[{"data":1,"prerenderedAt":674},["ShallowReactive",2],{"post-en-history-repeats-itself":3,"related-en-history-repeats-itself":232},{"id":4,"title":5,"body":6,"categories":211,"coverImage":213,"date":214,"description":215,"extension":216,"lang":217,"legacy":218,"meta":219,"navigation":220,"path":221,"seo":222,"stem":223,"sticky":218,"tags":224,"translation":24,"updated":230,"__hash__":231},"articles/en/history-repeats-itself.md","History repeats itself",{"type":7,"value":8,"toc":201},"minimark",[9,16,27,35,40,43,46,59,66,69,72,76,82,85,92,96,99,105,108,112,134,141,148,154,158,161,167,170,173,178,182,188,198],[10,11,12],"p",{},[13,14,15],"em",{},"Graph databases, MongoDB, dedicated vector databases. The pattern is always the same. Only the names change.",[10,17,18],{},[13,19,20,21,26],{},"(This is an English adaptation of my Italian essay ",[22,23,25],"a",{"href":24},"/la-storia-si-ripete","La storia si ripete",".)",[10,28,29,30,34],{},"In the ",[22,31,33],{"href":32},"/en/the-cms-in-c-i-never-wrote","previous post"," I told the story of how, in 2015, I chose \"boring old\" WordPress to build a platform used by tens of thousands of people. But there's another story from that era worth telling.",[36,37,39],"h2",{"id":38},"_2015-graph-databases-were-the-future-nosql-was-the-present","2015: graph databases were the future, NoSQL was the present",[10,41,42],{},"Every conference, every blog post, every enthusiastic dev. \"Foreign keys don't scale. Joins are old stuff. The world is made of connections.\" Neo4j, Cypher, slide decks full of colorful nodes and arrows.",[10,44,45],{},"On the same anti-relational-database war footing, millions of people piled onto a single combo that had practically become a synonym for \"startup\": Node.js + MongoDB.",[10,47,48,49,52,53,52,56],{},"I used to attend a lot of technical demo days where CTOs of other startups explained how they'd set up their stack. I still remember one opening with the classic line of the era: \"we're using Node paired with Mongo!\" Then he started describing the domain: ",[13,50,51],{},"\"the relationship with\"",", ",[13,54,55],{},"\"the relationship of\"",[13,57,58],{},"\"this is connected to that through that other thing.\"",[10,60,61,62],{},"I said: ",[63,64,65],"strong",{},"\"Wouldn't it have been crazy to store relationships in a... relational database?\"",[10,67,68],{},"Today I use Postgres for almost everything. When I need something extra, I reach for extensions on top of the same database: pgvector for vectors, TimescaleDB for time series, pg_textsearch for full text with ranking.",[10,70,71],{},"Graph databases: zero.",[36,73,75],{"id":74},"what-happened","What happened",[10,77,78,79],{},"Nothing dramatic. ",[63,80,81],{},"Postgres kept moving forward.",[10,83,84],{},"Recursive CTEs, jsonb, extensions for everything. It cannibalized 90% of the use cases without adding one more infrastructure component to maintain, sync, monitor, and pay for. The operational cost of a dedicated graph database isn't trivial: sharding is complicated, and the mental model differs from the relational one most developers have already internalized.",[10,86,87,88,91],{},"Result: a great tool for specific niches — enterprise-scale fraud detection, knowledge graphs in large organizations. For 95% of projects: ",[63,89,90],{},"overkill",".",[36,93,95],{"id":94},"the-same-story-earlier","The same story, earlier",[10,97,98],{},"Before graph databases, there was MongoDB.",[10,100,101,104],{},[13,102,103],{},"\"Joins don't scale\""," — repeated like a mantra by people who'd never actually had a scale problem. Two years later: an array of userIds inside the order document, some strings, some numbers, some ObjectIds, and 3% of the references missing, and nobody knows why.",[10,106,107],{},"MongoDB was genuinely useful for certain use cases, and I used it a lot myself (more often alongside something else than as a replacement). The problem was that it got used for everything, because it was new, because it was cool, because using it felt like a modern choice.",[36,109,111],{"id":110},"the-pattern-is-always-the-same","The pattern is always the same",[113,114,115,119,122,125,128,131],"ol",{},[116,117,118],"li",{},"New technology solves real problems in specific cases",[116,120,121],{},"It's presented with use cases from companies at a scale 99% of teams will never reach",[116,123,124],{},"It gets adopted as a general-purpose solution",[116,126,127],{},"Whoever uses the \"boring\" technology looks like a fossil",[116,129,130],{},"The reckoning