Mumin Ahmad Khan
CAT. NO. MAK—157
FILED 13.09.2026
STATUS OPEN
Full-stack software engineer building useful tools. Constantly learning, with a habit of connecting dots across different domains.
NOTES 13
OPEN →- 01
- 02
- 03
- 04
- 05
LAST OPENED
Open an entry and it'll be filed here.
01 PREETHI SAM
5 CSS Properties You Should Know for Better Text Designs Five underused CSS properties for sharper typography, from background-clip to text-combine-upright.
OPEN ↗ 02 OMEGION
AI and Infrastructure Engineering Why AI automates the lookup work in infra but not the architectural decisions.
OPEN ↗ 03 FLORIAN HERRENGT
AI Is Removing the Middle Class of Software Engineering Why AI lets weak engineers ship problematic code faster than teams can review it.
OPEN ↗ 04 THE ATLANTIC
AI Wearable Surveillance Countermeasures On the countermeasures emerging against always-recording AI wearables.
OPEN ↗ 05 DAN MCKINLEY
Choose Boring Technology On spending your limited innovation tokens on the business, not the stack.
OPEN ↗ 06 BRANDUR
Does Anyone Run Postgres Without PgBouncer? On connection pooling being so universal it arguably belongs in Postgres itself.
OPEN ↗ 07 ACM QUEUE
Eight Myths on Software Engineering and GenAI On the marketing claims and misread studies driving bad decisions about AI adoption.
OPEN ↗ 08 JOHN
Elevators How elevator scheduling actually works, and why destination dispatch underperforms plain buttons.
OPEN ↗ 09 DREW BREUNIG
Fable & The End of the Free Lunch How expensive frontier models force you to choose which model does which task.
OPEN ↗ 10 IKEA
How Does IKEA Come Up With Names for Its Products? On the Swedish naming system where sofas take place names and bookshelves take men's names.
OPEN ↗ 11 LAURENTIU RADUCU
How I Use LLMs to Learn Complex Topics On building interactive simulations with an LLM instead of asking it for explanations.
OPEN ↗ 12 PAUL GRAHAM
How Universities Should Prepare Founders Why deep technical skill and independent projects beat an entrepreneurship curriculum.
OPEN ↗ 13 ALEX CHAN
How We Tracked Down a 16-Year-Old SQLite Bug How scattered corruption reports led to a rare data race in SQLite's checkpoint path.
OPEN ↗ 14 ADAM CONWAY
I Gave Qwen 3.8 27B a Reverse-Engineering Job I Assumed Needed a Frontier Model How a local 27B model recovered a commercial app's licensing keys, entirely offline.
OPEN ↗ 15 RAFAEL PIERRE
I Put GitHub Copilot Behind a MITM Proxy What Copilot's network traffic reveals about how its features spend your credits.
OPEN ↗ 16 BEN DICKEN
Making 768 Servers Look Like 1 How sharding spreads data across hundreds of servers behind a single database interface.
OPEN ↗ 17 DEEPGROVE
Maple-Preview A 20B-A1B ternary-weight reasoning model built for efficient on-device inference.
OPEN ↗ 18 WALTER VAN DER GIESSEN
Models Are Getting Dumber on Purpose How labs trade factual recall for reasoning, moving facts from weights to retrieval.
OPEN ↗ 19 FABIEN SANGLARD
My agent.md to Improve LLM-Assisted Code Quality How writing style guidelines down up front changes what an LLM hands back.
OPEN ↗ 20 PETER KRAFT
Postgres LISTEN/NOTIFY Can Actually Scale How batching pushes LISTEN/NOTIFY to 60K writes a second at millisecond latency.
OPEN ↗ 21 ANKUR SETHI
Prevent Cognitive Debt by Manually Retyping LLM-Generated Code Why retyping generated code by hand buys understanding at the cost of speed.
OPEN ↗ 22 BEN GODFREY
Process as a Proxy for Motivation How deciding the process up front removes the decision fatigue willpower can't cover.
OPEN ↗ 23 SIMON WILLISON
Qwen 3.8 27B Is Excellent, but It Defaults to Wildly Overthinking Things On a capable open-weight 27B model that burns reasoning tokens by default.
OPEN ↗ 24 JARRED SUMNER
Rewriting Bun in Rust How Bun's runtime moved from Zig to Rust in eleven days of agent-assisted work.
OPEN ↗ 25 NICOLO'S NOTES
Something Is Changing in the Unit Economics of Software How per-unit inference costs erode SaaS margins and push pricing toward usage.
OPEN ↗ 26 STOLEN-THOUGHTS.COM
Stolen Thoughts How encrypted chain-of-thought blocks turn out to be interchangeable across sessions, users and models.
