I enjoyed it. Given you've given me much to read, I hope you will excuse me giving you much to read in return :) Reading it reminds me of the timelessness of such views. Both about the noise and on how to educate children. The ancient Greeks wrote the first anti-noise legislation, and Seneca complained about the chaos of living above a bathhouse. Later, in Rome, Juvenal said only the wealthy could afford to sleep, in contrast with how the chaos in the streets below caused insomnia. As for education, Aristotle wrote in his Rhetoric how young people have yet to be humbled by life, thinking they know everything while being sure about it. Plato wrote in the Republic how teachers both fear and flatter their students and how students despise them in return. I must admit western philosophers in general remain an understudied domain for me, including the Greek, and even more so with the enlightenment ones. Not for lack of trying or reading, but the underlying feeling they miss out on something fundamental which remains hard to either pinpoint or word out. A sort of ineffable human nature which resists being exposed by labels alone. I always felt more drawn towards the eastern ones, which made for a pleasant surprise in learning Schopenhauer read the Bhagavad Gita and the Upanishads. I am on my second read of the Gita and have yet to start on the Upanishads, so my views might also be incomplete in understanding them. They take far more time to embody than to study. I can, however, already draw interesting parallels between the Gita and Schopenhauer's views. He does make interesting points about observation and experience; a direct mirror to the two paths of contemplation and action. I also can attest first hand on the downsides of living near noise, and how much worse jackhammers sound. I lived for years with them ranging from 20 meters away to a few blocks away, always thinking "eh, it'll only last a bit longer, and I already quit work to self-fund this, I can't afford to move now", quite naively I might add! Although where I disconnect comes from both the presentation and the contradictions I can't help but notice. I feel I can skip over the former, others have written at length about it. The contradictions make for a fascinating study, at least to me. I think I can also skip over the "going mad" part, it seems to be a symptom of something else entirely. At least, from reading The Universal Computer, madness came to just about everyone who contributed significantly to the path leading to our modern thinking machines, regardless of their external environments. They too faced contradictions, with the exception their work forced these people to address them head on. Cantor in particular found paradoxes, which influenced Russell to find more paradoxes, and led to the latter quitting both Principia Mathematica and, well, mathematics in general. On the plus side, it led Whitehead to write Process and Reality, which inspired Rich Hickey in creating Clojure and Datomic, and in turn inspired me profoundly by using both to build with. My issue with western philosophy seems also split two, where I can see opposite factions of philosophers: the external critics and the internal critics. The latter applying their own frameworks back over themselves. The discerning factor I'm seeing here comes down to self-referentiality, which became the bread and butter of Hofstadter and makes a major link back to eastern philosophies going back millennia. The way I apply it asks whether a philosopher embodies the philosophy. The view it leads to appears to show pure thinkers and theorists on the external criticism side, while the engineers and artists sit comfortably in the internal one. In that regard, Schopenhauer seems to fall in the external critic camp. He documents his observations and experiences with surgical precision, but asks others to perform it rather than lead by example. He writes about avoiding errors, that mathematics make errors impossible, yet Gödel proved even basic arithmetic shows incompleteness and gauge theory boils down to using errors as compute power. Elon keeps proving errors make the impossible happen, and even Bob Ross called them "happy little mistakes". Schopenhauer also writes how ideas such as languages, natural sciences and history show little danger, which contrasts directly with "in the beginning was the word", the power of programming languages today and how historical narratives nudge entire cultures over time. Sartre makes another example of "we are what we will ourselves into" and then wills himself into ruminations and misery. I tend to avoid strict rules and absolutes, in part because I grew up with them, and studying figures like Alexander the Great shows he revolted so strongly against such an upbringing he ended up conquering a large portion of the world. Also in part because engineering experience taught me how rules and absolutes become the first things to break when hitting reality. Software which passes all the compilation checks and the full test suite still meets production as if the previous safeties act as mere suggestions. Also because a friend keeps pushing me to ship more and "yes those are all interesting ideas, but still very abstract, a demo would go a long way with this" which I end up grateful for even if it stings in the moment. It makes another interesting contradictory paradox, I think, in that Schopenhauer calls out the dangers of abstract ideas over first hand experiences and observations, while writing about it in abstract idea form. I still have no good answer to resolve this dilemma in myself. On one side I realize staying in the abstract shields me from shipping what I have in mind, on the other I realize going to market too soon won't solve the problems I experienced at work during the previous decades. On the one side I view countless hours of writing code and debugging, and on the other I view countless hours of studying abstract math towards what I believe leads to a much better way of writing software, from the ground up. And until both ends meet and merge, I feel uncomfortable shipping or going to market, as it would lower the potential from what I envision to what I have now. I end up approaching it the same way as music practice: most notes going into the void, most drafts discarded, most phrasings abandoned. Which leads me to the points Schopenhauer makes on selection by most capable people and following masters. I keep questioning how we arrive at the state of the art and mastery. Robert Greene's Mastery helped a ton to drive my views in this regard. Also the realization having extensive experience across just about every domain of software grounds me in a way such selection and mastery can now use reality itself as a co-author. Because nothing else breaks down ideas and prototypes faster than going beyond the theory and into field work. Examples pour in from history, such as the steam engine whose engineering problems led to discovering thermodynamics, to name only one. Which ends up, by doing both the engineering and the theoretical sides at once, being stuck between a rock and a hard place, whose pressure forces invention by the mother which we call necessity. And speaking of pressure, this now gets back to the original point: noise. I do my best to avoid complaining about it, and instead view it as a means to practice detachment as written in the Gita. Easier said than done, while also leading to accumulating personal motivation and discipline from every source I can get my hands on. Jocko's philosophy of GOOD, Elon's success after countless failures with the world's experts saying it can't be done, Carlos Castaneda's book series which I am still only halfway through fascinates me, the I Ching where failure gets reframed as simply the process of changes which science now calls calculus, Scott Adams' insane ability to re-frame any situation to his advantage, even your "read what you love until you love to read" got me back into reading which I enjoyed so much as a kid, and this list also grows endless real fast. I ask myself what lessons can life teach me in those moments, learn to appreciate what I have more than ever before, and how to do what I can given what I have. Even Arnold Schwarzenegger's Be Useful philosophy inspires me, especially the part about coming up with solutions if one wants to complain about issues. And then putting those solutions to the test and shipping them once they reach a tipping point of a working state. Possibly the hardest path I can imagine putting myself on, but as everyone I admire would say: easy isn't worth doing. And then when I can afford to move, I feel as if nothing will be able to distract me anymore, given what I've already been through. Everything serves a purpose. And as Jeff Bezos would say, today marks another Day 1.