arrives",[116,132,133],{},"The boring technology is still there; someone proposes a refactor \"to harden things\" (read: go back to solid schemas and contracts), and it takes a year",[10,135,136,137,140],{},"There's a concept called the ",[63,138,139],{},"Lindy effect",": a technology that's existed for twenty years will probably exist for another twenty. Coming from someone who's a tinkerer and an enthusiast by nature: know everything, experiment with everything, but in production... it depends.",[10,142,143,144,147],{},"Our job is to ",[63,145,146],{},"identify the contexts where a professional judgment call is actually needed",", factoring in risk areas like long-term support, the availability (and therefore cost) of people already skilled in that technology as the team grows, knowledge transfer, and the availability of extensions and libraries that are already battle-tested.",[10,149,150,153],{},[63,151,152],{},"And yes, I know reusing something you've used for 20 years is boring",", but sometimes it's also the most professional service we can render. For the thrill of novelty, a bit of patience is enough, and the universe will eventually send you the project where the new thing actually makes sense. In the meantime there are personal side projects, just for fun, as an outlet.",[36,155,157],{"id":156},"its-happening-again-now","It's happening again now",[10,159,160],{},"Pinecone, Weaviate, Chroma. Dedicated vector databases, presented as a necessary component of any AI architecture.",[10,162,163,164,91],{},"Same story here: it's not that they're useless. It's that in the vast majority of cases ",[63,165,166],{},"you don't need one",[10,168,169],{},"Hybrid full-text search with BM25 + pgvector solves most of it — with the advantage of having everything in the same database, relationships intact, zero synchronization, zero extra components to maintain. Up to roughly half a million records, the performance difference against a dedicated vector database is negligible. And most projects will never reach half a million records.",[10,171,172],{},"The cost of exotic infrastructure is always paid later, never up front. And \"later\" almost always coincides with the worst possible moment: real traction, investors watching, users complaining.",[10,174,175],{},[63,176,177],{},"History is cyclical. Only the names change.",[36,179,181],{"id":180},"my-mantra","My mantra",[183,184,185],"blockquote",{},[10,186,187],{},"Minimum diff, less moving parts.",[10,189,190,191,194,195],{},"I carry one rule with me across projects: ",[63,192,193],{},"minimum diff, less moving parts",". Every additional component is a dependency, something that can break, something with side effects that are sometimes implicit. Something someone has to understand, maintain, upgrade. ",[63,196,197],{},"The cost is never only technical — it's cognitive, operational, economic.",[10,199,200],{},"Choosing the most straightforward technology that solves the problem isn't a shortcut. It's the job.",{"title":202,"searchDepth":203,"depth":203,"links":204},"",2,[205,206,207,208,209,210],{"id":38,"depth":203,"text":39},{"id":74,"depth":203,"text":75},{"id":94,"depth":203,"text":95},{"id":110,"depth":203,"text":111},{"id":156,"depth":203,"text":157},{"id":180,"depth":203,"text":181},[212],"dev","https://i2.wp.com/enricodeleo.s3.eu-south-1.amazonaws.com/images/la-storia-si-ripete.png","2026-07-12T10:00:00.000Z","In 2015, graph databases were the future. Today I use Postgres for almost everything. In between: MongoDB, mixed ObjectIds, and the usual bill that arrives later. It's happening again with vector databases.","md","en",false,{},true,"/en/history-repeats-itself",{"title":5,"description":215},"en/history-repeats-itself",[225,226,227,228,229],"postgres","database","engineering","tech-debt","build-in-public",null,"4a6NwbGmxImrjuF8NnRM9gJr2yZIO2snsYaFjlFG3SY",[233,592],{"id":234,"title":235,"body":236,"categories":580,"coverImage":582,"date":583,"description":584,"extension":216,"lang":217,"legacy":218,"meta":585,"navigation":220,"path":586,"seo":587,"stem":588,"sticky":218,"tags":589,"translation":249,"updated":230,"__hash__":591},"articles/en/speed-without-judgment.md","Speed without judgment",{"type":7,"value":237,"toc":571},[238,243,251,254,261,264,267,272,279,283,286,293,307,314,320,323,327,333,340,343,354,357,364,372,375,379,382,389,400,403,410,415,418,421,428,432,435,438,441,444,450,453,458,461,465,471,474,481,484,495,499,505,508,519,524,527,530,535,539,542,549,552,563,566],[10,239,240],{},[13,241,242],{},"Vibe