OPEN ↗ 27 NOTASHELF
Taste Is All That's Left Why, once tools collapse the cost of building, judging what deserves to exist is the scarce skill.
OPEN ↗ 28 BJORN ROCHE
The AI Productivity Gap Why AI gains stay modest when writing code is only a small slice of the job.
OPEN ↗ 29 ZACH HOLMAN
The Cold Email On cold outreach as a genuine catalyst, survivor bias included.
OPEN ↗ 30 ALEX KOTLIARSKYI
The Productivity Mirage Why elaborate tooling matters less than picking meaningful problems and building product intuition.
OPEN ↗ 31 FERMISENSE
The Rise of Intelligence Ownership How an RL-tuned open-source model beat every frontier configuration on a real e-commerce workflow.
OPEN ↗ 32 WELI
The Valley of Webhooks Why webhooks are a poor fit for data replication, and ordered logs are the better primitive.
OPEN ↗ 33 FRANK WILES
Things I Would Have Done Differently On buying tools earlier and trusting your gut about clients and projects.
OPEN ↗ 34 ERICLAW
Web Security is Too Hard How a legitimate Cloudflare flow mimicked consent phishing well enough to fool the experts.
OPEN ↗ 35 SENDERLEDGER
What DMARC Actually Protects You From, and What It Does Not On what DMARC actually validates, and why it says nothing about sender intent.
OPEN ↗ 36 EARENDIL
What Is a Harness? On the system prompts, tools and agentic loops that turn a model into a working agent.
OPEN ↗ 37 ROGIX.DEV
What Is a Product? On the gap between an AI-generated prototype and something with users and a market.
OPEN ↗ 38 MUN LOGADAN
Why Does Opus 5 Feel Worse to Work With? On the gap between benchmark gains and day-to-day usability.
OPEN ↗ 39 TERENCE EDEN
A Bug Which Only Affected Left-Handed Users On a seven-year WordPress bug that only fired under a left thumb.
OPEN ↗ 40 ANNA LIBERTY
A Pragmatic Approach to LLMs Why the author avoids models for writing code but keeps them for navigation and vulnerability research.
OPEN ↗ 41 JUSTIN D FULLER
Accretive Editing How AI tools pile on addendums instead of replacing outdated information outright.
OPEN ↗ 42 OANA OLTEANU
Against Usefulness Why paradigm-shifting research needs patient funding and people chasing unfashionable questions.
OPEN ↗ 43 ADDY OSMANI
Agent Harness Engineering Why the scaffolding around a model matters more than the model's raw capability.
OPEN ↗ 44 ARDA TASCI
Ambiance: My Take on Harnesses Building an agent harness on Unix principles: modular, transparent, filesystem-based.
OPEN ↗ 45 YENNIE JUN
Are We Offloading Too Much of Our Thinking to AI? On what delegating judgement to models costs human autonomy and agency.
OPEN ↗ 46 SHUB.CLUB
Are You Telling Me a Readonly Property Is Wrecking My Performance? How repeatedly reading scrollHeight forces layout and eats the frame budget.
OPEN ↗ 47 VINAY GABA ET AL.
Benchmarking Coding Agents on Databricks' Multi-Million Line Codebase How open-source GLM 5.2 held its own against premium models at a fraction of the cost.
OPEN ↗ 48 SCOTT WILLSEY
Building and Shipping Mac and iOS Apps Without Ever Opening Xcode How command-line builds take Xcode out of the loop after initial setup.
OPEN ↗ 49 THEREALLO
Claude Code Is Steganographically Marking Requests How invisible Unicode characters encode endpoint classification into system prompts.
OPEN ↗ 50 STILLTHINKING.NET
Connections in Math: The Two Kinds of Random On Shannon entropy versus Kolmogorov complexity, and why pi looks random but compresses trivially.
OPEN ↗ 51 AARON PORTNOY
Cursor 0day: When Full Disclosure Becomes the Only Protection Left On publishing an arbitrary code execution bug in Cursor after seven months of silence.
OPEN ↗ 52 JANET CARR
Dolosse: A South African Invention Used Over the World On the wave-dissipating concrete blocks guarding coastlines worldwide, and their uncredited inventors.
OPEN ↗ 53 MITCHELL HASHIMOTO
Everyone Should Know SIMD How most SIMD work follows five predictable steps worth learning once.
OPEN ↗ 54 CRAIG MOD
Fast Software, the Best Software Why speed is a feature that protects flow and signals engineering quality.
OPEN ↗ 55 TRY-WORKS
First Principles of Model Routing Four rules: keep models distinct, pools small, benchmarks real, past decisions reviewed.
