Blog
Writing on GPUs, LLMs, MLOps, Kubernetes — and mindset · 3517 posts
#2026-03 765#english 592#culture 264#deep-dive 254#kubernetes 247#career 229#ai 216#llm 208#devops 193#2026-04 146#security 141#database 114#observability 113#communication 109#history 107#architecture 100#productivity 96#finance 88#economy 84#mindset 81#psychology 80#ai-papers 79#food 78#it 78#travel 78#deep-learning 77#japanese 77#networking 77#performance 72#business-travel 70#linux 70#gpu 69#ai-agent 66#cs-fundamentals 63#postgresql 60#rag 58#self-improvement 55#learning 53#mlops 53#ai-platform 51
One Issue, the Whole Supply Chain — How an Agent Inside CI Broke, and What the Defenses Actually Bought
The Claude Code GitHub Actions vulnerability that GMO Flatt Security researcher RyotaK disclosed on June 1, 2026 traces, end to end, how an agent dropped into a CI pipeline can become the channel that hands over an entir
2026-07-16 · 21 min read #security#ai#prompt-injection#supply-chain#ci-cdDebian Has Started Blocking Unreproducible Packages From Migrating to Testing — What the Gate Actually Blocks, and What It Doesn't
On May 9, 2026, Debian's release team wired its migration software, britney, to reproducibility checks. The sentence that hit the mailing list the next day was "Debian must ship reproducible packages," but what's actuall
2026-07-16 · 20 min read #devops#reproducible-builds#debian#supply-chain#ci-cdWhat ROV Blocked, What ASPA Still Can't — the June 2026 Telegram Hijack and the Next Layer of Routing Security
On June 16, 2026, India's Rcom (AS18101) hijacked Telegram's prefixes, and Kentik's measurement shows the hijacked path reached only 1.6% of Kentik's own BGP sources — a case where RPKI ROV actually worked. But that's be
2026-07-16 · 23 min read #network#bgp#rpki#routing#securityClang 22's MSVC ABI Change — the Vector Deleting Destructor and the delete[] Heap Corruption Still Alive in 22.1.x
Clang 22.1.0 (2026-02-24) changed destructor codegen in two places when targeting the MSVC ABI. One aligns ::delete handling with MSVC; the other is vector deleting destructor support, closing an issue opened in 2014 aft
2026-07-16 · 17 min read #llvm#clang#cpp#compilerPySpark 4.2 Puts Python UDFs on Arrow by Default — the Story of 132 out of 225 Type-Coercion Cells Changing
Apache Spark 4.2.0, released on July 14, 2026, flipped the default of spark.sql.execution.pythonUDF.arrow.enabled from false to true under SPARK-54555. Vendor blogs describe this as a faster columnar path without rewriti
2026-07-16 · 16 min read #spark#python#arrow#data-engineeringbcachefs Drops the Experimental Tag from Erasure Coding — and What It Actually Means
On September 29, 2025, Linus Torvalds erased 117,483 lines of bcachefs code from the kernel in a single commit, and bcachefs became a filesystem shipped as a DKMS module, like ZFS. That was the end of the drama — the rea
2026-07-16 · 17 min read #linux#kernel#filesystem#bcachefs#storageThe 3-Hour axios Account Takeover — Provenance Was On, and Nobody Checked It
At 00:21 UTC on March 31, 2026, a malicious version 1.14.1 landed on axios, a package with over 80 million weekly downloads. What makes this case interesting is that axios was already using npm trusted publishing — the l
2026-07-16 · 18 min read #security#supply-chain#npm#nodejs#devsecopsOpenTofu 1.12's Dynamic prevent_destroy — Where the Lifecycle Block's Static Constraint Loosens, and What It Costs
Terraform's lifecycle block accepts only literal values by design. HashiCorp's docs explain why: lifecycle settings are processed while building the dependency graph, before variables can be evaluated — and the request t
2026-07-16 · 12 min read #opentofu#terraform#iac#devops#state-managementCXL Memory Tiering: How Far Has the Kernel Come — What DAMON Merged, and What the Numbers Didn't Measure
Even when CXL hardware is plugged into a server, it's the kernel that actually decides which pages are hot and which are cold and moves them. As of July 2026, Linux mainline has received DAMON/DAMOS-based tiering pieces