coding didn't invent badly written software. It just removed the friction.",[10,244,245],{},[13,246,20,247,26],{},[22,248,250],{"href":249},"/velocita-senza-giudizio","Velocità senza giudizio",[10,252,253],{},"I once saw a backend with authentication hardcoded into the frontend. Admin password in plain text, visible to anyone who opened DevTools. That wasn't vibe coding — it was 2018, human devs, hired off Upwork.",[10,255,256,257,260],{},"I mention it because what's happening now ",[63,258,259],{},"isn't new",". It's the same thing, faster.",[10,262,263],{},"I've seen Google Sheets used as a database. It collapsed at four concurrent users. The founders were surprised.",[10,265,266],{},"I've seen features declared \"done\" that were empty API endpoints with nothing behind them. I've seen human developers do it. I've seen Claude Opus 4.6 — Anthropic's most capable model right now — do it too. I've seen tests written to pass, not to verify.",[183,268,269],{},[10,270,271],{},"A system that certifies itself is more dangerous than one that fails openly.",[10,273,274,275,278],{},"The problem isn't the AI. It's that the speed it offers can make you forget that building software is ",[63,276,277],{},"fundamentally a cognitive act"," — not a writing one.",[36,280,282],{"id":281},"software-is-not-text","Software is not text",[10,284,285],{},"This is the thing the enthusiasm around vibe coding systematically forgets.",[10,287,288,289,292],{},"Software is not text. It's a ",[63,290,291],{},"formal model of reality"," — a system of rules describing how a domain behaves, how states change, how exceptions are handled, how time, concurrency, and failure are treated.",[10,294,295,296,299,300,303,304,91],{},"Writing code was never the hard problem. The hard problem is understanding ",[13,297,298],{},"what"," to write — and ",[13,301,302],{},"why",". It's identifying the domain's boundaries, the invariants that must hold, the edge cases the business never told you about because it didn't know it had to, the implications a given architectural choice will have six months from now when the requirements change — ",[63,305,306],{},"because they always change",[10,308,309,310,313],{},"That understanding isn't acquired quickly. It's acquired through iteration, conversation, mistakes, corrections, nights spent wondering why the thing isn't behaving the way it should. It's what Michael Polanyi called ",[13,311,312],{},"tacit knowledge"," — knowing that can't be fully transferred into words, that lives in accumulated experience, in recognized patterns, in intuitions built across dozens of projects.",[10,315,316,319],{},[63,317,318],{},"The AI doesn't have this knowledge."," It has the statistical average of what's been written on the internet about problems similar to yours.",[10,321,322],{},"Which is not the same thing.",[36,324,326],{"id":325},"the-trap-of-apparent-perfection","The trap of apparent perfection",[10,328,329,330,91],{},"There's something specific about AI that makes all of this worse: ",[63,331,332],{},"the code it produces always looks well written",[10,334,335,336,339],{},"It's well named, has comments in the right places, follows conventions, uses the correct patterns. It doesn't look like the code of someone who doesn't know what they're doing — it looks like the code of someone who knows ",[13,337,338],{},"exactly"," what they're doing.",[10,341,342],{},"That's the problem.",[10,344,345,346,349,350,353],{},"Badly written wrong code stops you. You see it, you feel it, it makes you suspicious. It triggers your critical instinct before you've even understood what it does. ",[63,347,348],{},"Well-written wrong code"," sails through code review without tripping a single alarm, passes the tests, ships to production. And when it breaks — ",[13,351,352],{},"because it breaks"," — its correct-looking form makes it harder, not easier, to locate the conceptual problem underneath.",[10,355,356],{},"You're looking for a bug in code that doesn't look buggy. You're looking for a wrong assumption in code that looks reasonable.",[10,358,359,360,363],{},"In 1986 Fred Brooks wrote that the essential difficulty of software isn't accidental — it isn't about syntax, compilers, typing speed. It's about the ",[63,361,362],{},"intrinsic conceptual complexity"," of the systems we build. Forty years later, AI has all but eliminated the accidental difficulty. The essential one is untouched.",[183,365,366],{},[10,367,368,371],{},[13,369,370],{},"\"No silver bullet.