycle: define what you're trying to do, execute with agents, log everything, collect failures, and let the system improve itself.

BUYING DEEPMIND

Google bought DeepMind for $650 million at the end of January 2014. It poured billions into DeepMind's research, but it took a decade for the quest for superintelligence to go into overdrive. This article tells the inside story of how Google ended up purchasing the startup and the specific terms of the final deal.

time by 70%.

AI UNMASKED OUR WORK AS SCAFFOLDING

AIs with Agent Skills can execute tasks as good or better than most people who do that work professionally.

STOP HOLDING OUT HOPE, LIQUID GLASS WILL BE MANDATORY IN IOS 27

Apple has made it clear that Liquid Glass is absolutely moving forward, evolving, and expanding across the ecosystem.

JSONGREP IS FASTER THAN {JQ, JMESPATH, JSONPATH-RUST, JQL}

jsongrep takes a query and a JSON input and prints every value whose path through the document matches the query.

OPERATIONAL BENEFITS OF DATABASES BUILT ON OBJECT STORAGE

Object storage is cost-effective, and the independence of compute from storage allows flexible scaling of compute resources.

WHY ARE EXECUTIVES ENAMORED WITH AI BUT ICS AREN'T?

Executives built their careers around managing non-deterministic systems, while ICs operate in a more deterministic world and are evaluated on their ability to deliver precise, reliable output.

PGMICRO

pgmicro is an in-process reimplementation of PostgreSQL that is backed by a SQLite-compatible storage engine.

See it in action and request a demo

CLAUDE MYTHOS

'Mythos' is the name for a new tier of Anthropic models that are larger and more intelligent than Opus. The models get dramatically higher scores on tests of software coding, academic reasoning, and cybersecurity compared to Claude Opus 4.6. Mythos is a large, compute-intensive model that is very expensive to use and serve. Anthropic is working on making the model much more efficient before any general release.

META TESTS AVOCADO 9B, AVOCADO MANGO AGENT, AND MORE

Meta's Avocado model has been pushed back to at least May as it still falls short of leading systems from competitors. The company appears to be running parallel experiments with multiple Avocado variants. The model appears to be able to solve complex math problems that earlier Llama models could not, but these problems have already been solved by other labs months earlier. Meta's AI leadership has reportedly discussed temporarily licensing Google's Gemini technology. Some requests within Meta AI are already being routed through Gemini models.

ANTHROPIC'S CLAUDE POPULARITY WITH PAYING CONSUMERS IS SKYROCKETING

Anthropic is more popular with customers than ever. Claude is gaining paid subscribers in record numbers. Paid subscriptions have more than doubled this year. The majority of new subscribers were in the lowest tier. OpenAI is still gaining new paid subscribers at a rapid rate and remains the biggest consumer AI platform.

FUNCTION CALLING HARNESS: FROM 6.75% TO 100%

AutoBe is an open-source AI agent that takes a single natural language conversation and generates a complete backend. qwen3-coder-next has a 6.75% function calling success rate when asked to generate API data types for a shopping mall backend. AutoBe boosts that success rate up to over 99.8%. It uses a harness where type schemas constrain outputs, compilers verify results, and structure feedback pinpoints compactly where and why something went wrong so the agent can correct itself. This post dissects the engineering behind AutoBe.

AI'S CAPABILITY IMPROVEMENTS HAVEN'T COME FROM IT GETTING LESS AFFORDABLE

AI's capability improvements at the frontier have not led to increased inference costs relative to human labor. Despite rising per-task inference costs, current models achieve tasks at roughly 3% of human costs without any upward trend in median cost ratios. Models can continue advancing even under strict cost constraints, enabling profitable automation with AI cost ratios remaining well below human levels.

THE CAPABILITY OVERHANG IN AI

Coding agents outperform other domains because codebases provide a self-contained environment of critical context, unlike fragmented knowledge work spread across video calls and legacy systems. Enterprise adoption remains stalled by the three hard problems of context fragmentation, complex access control, and a rapidly shifting architecture landscape.

SCHEDULE TASKS ON THE WEB

Claude Code on the web users can now schedule tasks. The tasks will run on Anthropic-managed infrastructure, so they will keep working even if users turn off their devices. Scheduled tasks are available to all Claude Code on the web users. Example tasks include reviewing open pull requests each morning, analyzing CI failures overnight and surfacing summaries, syncing documentation after PRs merge, and running dependency audits every week.

LAT.MD

lat.md is a spec that agents keep in sync with the code base that helps them understand big ideas and key business logic. It ensures that corner cases have proper high-level tests that matter and can speed up coding by saving agents from endless grepping. The spec uses plain Markdown, with Wiki links connecting concepts into a navigable graph.

WHAT PRETEXT REINFORCED ABOUT AI LOOPS

Pretext is a fast, accurate, comprehensive text measurement algorithm that can lay out web pages without leaning on DOM measurement and reflow. It was created using AI agent workflows. The particular loop that was used in developing the tool (constrain -> measure -> isolate -> classify -> test -> reject -> keep only what survives broad pressure) made the engineering rigorous. This article analyzes the loop to see what makes it so successful.

XAI'S LAST COFOUNDER LEAVES

All remaining co-founders of xAI reportedly departed, marking the complete exit of the original founding team.

THINGS I LEARNED AT OPENAI

OpenAI alumni emphasize the significance of creating effective evaluations and benchmarks, noting that the best benchmarks drive collective optimization efforts. Post-training data design and model alignment are critical for unlocking new AI capabilities, particularly in subjective attributes like empathy or creativity. Fast iteration, choosing the right problems, and leveraging internal tooling are key competitive advantages in AI research.

THREE WEEKS AGO THERE WERE RUMORS THAT ONE OF THE LABS HAD COMPLETED ITS LARGEST EVER SUCCESSFUL TRAINING RUN

That lab was likely Anthropic, which trained Mythos.

LIVE TRANSLATE COMES TO HEADPHONES ON IOS

Google rolled out real-time translation through headphones on iOS, expanding support to more countries and 70+ languages while preserving speaker tone and cadence.

WHY AI IS NOT KILLING THE CYBERSECURITY INDUSTRY, BUT EXPANDING IT EXPONENTIALLY - THOUGHTS FROM RSA

AI-powered exploitation can attack systems at a scale never possible before.

APPLE CONFIRMS MAC PRO IS DISCONTINUED AFTER TWO DECADES

Apple has discontinued the Mac Pro, ending its long-standing tower workstation line and shifting pro users toward the more compact and powerful Mac Studio, whose Apple silicon performance now covers most professional needs. The move reflects Apple's broader strategy of prioritizing integrated, efficient hardware over modular expandability, with Thunderbolt handling most expansion needs while specialized workflows (like CUDA) move to non-Mac systems.

GOOGLE, COHERE LAUNCH NEW AUDIO AI MODELS

Google and Cohere launched new AI models optimized for audio processing: Google's Gemini 3.1 Flash Live, designed for customer service automation, and Cohere's model focused on speech transcription. Gemini 3.1 Flash Live can detect user emotions, process images alongside speech, and scored 90.8% on benchmarks, while Cohere Transcribe achieved a 5.42% word error rate across multiple languages. Both models offer significantly improved performance over their predecessors and are available for companies to integrate into their applications.