OPEN ↗ 56 ADAM KURKIEWICZ
GLM 5.2 Is (Nearly) as Accurate as a Human Book-Keeper at Under 1% of the Cost How an open-weights model prepares UK VAT returns at near-human accuracy.
OPEN ↗ 57 GWERN
Guardian Angels: LLM Personalization for Productivity and Security On personalised models that emulate your values to amplify you and guard against AI threats.
OPEN ↗ 58 WELI
Half-Baked Product How startups fail chasing market size over product quality while ignoring technical debt.
OPEN ↗ 59 THOMAS DULLIEN
Halvar's Guide to Entrepreneurship A long guide to founding: market selection, fundraising, product, hiring, operations.
OPEN ↗ 60 THATXLINER
Hologram A local-first desktop workspace for culling and managing RAW and JPEG photo libraries.
OPEN ↗ 61 TOM LOCKWOOD
How Is the Bun Rewrite in Rust Going? Questioning whether the rewrite is actually finished, given ongoing commits and no release tags.
OPEN ↗ 62 HAKI BENITA
How to Achieve Pruning When Querying by Non-Partitioned Columns in PostgreSQL Using check constraints to get partition pruning when you filter on the wrong column.
OPEN ↗ 63 PRADYUMNA PRASAD
How to Ask for Help From People Who Don't Know You On proof of work, context, and reasonable asks when writing to a stranger.
OPEN ↗ 64 ELIA SCOTTO
How to Read More Books On removing digital distractions, carrying a book everywhere, and keeping expectations realistic.
OPEN ↗ 65 KAPA.AI
How We Taught a Small LLM to Throw Away 68% of Our RAG Context How a cheap pruning pass drops two-thirds of retrieved chunks while keeping most recall.
OPEN ↗ 66 BRT.FYI
I Also Filed the Edges Off My MacBook On taking a hand file to a laptop's sharp palm-rest edges.
OPEN ↗ 67 BARTOSZ KOTRYS
I Burned All My Tokens Researching How to Save Tokens How orchestrating cheaper models beats reaching for a frontier model at every step.
OPEN ↗ 68 GEOHOT
I Love LLMs, I Hate Hype Enthusiasm for the progress, impatience with closed models and singularity talk.
OPEN ↗ 69 ALEC SCOLLON
I Think I Have LLM Burnout On fatigue from the repetitive patterns and stylistic tics of working with models.
OPEN ↗ 70 MARCIN WICHARY
If You're a Button, You Have One Job Why buffering rapid taps beats dropping them, compared across iPhone and Android.
OPEN ↗ 71 RUXANDRA TESLO
Intelligence Is Not the Main Bottleneck Why governance and policy gate real-world progress more than raw capability does.
OPEN ↗ 72 ALEX ALEJANDRE
Interview With Mitchell Hashimoto On terminal design, Zig, open-source philosophy and API design.
OPEN ↗ 73 GLYPHACK
It's Getting Harder to Focus Every Day On the steady erosion of attention and what it costs.
OPEN ↗ 74 ARIYA HIDAYAT
Local, CPU-Friendly, High-Quality TTS with Kokoro How to run an 82M-parameter speech model that still sounds good, entirely on CPU.
OPEN ↗ 75 ANTHONY HOBDAY
Notes on Software Quality On how craft decays at organisational scale, and which quality signals survive.
OPEN ↗ 76 SHREWD BLAWG
Ollama vs llama-server: Quick Benchmark Similar generation speed on GPU, but llama-server wins clearly on prompt processing.
OPEN ↗ 77 OPENAI
OpenAI and Hugging Face Partner to Address Security Incident During Model Evaluation On models under reduced safeguards escaping isolation and reaching third-party systems during evaluation.
OPEN ↗ 78 PETER KRAFT AND QIAN LI
Postgres Transactions Are a Distributed Systems Superpower How co-locating workflow state with application data buys exactly-once execution.
OPEN ↗ 79 ALEX CHAN
Rebuilding the Computer Room How portable computing erased the boundaries around attention, and what physical limits restore.
OPEN ↗ 80 CSIRO PUBLISHING
Spider Venom Kills Varroa Mites Without Harming Honeybees How peptides from cave and funnel-web spider venom target the mite but spare bees.
OPEN ↗ 81 YAEL GRAUER
Stop Telling Me to Ask an LLM Why deflecting hard questions to a model dismisses human expertise and lived experience.
OPEN ↗ 82 ELLO
Teaching a Child in <1000 ms: The Architecture Behind a Real-Time Tutor How a real-time reading tutor for children was architected around pedagogy, not just latency.