\"","\n— Fred Brooks, 1986",[10,373,374],{},"And now that the accidental difficulty is gone, the essential difficulty is more exposed than ever. But it's also easier not to see, because the output hiding it looks so convincing.",[36,376,378],{"id":377},"the-model-of-reality","The model of reality",[10,380,381],{},"When you ask AI to build something, what you get back isn't a solution.",[10,383,384,385,388],{},"It's an ",[63,386,387],{},"interpretation"," of the problem you described.",[10,390,391,392,395,396,399],{},"The AI builds a model of reality based on what you told it — and what you told it is inevitably incomplete, because the domain knowledge in your head doesn't transfer whole into a prompt. The gap between the problem you have ",[13,393,394],{},"in your head"," and the problem the AI ",[13,397,398],{},"understood"," is always there. Sometimes it's negligible. Sometimes it's the difference between a system that holds and one that collapses at the first real edge case.",[10,401,402],{},"Speed doesn't give you time to measure that gap. Worse: speed, combined with the apparent perfection of the output, actively creates the illusion that the gap doesn't exist.",[10,404,405,406,409],{},"This is why ",[63,407,408],{},"\"it works\" is never enough"," as an acceptance criterion.",[10,411,412],{},[13,413,414],{},"Works against what?",[10,416,417],{},"Against the tests the AI wrote to pass — built around the same cases it had in mind when it wrote the code, in a closed loop that includes no external reality? Against a demo with clean data and happy paths? Or against the actual domain logic, with all the edge cases only you know about because you've lived inside that domain?",[10,419,420],{},"Salvatore Sanfilippo — antirez, the creator of Redis — raised an analogous problem about code review: the mental state of the person who wrote the code doesn't transfer to the reviewer. The reviewer sees the text, not the reasoning that produced it. Either they flag superficial things, or they don't understand enough to see the real problem.",[10,422,423,424,427],{},"With AI, the problem radicalizes: ",[63,425,426],{},"the AI has no mental state to transfer",". It produced output optimizing for the internal coherence of the text, not for correspondence with your reality. The reviewer — you — has to close that gap alone. If you don't stop to do it, nobody will.",[36,429,431],{"id":430},"let-it-sink","Let it sink",[10,433,434],{},"There's a practice every engineer with enough scar tissue has internalized: sleep on it.",[10,436,437],{},"Not out of laziness. Out of cognitive necessity.",[10,439,440],{},"The human brain doesn't process complex problems linearly and sequentially. It processes in parallel, often below the threshold of consciousness. That's why the best solutions arrive in the shower, in the car, first thing in the morning — not in front of the screen while you're actively hunting for the answer. The things you don't see in the moment surface later, when you come back with fresh eyes and the brain has had time to connect patterns that active focus can't see.",[10,442,443],{},"It's not mysticism. It's neuroscience. The brain's diffuse mode is essential for insight, for deep understanding, for noticing the inconsistencies that focused mode skips because it's too deep inside the problem.",[10,445,446,447],{},"With AI, that practice becomes even more important, and even easier to skip. Because in a few minutes you have something in front of you that looks complete. The effort you spent is minimal, so the sense of investment is low, so the bar for calling it done drops with it. ",[63,448,449],{},"You didn't struggle to get there, so you don't feel the need to protect yourself from the possibility that the struggle was wasted.",[10,451,452],{},"But \"looks complete\" and \"is correct\" are two different things — and the speed at which AI produces output has made that distinction subtler than ever.",[183,454,455],{},[10,456,457],{},"Stopping isn't losing the advantage of speed. It's the moment you verify that speed took you in the right direction.",[10,459,460],{},"Without that moment, you're just going fast toward the wrong place — and you get there sooner.",[36,462,464],{"id":463},"code-review-isnt-enough","Code review isn't enough",[10,466,467,468],{},"The instinctive response to all of this is: ",[13,469,470],{},"\"that's what code review is for.