[{"data":1,"prerenderedAt":955},["ShallowReactive",2],{"post-en-a-port-is-not-a-translation":3,"related-en-a-port-is-not-a-translation":514},{"id":4,"title":5,"body":6,"categories":493,"coverImage":496,"date":497,"description":498,"extension":499,"lang":500,"legacy":501,"meta":502,"navigation":503,"path":504,"seo":505,"stem":506,"sticky":501,"tags":507,"translation":24,"updated":512,"__hash__":513},"articles/en/a-port-is-not-a-translation.md","A port is not a translation",{"type":7,"value":8,"toc":480},"minimark",[9,16,27,40,48,51,56,59,64,71,84,87,91,94,97,104,107,110,121,124,128,131,139,146,153,156,159,175,182,186,189,249,262,276,279,282,286,289,331,334,337,341,344,350,357,363,366,377,382,386,393,396,411,414,417,421,424,427,430,436,439,442,446,475],[10,11,12],"p",{},[13,14,15],"em",{},"A port is not a translation. It's an exercise in discovering what, in a system, is actually the contract. It's almost never what you think.",[10,17,18],{},[13,19,20,21,26],{},"(This is an English adaptation of my Italian essay ",[22,23,25],"a",{"href":24},"/bullmq-rs-porting","L'interfaccia che non si vede",".)",[10,28,29,30,39],{},"A while ago a PR I opened against ",[22,31,35],{"href":32,"rel":33},"https://github.com/bogardt/bullmq-rs",[34],"nofollow",[36,37,38],"code",{},"bullmq-rs"," — a Rust implementation of BullMQ — got merged upstream. The diff is large: roughly eleven thousand lines added, two thousand removed, shipped as a breaking v2.0.0 release.",[10,41,42,43,47],{},"Rereading the PR description after the fact, I realized the most informative section wasn't the list of what the PR included. It was the list of ",[44,45,46],"strong",{},"what I had deliberately left out",".",[10,49,50],{},"That's the only part worth writing about.",[52,53,55],"h2",{"id":54},"what-happened","What happened",[10,57,58],{},"BullMQ is a Node library for distributed queues on top of Redis. It's widely used, it has a mature ecosystem around it — Bull Board, dashboards, monitoring — and it has a de facto specification for how data is structured inside Redis. There is no formal standard: there's the library's code, and there's the behavior that external tooling expects to observe.",[10,60,61,63],{},[36,62,38],{}," is the Rust implementation. The idea is simple: if you want workers in Rust, you should be able to write workers in Rust without rebuilding the infrastructure around them. The dashboard stays. Jobs created from Node are readable from Rust, and vice versa. Metrics keep working. You migrate one service at a time without turning anything else off.",[10,65,66,67,70],{},"In the previous version of the library, the Rust API had converged toward BullMQ's — but the ",[44,68,69],{},"wire format",", the exact shape of the data landing in Redis, diverged in subtle ways. Lists where there should have been sorted sets. Hash fields with different names. Stream events whose payloads didn't match what Node tooling expects to read.",[10,72,73,74,77,78,83],{},"The concrete result: external tooling didn't work. Bull Board couldn't inspect queues created from Rust — Redis answered its list operations with ",[36,75,76],{},"WRONGTYPE",". A Node worker sharing a queue with a Rust worker couldn't see the other's jobs. (That was ",[22,79,82],{"href":80,"rel":81},"https://github.com/bogardt/bullmq-rs/issues/2",[34],"the issue"," that started all of this.)",[10,85,86],{},"The Rust API was clean. What came out of Redis was something else.",[52,88,90],{"id":89},"the-scenario-this-is-for","The scenario this is for",[10,92,93],{},"To be concrete, the typical scenario looks like this.",[10,95,96],{},"You run a Node system with BullMQ in production. It works. Bull Board shows you the queues, the alerts are wired, the SRE knows where to look when something smokes at night.",[10,98,99,100,103],{},"At some point one job type stops keeping up. Video encoding. Image processing. A CPU-bound pipeline. The Node worker does what it can, but the throughput isn't there anymore. You'd like to rewrite ",[13,101,102],{},"just that worker"," in Rust and leave the rest of the infrastructure where it is.",[10,105,106],{},"Without wire compatibility you have two options. Rewrite everything in Rust and lose the Node tooling. Or maintain a translation layer between the two, and carry the complexity of keeping it aligned with every BullMQ release.",[10,108,109],{},"With wire compatibility there's a third option. The same job produced by a Node producer gets processed by a Rust worker. The dashboard doesn't know two runtimes are calling it. The metrics don't change shape. The SRE paged at 3 AM sees exactly what they saw before.",[10,111,112,113,116,117,120],{},"There's also a property that emerges naturally in Rust and simply doesn't exist in Node: the job payload is typed. You write ",[36,114,115],{},"Job\u003CEncodingTask>"," and the worker receives an ",[36,118,119],{},"EncodingTask"," already deserialized and checked at compile time. Change the schema and the compiler stops you before the deploy. For jobs where payload correctness actually matters — payments, expensive side effects, critical state transitions — that's a concrete guarantee, not an