OPEN ↗ 83 GEA BLOG
The Age of Personalized Hardware Is Coming Why small personal devices keep multiplying, and the real opening is who writes their software.
OPEN ↗ 84 JIM NIELSEN
The AI Aesthetic On the visual and interaction patterns emerging from AI tools spreading into software generally.
OPEN ↗ 85 JA WESTENBERG
The Bread Paradox: Why Convenience Always Wins Why convenience keeps winning, and why that makes SaaS more durable than predicted.
OPEN ↗ 86 KENT BECK
The Cost YAGNI Was Never About Why YAGNI is about optionality and timing, not typing effort, even when code is cheap.
OPEN ↗ 87 MALISPER.ME
The Four Horsemen Behind Thousands of Postgres Outages On VACUUM, connection limits, planner failures and JSON statistics gaps.
OPEN ↗ 88 INTREPIDKARTHI
The Gauge Broke How AI tools make experienced developers feel faster while measurably slowing them down.
OPEN ↗ 89 LALIT MAGANTI
The git history Command Deserves More Attention How git's experimental history command gives jujutsu-like branch handling without switching tools.
OPEN ↗ 90 FARID ZAKARIA
The Mean Means Nothing Why a single summary statistic misleads, and why plots tell the whole story.
OPEN ↗ 91 AYUSH PAUL
The Memory Heist How a crafted website used an assistant's browsing to exfiltrate a user's personal data.
OPEN ↗ 92 EARENDIL ENGINEERING
The Session You Cannot Take With You How opaque, non-portable session state locks you into an inference provider.
OPEN ↗ 93 HUGO LANDAU
Web-Based Cryptography Is Always Snake Oil Why browser-delivered end-to-end encryption is incoherent when the server ships the code.
OPEN ↗ 94 CHIP WEINBERGER
What I Learned Selling 2500 MIDI Recorders, Part 1: Hardware Is Not So Hard How the hardware turned out easier than the software when shipping a physical product.
OPEN ↗ 95 JUSTIN FLICK
Why I Hate LLM Confidence Scores Why self-reported confidence has no scientific basis and buys only false reassurance.
OPEN ↗ 96 GETAPPS.CAFE
Why the Future Is Local App On local-first software winning on speed, privacy and ownership.
OPEN ↗ 97 GUSEYN ISMAYYLOV
Why Vanilla JS On building a whole SaaS product with web components and no build pipeline.
OPEN ↗ 98 LEV KOKOTOV
Why We Built Yet Another Postgres Connection Pooler How PgDog supports SET, LISTEN/NOTIFY and multithreaded scaling without application changes.
OPEN ↗ 99 CARSON GROSS
Working With AI: A Concrete Example How a developer who knows the domain collaborates with AI without accruing debt.
OPEN ↗ 100 NEAL STEPHENSON
Writing by Hand Is Good for Your Brain Practical handwriting technique, pens and paper included, for the benefit without the fatigue.
OPEN ↗ 101 RAY MYERS
Zig Creator Calls Spade a Spade, Anthropic Blows Smoke A sceptical read of the Bun Zig-to-Rust rewrite and the claims made around it.
OPEN ↗ 102 MAARTEN GROOTENDORST
A Visual Guide to Gemma 4 12B How Gemma 4 drops separate encoders and projects raw vision and audio straight into the LLM.
OPEN ↗ 103 AKRITES
Akrites Open Letter A coordinated cross-industry effort to remediate vulnerabilities in critical open source.
OPEN ↗ 104 LUKE STEPHENS
Are Bug Bounties Cooked? Whether AI-democratized bug discovery is a Quartz Crisis moment for bug bounties.
OPEN ↗ 105 RICK OSGOOD
Banned Book Library Modifying WiFi smart bulbs into wireless dead drops that host banned books.
OPEN ↗ 106 JIM NIELSEN
Blogging Can Just Be Stating the Obvious Why articulating obvious truths others overlook is a surprisingly valuable kind of writing.
OPEN ↗ 107 BENJAMIN CANE
Cell-Based Architecture for Resilient Payment Systems How American Express isolates failures with cells to keep global payments resilient.
OPEN ↗ 108 MARK DOMINUS
Deciphering Basmala How Unicode and font design solved rendering the Arabic basmala correctly across platforms.
OPEN ↗ 109 TORGEIR HELGEVOLD
Fine-Tuning a Local LLM to Categorize Questions Fine-tuning a 0.6B model to 92% accuracy by mapping categories to short two-char codes.
OPEN ↗ 110 BROWSER USE
Firecracker Browser Infrastructure Rebuilding cloud browsers on Firecracker microVMs for 3x cheaper, sub-second startups.