\"",[10,472,473],{},"But code review, as practiced on most teams, doesn't solve the problem we're describing.",[10,475,476,477,480],{},"Traditional code review is great at catching syntax bugs, convention violations, obvious security issues. It's much less effective at evaluating whether the ",[63,478,479],{},"architectural decisions are correct",", whether the domain model reflects reality, whether the implicit assumptions in the code are valid.",[10,482,483],{},"With AI the problem is amplified: the generated code is syntactically unimpeachable. The reviewer scans, finds nothing that trips the standard alarms, approves. The conceptual problem underneath goes through undetected.",[10,485,486,487,490,491,494],{},"Useful code review in this context isn't ",[13,488,489],{},"\"is this code well written?\""," — that answer is almost always yes. It's ",[13,492,493],{},"\"does this code do what it should do in the context of the real domain?\""," — and to answer that you need the domain knowledge the AI doesn't have, you need to have understood the requirements well enough to know what to check, you need to have let the solution sit long enough to see it with critical distance.",[36,496,498],{"id":497},"the-debt-you-dont-see","The debt you don't see",[10,500,501,502,91],{},"Technical debt generated without judgment has one specific trait: ",[63,503,504],{},"it's silent",[10,506,507],{},"It doesn't show up immediately. It accumulates in decisions the AI made by default, that weren't your decisions — but are now in the code, and the code is in production. In architectures that hold for the use case you described but not the next one. In implicit assumptions the AI made filling the gaps in your prompt with the average of what it saw in training.",[10,509,510,511,514,515,518],{},"Ward Cunningham, who coined the term \"technical debt,\" meant something specific: ",[13,512,513],{},"deliberate"," debt, consciously taken on to move faster now with the intention of paying it back later. Not ",[13,516,517],{},"accidental"," debt — the kind you accumulate without knowing it because you didn't understand the problem well enough.",[10,520,521],{},[63,522,523],{},"Deliberate debt is a legitimate tool. Accidental debt is capitalized ignorance.",[10,525,526],{},"Vibe coding without judgment produces almost exclusively accidental debt. Because speed prevents understanding, and without understanding you can't even consciously choose to go into debt — you just go into debt, without knowing it, without being able to quantify it, without a plan to pay it back.",[10,528,529],{},"And when it surfaces, it surfaces at the worst possible time. When there's traction. When there are real users. When the cost of fixing it is an order of magnitude higher than it would have been at the start.",[183,531,532],{},[10,533,534],{},"The bill always arrives. Speed only determines how fast you run it up.",[36,536,538],{"id":537},"speed-is-a-means","Speed is a means",[10,540,541],{},"AI in software development is extraordinary. I use it every day.",[10,543,544,545,548],{},"But speed is not the goal — it's a means. The goal is to build something that ",[63,546,547],{},"reflects the logic you had in your head",", that holds up under the real domain's edge cases, that someone can understand and change a year from now without cursing whoever wrote it.",[10,550,551],{},"If speed keeps you from verifying that this is true, you're not going faster. You're just failing faster. With more lines of code. With more confidence. With less chance of noticing before it becomes expensive.",[10,553,554,555,558,559,562],{},"The difference between an engineer who uses AI well and one who uses it badly isn't technical. ",[63,556,557],{},"It's cognitive."," It's the ability to stop, let things settle, come back with critical distance, and ask: does this actually reflect what I wanted? Or am I looking at something that ",[13,560,561],{},"looks like"," what I wanted?",[10,564,565],{},"Those are uncomfortable questions to ask when you've just had a productive afternoon and the screen is full of new code and the tests are passing.",[10,567,568],{},[13,569,570],{},"Ask them anyway.",{"title":202,"searchDepth":203,"depth":203,"links":572},[573,574,575,576,577,578,579],{"id":281,"depth":203,"text":282},{"id":325,"depth":203,"text":326},{"id":377,"depth":203,"text":378},{"id":430,"depth":203,"text":431},{"id":463,"depth":203,"text":464},{"id":497,"depth":203,"text":498},{"id":537,"depth":203,"text":538},[212,581],"ai","https://i2.wp.com/enricodeleo.s3.eu-south-1.amazonaws.com/images/velocita-senza-giudizio.png","2026-07-12T12:00:00.000Z","Vibe coding didn't invent badly written software. It just removed the friction. And