the lens rides in (quixotically) on a cap that bounds what it may read and write. foreign code in your transaction with exactly the authority you granted and not a datum more. its the part eBPF and datomic never had

the survivors restricted the code (verified bytecode, fixed op menus, sandboxed lua) but none of them carry authority. datomic tx fns run as root over the whole db, eBPF doesn't know who you are. so nobody ships computation across org boundaries, and the five coping mechanisms above are what you get instead

the fix has been known since BPF (1992): ship the computation to the data. redis lua, datomic tx fns, stored procedures, eBPF. the modify step runs inside the data's own transaction so there is nothing to be stale against. obviously

First words of agent running in a live Elixir image on a Datomic-style "fact log"

If we're talking about weights, I'd recommend something like Datomic to get a scrub history and do a lock on anything delete oriented (by nature it doesn't do that given it's immutable on record). You'd want special constraints and care to avoid bypass attempts. Just an idea.

In my experience as perf engineer, CPU usage has been the main resource services content for in clusters and I worked with DBs that keep *a lot* of data in memory (Manhattan at Twitter and Datomic at Nubank). Even file system usage is typically more problematic I can see use cases where the JVM’s profile becomes problematic like lambda functions but the JVM footprint tax isn’t a major issue in my experience, at least for large-scale systems

Democratizing Machine Learning at Netflix: Building the Model Lifecycle Graph When Netflix began investing in ML over a decade ago, it was focused on a single domain: personalization. Today, ML runs across personalization, studio, payments, ads, and more, each with its own tech stack, metrics, and org structure. That growth introduced a new challenge: enabling cross-pollination of models and data across domains. Without any discovery infrastructure, ML practitioners couldn't easily collaborate or share work across business verticals. Their tools existed in silos. The model registry was unaware of which A/B tests were using its models. The pipeline orchestrator was unaware of downstream model dependencies. Consider content embeddings: Studio teams created sophisticated embeddings for production workflows that could also be valuable for Ads (context matching) and Personalization (recommendations). Making that cross-pollination happen was extraordinarily difficult. The Netflix team's answer was the Metadata Service (MDS), which builds a Model Lifecycle Graph that indexes and connects ML-related entities across the company, including models, features, pipelines, experiments, and datasets. The graph answers cross-domain questions such as "Which experiments are running this model?" or "Which models share these features?" It was designed to make every ML asset at Netflix discoverable, understandable, and reusable by every ML practitioner, regardless of their team or domain. The architecture flows from event ingestion (via Kafka and AWS SNS/SQS) through entity enrichment, normalization, and storage in Datomic (for graph traversals) and Elasticsearch (for full-text discovery). Background enrichment jobs then walk the graph to infer relationships that no single source system could see on its own. The result: a query that once required manual checks across four separate systems (model registry, pipeline orchestrator, experimentation platform) is now a single graph traversal. By Saish Sali, Nipun Kumar, Sura Elamurugu netflixtechblog.com/democrat… #MLOps #MachineLearning #DataEngineering #DataLineage -- The Year of the Graph's Spring 2026 newsletter issue on all things #KnowledgeGraph, #GraphDB, Graph #Analytics / #DataScience / #AI and #SemTech is coming soon. Subscribe and follow to be in the know. Reach out if you'd like to be featured 👇 yearofthegraph.xyz/newslette…