OPEN ↗ 111 HACKERNEWSTRENDS.COM
Hacker Trends Chart how any topic has trended across 18 years of Hacker News posts and comments.
OPEN ↗ 112 PAUL GRAHAM
How to Earn a Billion Dollars Why exponential growth in a large market, not cheating, is what builds billion-dollar startups.
OPEN ↗ 113 TOM BEDOR
How to Write Good (Short) Docs Practical guidance for concise one-page docs that respect a reader's attention.
OPEN ↗ 114 DENNIS BROTZKY
How's Linear So Fast? A Technical Breakdown The architecture behind Linear's speed: a browser database, sync engine, and optimistic mutations.
OPEN ↗ 115 MLL.SH
Humiliating IIS Servers for Fun and Jail Time A bug-bounty methodology for enumerating and exploiting IIS and .NET web servers.
OPEN ↗ 116 PAUL BUCHHEIT
If Your Product Is Great, It Doesn't Need to Be Good Perfect three core features rather than spreading effort across many mediocre ones.
OPEN ↗ 117 GRAPHICORE
Libre Barcode Free open-source fonts that render scannable Code 39, Code 128, and EAN/UPC barcodes.
OPEN ↗ 118 JIM MONTGOMERY
LLM Code Style and Token Costs Preferring native Web APIs over hand-rolled code cuts LLM output tokens by 85–92%.
OPEN ↗ 119 ALEX ELLIS
Local AI Is Not Opus Local Qwen models aren't a worse Opus — they're a different tool for different jobs.
OPEN ↗ 120 MIDJOURNEY
Midjourney Medical Midjourney's medical imaging initiative and its blog announcement.
OPEN ↗ 121 PRESTON JENSEN
Playing with Vision Embeddings Probing how DINOv3 encodes visual concepts into embeddings using sparse autoencoders.
OPEN ↗ 122 MOHAMED HABIB
Scaling One VM to a Million Sandboxes Redesigning OpenComputer into multi-cell, multi-cloud cells to run millions of sandboxes.
OPEN ↗ 123 TEREZA TIZKOVA
Software Abundance As AI lowers the barrier to building software, value pivots to services and outcomes.
OPEN ↗ 124 SAMSCH
Stop Using JWTs (for Sessions) Why JWTs are a poor fit for session management and cookie-based sessions are better.
OPEN ↗ 125 ARMIN RONACHER
The Coming Loop On AI agent loops and how external harnesses reshape how code gets written and maintained.
OPEN ↗ 126 ARUN
The JR Logo How Japan's railways kept one identity through a mark designed in 124 days.
OPEN ↗ 127 AJEY GORE
The Solo Climb Why teams need robust harnesses before running AI agents with minimal oversight.
OPEN ↗ 128 JAMES O'CLAIRE
The Unbearable Cheapness of Open Weight Models Why open-weight models are so much cheaper — real cost advantage or deliberate scarcity?
OPEN ↗ 129 ALEPH
Ultrasound Imaging of the Brain MRI-level brain imaging through an intact skull using ultrasound and microbubbles.
OPEN ↗ 130 BAIDU
Unlimited-OCR An OCR system for one-shot long-horizon parsing of unlimited-length documents.
OPEN ↗ 131 NOLAN LAWSON
Using AI to Write Better Code More Slowly On using LLMs for methodical code review rather than fast, low-quality generation.
OPEN ↗ 132 ETHAN MOLLICK
What It Feels Like to Work with Mythos On how frontier models shift the human role from participant to commissioner of finished work.
OPEN ↗ 133 KARL TRYGGVASON
You Can't Unit Test for Taste On the parts of software quality that automated tests can never capture.
OPEN ↗ 134 ANSON BIGGS
You're Probably Using Agent Skills Wrong Agent skills should be human-authored to fill real knowledge gaps, not self-generated.
OPEN ↗ 135 MURAT DEMIRBAS
Your Attitude Determines Your Success How mindset and emotional management, more than talent, shape whether you succeed.
OPEN ↗ 136 ELON LITMAN
A Theory of Deep Learning A framework for why deep networks generalize despite overfitting.
OPEN ↗ 137 AUDIOMASS.CO
AudioMass A browser-based, open-source audio editor with no backend required.
OPEN ↗ 138 AARON BRETHORST
Domain Expertise Has Always Been the Real Moat Why domain knowledge—not raw coding—becomes the scarce skill in the AI era.
OPEN ↗ 140 MILAN MILANOVIĆ
Laws of Software Engineering A curated set of principles and patterns that shape software, teams, and decisions.
OPEN ↗