speed, without the time to evaluate, only produces faster mistakes.",{},"/en/speed-without-judgment",{"title":235,"description":584},"en/speed-without-judgment",[581,590,227,229],"vibe-coding","1EiIosWChyFyr3gLffNLNa3AnQzQaH3u9ym3ejZxupc",{"id":593,"title":594,"body":595,"categories":661,"coverImage":662,"date":663,"description":664,"extension":216,"lang":217,"legacy":218,"meta":665,"navigation":220,"path":666,"seo":667,"stem":668,"sticky":220,"tags":669,"translation":603,"updated":230,"__hash__":673},"articles/en/a-human-if-needed.md","A human, if needed",{"type":7,"value":596,"toc":656},[597,605,608,611,614,617,621,624,627,633,637,640,643,647,650,653],[10,598,599],{},[13,600,20,601,26],{},[22,602,604],{"href":603},"/un-umano-se-serve","Un umano, se serve",[10,606,607],{},"Every time a new model ships, the feed fills up with reviews. \"I tried it. Here are my impressions.\" Screenshots, homegrown benchmarks, verdicts.",[10,609,610],{},"I don't understand what feels special about chatting with the same model a hundred million other people have access to. It's like reviewing electricity.",[10,612,613],{},"To be clear: I'm not saying models don't matter. They matter to the point that any frontier LLM is, frankly, already magic. I'm saying the opposite: it's precisely because they're magic available to everyone, at the same price, on the same day, that they've stopped being an interesting topic. When everyone has the same capability, capability stops being the variable.",[10,615,616],{},"The variable is what you build around it.",[36,618,620],{"id":619},"the-question-i-care-about","The question I care about",[10,622,623],{},"The question I care about isn't which model is best this month. It's what this technology does to organizations. And there I see something specific, something I think is still underrated.",[10,625,626],{},"Today companies \"adopt AI\": they take the existing org chart — people, roles, hierarchies — and bolt tools onto people. A copilot for the developer, an assistant for marketing, a chatbot for support. The human stays the default executor; the AI assists.",[10,628,629,630],{},"The AI-native company of tomorrow will do the exact opposite: ",[63,631,632],{},"an agent for every role, and a human if needed.",[36,634,636],{"id":635},"the-inverted-org-chart","The inverted org chart",[10,638,639],{},"This isn't a provocation. It's the logical consequence of where we already are. If an agent covers the first pass at the code, the first draft of the report, the first response to the customer, the role is no longer a person with a tool: it's an automated process with an escalation point. The org chart inverts. The human stops being the workforce and becomes the layer of judgment — the one you escalate to when judgment actually changes the risk.",[10,641,642],{},"That changes very concrete questions. \"Where do we need hands\" becomes \"where do we need judgment.\" And judgment isn't hired, it's accumulated. The people who move up are the ones who have already seen the failure modes, who can tell a plausible output from a correct one, who know when to stop the machine. Thresholds, review, escalation stop being decorative compliance and become the actual design of the organization.",[36,644,646],{"id":645},"the-flip-side","The flip side",[10,648,649],{},"There's a flip side too, and it needs saying: an agent per role without a real escalation path isn't an organization, it's automation hiding complexity. And the human \"if needed\" only works if, when needed, they're actually there — with the context, the authority, and the time to intervene. Otherwise you've built an extremely fast factory with an alarm bell nobody hears.",[10,651,652],{},"The agent executes. The human responds — in both senses: they respond when needed, and they answer for the consequences.",[10,654,655],{},"Everything else — which model, which benchmark, which review — is noise.",{"title":202,"searchDepth":203,"depth":203,"links":657},[658,659,660],{"id":619,"depth":203,"text":620},{"id":635,"depth":203,"text":636},{"id":645,"depth":203,"text":646},[581],"https://i2.wp.com/enricodeleo.s3.eu-south-1.amazonaws.com/images/umano-se-serve.png","2026-07-12T11:00:00.000Z","This month's model is a commodity. The real question is what AI does to organizations: an agent for every role, a human if needed.",{},"/en/a-human-if-needed",{"title":594,"description":664},"en/a-human-if-needed",[581,670,671,672,227],"agents","organization","ai-native","PhZuSu3jlmuZ6PtFT1vI-fLdztj_T-hgZV3LarO_VCg",1783814584652]