does anybody want to rewrite Datomic in Rust?

自社製品の事例：イベントソーシングで変更履歴を残しつつ、各レコードの最新スナップショットを参照系に用いるアーキテクチャ いわゆる追記型や不変性で、関数型プログラミングにも向いていると思います。 最初にこの設計手法を知ったのは、ClojureとDatomicで作られたCapital Oneの事例から。

Resposta curta: não foi aquisição, foi hiring. Os talentos que fizeram o bun são raros. Outro ponto é o controle do roadmap, ao menos uma forte influência. Isso não é raro, no Brasil o Nubank fez o mesmo com a Cognitec, empresa que criou o Clojure e Datomic

Once people started using automated primary keys they fell into a land compatible with Datalog (every tuple is unique) but still only accessible with cumbersome SQL. Smarter people than me have tried to fight the power (Logica, Datomic), no dice.

A Datomic-like DB on SlateDB: Triplox! finnvolkel.com/triplox-log-1…

Datomic Pro 1.0.7266 now available! forum.datomic.com/t/datomic-… #Clojure #Datomic

Blogged about Triplox, a system I'm building on top of SlateDB. finnvolkel.com/triplox-log-1… #slatedb #datomic

$100 they will reinvent datomic at best

$NU's co-founder and former CTO Edward Wible, alongside Distinguished Engineer Lucas Cavalcanti, recently broke down insights into the structural architecture that allowed the digital bank to scale. As Cavalcanti noted from his prior industry experience, most traditional bank tech stacks age like milk. While Nubank's founding team initially wrote some early services in Ruby back in 2013 due to familiarity, they quickly hit scaling bottlenecks from the global interpreter lock and an immature library ecosystem. But the realization came from studying large system failures. The root cause of accidental complexity in massive codebases is mutable state and side effects. If you are building a retail bank that will eventually handle hundreds of millions of transactions, the database cannot just overwrite historical context. This is where Nu made an early, contrarian bet that became their strength. To avoid building a legacy system that would eventually rot, they anchored their core using Clojure as their backend language and Datomic as their database. Because Clojure runs on the JVM, it gave them instant access to decades of battle-tested Java libraries. This allowed the engineering team to focus entirely on business logic rather than building basic infrastructure from scratch. And then there is Datomic. When data changes in Datomic, it does not overwrite the old data. It appends the new state, preserving the complete history of prior states. For a heavily regulated financial institution, this immutable database is the holy grail. It creates a native, unalterable audit trail. Engineers, regulators, and auditors can rely on this database to view immutable snapshots of the ledger at any exact moment in time. It natively solves the hardest parts of double-entry accounting and allows engineers to replay chronological transactions to recover from errors without data corruption. This tech advantage was so critical to their compounding growth that Nubank outright acquired Cognitect, the creators of Clojure and Datomic, to help secure their moat. Today, this architecture processes hundreds of millions of transactions a day across one of the world's most active payment networks. Nu operates with horizontal read scalability that allows machine learning and analytics to run in real-time without breaking production systems. The tech stack is the structural advantage driving their unit economics, allowing them to ship products to over 130M people with nearly zero friction.

@nubank is one of the biggest digital banks in the world, serving over 130 million customers in Brazil, Mexico, and Colombia. Their tech stack? Clojure and Datomic. Learn why: ➡️ youtube.com/watch?v=pleo1VA9… @datomic_team

I don't believe so, but datomic or something like it could serve as foundational component of a distributed runtime. Distributed resources aren't limited to state. Distributed sensors, actuators, drone swarms, etc. should be contemplated for the notion of distributed runtimes.