2026-07-16 · 19 min read #hardware#cxl#memory#linux-kernel#performanceElixir 1.20's Gradual Type System — The Reality of Finding 'Verified Bugs' Without Annotations
Elixir v1.20, released on June 3, 2026, closed out the first milestone of the set-theoretic type system work that began in 2022. Every Elixir program is now gradually type-checked without a single type annotation, and th
2026-07-16 · 17 min read #elixir#type-system#beam#compiler#static-analysisEROFS Native Layers in containerd 2.3 — What You Get in Exchange for Removing Unpack
containerd 2.3, released on April 30, 2026, opened a path to pull an EROFS filesystem image straight from the registry and mount it as-is. Since there's no tar to unpack, the entire unpack step disappears, leaving behind
2026-07-16 · 18 min read #container#containerd#oci#storage#linuxThe Craft of a Forward Deployed Engineer: Requirement Archaeology, Domain Modeling, and Optimizing for Adoption
A previous post mapped the knowledge an FDE needs — Linux, networking, Kubernetes, databases, security. That map gets you in the door. It says nothing about what you actually do once you are inside a customer's building,
2026-07-15 · 10 min read #career#fde#solutions-engineering#software-engineering#palantirFrom Documents to a Knowledge Graph: An Honest Pipeline
'Extract a knowledge graph from your documents' looks like a single LLM call in a demo. But turning a customer's documents into a graph you can actually query is a six-stage pipeline, and most of the cost and pain lives
2026-07-15 · 10 min read #knowledge-graph#ai#llm#data-engineeringA Map of Knowledge Graph Tools and Frameworks: What to Pick, When
Graph databases, the RDF and ontology stack, and graph RAG frameworks have exploded over the last two years. Link lists are already abundant, so this post draws an honestly-opinionated map instead. For each category: wha
2026-07-15 · 9 min read #tools#knowledge-graph#graph-rag#ontology#frameworksTurning a Customer Domain Into a Model — From Ubiquitous Language to Ontology
A customer's domain knowledge usually lives not in documents but as tacit knowledge inside people's heads. One core job of a Forward Deployed Engineer is turning that scattered tacit knowledge into an explicit model the
2026-07-15 · 11 min read #ontology#knowledge-graph#domain-driven-design#architecture#forward-deployed-engineerGraph RAG, Explained: What It Is and When It Earns Its Cost
The standard RAG recipe — chunk, embed, retrieve top-k — works when the answer sits inside a single chunk, but it stalls structurally on multi-hop questions and on global sensemaking questions that span the whole corpus
2026-07-15 · 8 min read #rag#graph-rag#knowledge-graph#ai#llmFive Habits for Working Well with AI Coding Tools
The same tool produced a 55.8% gain in one experiment and a 19% loss in another. What flipped the sign was not the tool, but how it was used. Five habits drawn from METR, the GitHub Copilot RCT, the Stack Overflow survey
2026-07-12 · 10 min read #ai#productivity#software-engineering#developer-experienceWhat to Learn Deeply, What to Skim — A Learning Strategy for the AI Era
When an AI answers almost anything in three seconds, what is still worth learning deeply? Cognitive psychology has an uncomfortable answer: memory is built by pulling information out of your own head, not by having it ha
2026-07-12 · 12 min read #career#learning#ai#software-engineeringFrom Senior to Staff: What Actually Gets Evaluated
Plenty of engineers believe that shipping more, faster, is the path to promotion. But open an actual public engineering ladder — Dropbox, CircleCI, Rent the Runway — and there is no axis called output. There is scope, co
2026-07-12 · 12 min read #career#software-engineering#growth#staff-engineerWhen Code Generation Gets Cheap, Which Skills Go Up in Value — What Depreciates and What Appreciates
When code generation becomes cheap and abundant, value does not vanish — it moves to whatever is still a bottleneck. Right now that bottleneck is verification, judgment, and integration. Building on Jason Wei's asymmetry
2026-07-12 · 14 min read #career#software-engineering#ai#skills#code-review