TL;DR Traditional testing assumes determinism - given input X, function f always returns Y - but LLMs are non-deterministic, which breaks assertion-based testing at its foundation The same agentic task run twice may produce different but equally correct code, making exact-output assertions brittle and often useless The new paradigm shifts from “test the code” to “verify the intent”: property-based testing, LLM-as-a-Judge evaluation, golden datasets for regression, and human review for overall correctness Structured outputs enforce syntactic correctness at generation time, but semantic correctness - whether the output actually solves the right problem - still requires layered verification on top The future of AI quality assurance is designing robust evaluation frameworks and measuring properties of acceptable outputs, not writing exhaustive unit tests for code the model may generate differently next time AI agents like Cline are now the primary “builders” of software in many workflows, executing complex engineering plans from high-level specifications. As I have argued in “The Architect vs The Builder”, the human role is shifting from execution to architectural oversight and defining intent. The patterns that determine whether agents stay shipped are covered in “AI agents that actually work”, and the wider safety framing sits in “AI safety from first principles”. ...

The Space Between Knowing and Becoming There is a particular moment in the journey of personal transformation that nobody quite prepares you for. It comes after the awakening - after you have seen clearly what matters, remembered who you are, and felt the profound sense that something fundamental has shifted inside you. And then reality arrives. The bills still need to be paid. The old habits are still there, waiting at 3am. The people around you haven’t changed, even though you have. The clarity you felt so vividly in that moment of insight begins to blur against the texture of ordinary Tuesday afternoons. ...

TL;DR Career-Ops is an open-source tool built on Claude Code that inverts the job search power dynamic - giving candidates AI-powered evaluation and application tools to match what companies use to filter them Each opportunity is scored across 10 weighted dimensions on an A-F scale, producing a structured comparison that replaces the ad hoc spreadsheet most candidates rely on The system generates ATS-optimized resumes dynamically tailored to each job description and auto-discovers new postings from 45+ pre-configured job boards A key design principle is human-in-control: nothing auto-submits, the AI recommends and the candidate decides, making it a decision-support system rather than an automation Career-Ops is a clean example of the broader pattern of AI tools that amplify individual judgment rather than replace it - worth studying for its architecture as much as its use case The job search has long been a one-way mirror - companies deploy AI to filter applications while candidates manually juggle spreadsheets, tailor cover letters, and hope their resume gets past the automated screener. Career-Ops flips that script entirely. Built on Claude Code, it’s an open-source system that gives job seekers their own AI advantage: intelligent evaluation of opportunities, automated customized applications, and systematic candidate strategy. ...

Amazon Web Services recently introduced AWS S3 Files, a service that addresses a persistent challenge in cloud computing - how to give file-based applications direct access to object storage without duplicating data or building custom connectors. The Problem S3 Files Solves Traditionally, applications designed around file systems faced a difficult choice when working with Amazon S3: Use object APIs - Build custom integration code and refactor applications Duplicate data - Copy data between S3 and separate file systems, creating sync challenges and increased costs Accept performance trade-offs - Work with slower, network-dependent access patterns S3 Files eliminates these constraints by providing a native file system interface directly over S3 data. ...

TL;DR The Lunar Gateway is infrastructure, not a destination: a way station in lunar orbit that separates the hard problem (Earth to lunar orbit) from the repeated one (orbit to surface) Without it, each Artemis landing needs at least two SLS launches plus a Starship - roughly four Moon landings per decade; with it, landings can become routine Assembly is modular: the Power and Propulsion Element arrives around 2028-2029, followed by the HALO habitation module, with operations expanding through the 2030s The Gateway is to the Moon what the ISS is to Earth orbit, and it doubles as a staging point for the lunar far side and eventually deeper space The honest risks are funding uncertainty, lander development, and international coordination - the technology is the more solved part The Gateway Concept When most people think of returning to the Moon, they imagine Artemis astronauts landing, collecting samples, and returning home - just like Apollo. That’s the goal for Artemis III and IV. ...

TL;DR BASIC shaped how a generation thought about programming not because it was good, but because it was the only language shipped with the C64, Spectrum, BBC Micro, and Apple II - the machine booted straight into it The gateway effect was real: zero installation, instant feedback, and a language readable enough that magazine listings could be typed in and modified by children Its flaws left marks too - type coercion taught loose thinking, and GOTO-heavy structure fuelled the structured-programming backlash Platform-fragmented dialects meant BASIC skills never fully transferred, an early lesson in lock-in BASIC’s real legacy is the idea that a computer should be programmable by its owner out of the box - something modern computing has largely abandoned When you powered on a Commodore 64 in 1983, the first thing you saw was: ...

TL;DR The ability to say no to a buildable idea is rarer and more valuable than knowing how to build well Ideas are abundant; attention, energy, and years are not - every yes is an implicit no to everything else you could have built Discernment is a trainable skill: interrogate who the thing is for, what it costs to maintain, and whether you would still want it in five years The compound effect runs both ways - focused builders compound depth, scattered builders compound half-finished repositories The hardest no is to the idea that is genuinely good but still not yours to build Every builder knows the feeling. You have an idea. It’s clever. It could be useful. You start sketching it out, planning the architecture, imagining how it would work. ...

TL;DR The SID (Sound Interface Device), designed by Bob Yannes at MOS Technology in 1981, put a genuine synthesizer - oscillators, a filter, envelope generators - inside the 1982 Commodore 64 It was originally designed as a general-purpose synth-on-a-chip to rival professional instruments, not as a computer sound chip Three voices was a brutal constraint, and composers like Rob Hubbard and Martin Galway turned it into an aesthetic - fast arpeggios simulating chords became the signature C64 sound No other chip - not Atari’s TIA, not the Genesis’s Yamaha FM - achieved the SID’s cultural dominance The chip outlived its platform: SID-based music, hardware clones, and chiptune culture are still active today The Commodore 64, released in 1982, had one feature that set it apart from every other personal computer: it had a synthesizer on a chip. Not a speaker driver. Not a simple sound generator. An actual synthesizer - with oscillators, filters, envelope generators, the same components used in professional synthesizers costing thousands of dollars. ...

TL;DR Claude Mythos Preview set new records across coding, mathematics, and reasoning: 93.9% on SWE-bench Verified, 97.6% on USAMO 2026, and leads GPT-5.4 on every shared benchmark The USAMO result - a 55-point jump over Claude Opus 4.6 - suggests genuinely different reasoning capabilities, not just incremental improvement, and Anthropic screened against memorization concerns Despite dominating benchmarks, Mythos is not publicly available because it autonomously discovered thousands of zero-day vulnerabilities across every major OS and browser Access is restricted to 12 major tech and finance companies via Project Glasswing, a defensive cybersecurity research initiative backed by $100M in Anthropic usage credits The wider implication: we have entered an era where “the best model” and “the publicly available model” may be permanently different things, with security becoming a deployment constraint alongside capability Anthropic released Claude Mythos Preview on April 7, 2026 - and immediately announced it won’t be publicly available. ...

The views in this post are my own personal reflections on the data industry, written in my own time. They are not about any specific employer, team, or colleague, past or present, and do not draw on any non-public information. In 2026, the technical gap between Databricks and Snowflake has narrowed to a sliver. Both offer world-class serverless compute, both support Iceberg/Delta as first-class citizens, and both have integrated AI agents that can write SQL better than your average intern. ...