aesthetic detail.",[10,122,123],{},"It's not a keynote use case. It's the kind that matters after the deploy.",[52,125,127],{"id":126},"hyrums-law-applied-to-porting","Hyrum's Law, applied to porting",[10,129,130],{},"There's a formulation attributed to Hyrum Wright, who spent years at Google working on large-scale refactoring tooling:",[132,133,134],"blockquote",{},[10,135,136],{},[13,137,138],{},"With a sufficient number of users of an API, it does not matter what you promise in the contract: all observable behaviors of your system will be depended on by somebody.",[10,140,141,142,145],{},"With enough users, ",[44,143,144],{},"what you claim to promise doesn't matter",". What matters is everything the system does, including the parts you never would have chosen to promise. Somebody, somewhere, is leaning on every one of them.",[10,147,148,149,152],{},"In a port, this has a very precise corollary: the real interface of the system you're porting ",[44,150,151],{},"is not its API",". It's the set of externally observable behaviors — data structures, serialization formats, side effects, event ordering — that someone outside the system has built on to make their own things work.",[10,154,155],{},"For BullMQ, that interface is the Redis layout. Not a function signature. The exact shape of a hash, of a sorted set, of a stream entry.",[10,157,158],{},"When you port a system like this, every choice falls into one of two categories:",[160,161,162,169],"ol",{},[163,164,165,168],"li",{},[44,166,167],{},"What you cannot change"," — the wire format. Changing it breaks everything sitting on top without the caller noticing, until the next deploy when they discover the dashboard is empty or half the jobs are missing.",[163,170,171,174],{},[44,172,173],{},"What you can change"," — the API in the target language. That one is free. Nobody outside is observing it.",[10,176,177,178,181],{},"Confusing the two categories is the most reliable way to produce a port that ",[13,179,180],{},"looks finished"," and isn't.",[52,183,185],{"id":184},"what-the-pr-put-in","What the PR put in",[10,187,188],{},"The first part of the work was flattening the wire format under BullMQ v5's:",[190,191,192,205,226,229],"ul",{},[163,193,194,195,198,199,198,202],{},"lists for ",[36,196,197],{},"wait",", ",[36,200,201],{},"active",[36,203,204],{},"paused",[163,206,207,208,198,211,198,214,198,217,198,220,198,223],{},"sorted sets for ",[36,209,210],{},"prioritized",[36,212,213],{},"delayed",[36,215,216],{},"completed",[36,218,219],{},"failed",[36,221,222],{},"waiting-children",[36,224,225],{},"marker",[163,227,228],{},"a hash for metadata, a stream for events",[163,230,231,232,198,235,198,238,198,241,244,245,248],{},"job hash fields with the same names and formats as Node (",[36,233,234],{},"atm",[36,236,237],{},"ats",[36,239,240],{},"processedOn",[36,242,243],{},"pb",", JSON-encoded ",[36,246,247],{},"opts",")",[10,250,251,252,254,255,198,257,254,259,261],{},"On top of that, ports of the atomic Lua scripts BullMQ uses for state transitions. A job transition — ",[36,253,197],{}," to ",[36,256,201],{},[36,258,201],{},[36,260,216],{},", recovering a stalled job — cannot be implemented as a sequence of Redis commands. It has to be atomic, and BullMQ gets atomicity by delegating to Lua. Rewriting those by hand in Rust would mean inventing race conditions the original code doesn't have. So you port them line by line, and verify the observed behavior is identical.",[10,263,264,265,268,269,272,273,275],{},"Above that, the runtime: a marker-based worker loop on ",[36,266,267],{},"BZPOPMIN",", token-based locks with TTLs, stalled-job recovery, a ",[36,270,271],{},"moveToFinished"," fast path, and a prefetch fix so jobs don't end up orphaned in ",[36,274,201],{}," when a worker dies uncleanly.",[10,277,278],{},"Above that, the Rust API surface: Queue, Job, QueueEvents, FlowProducer with cross-queue parent/child trees.",[10,280,281],{},"That's enough to use in production, and enough to talk to an existing Node cluster with no translation layer in between.",[52,283,285],{"id":284},"what-the-pr-deliberately-left-out","What the PR deliberately left out",[10,287,288],{},"A list, in the body of the PR:",[190,290,291,298,316,319,322,325,328],{},[163,292,293,294,297],{},"no ",[36,295,296],{},"JobScheduler"," (cron-style repeatable jobs)",[163,299,300,301,198,304,198,307,198,310,198,313,248],{},"no bulk queue operations (",[36,302,303],{},"addBulk",[36,305,306],{},"clean",[36,308,309],{},"obliterate",[36,311,312],{},"retryJobs",[36,314,315],{},"promoteJobs",[163,317,318],{},"no worker-level pause/resume (queue-level: yes)",[163,320,321],{},"no rate limiting",[163,323,324],{},"no deduplication",[163,326,327],{},"no debouncing",[163,329,330],{},"no metrics",[10,332,333],{},"All things BullMQ has. All things that would need to be added.",[10,335,336],{},"But not in this PR.",[52,338,340],{"id":339},"why","Why",[10,342,343],{},"Because a port that tries to do everything in one shot ends one of two ways: it never gets merged, or it gets merged with hidden compromises.",[10,345,346,347],{},"Hidden compromises are the ones you don't see in the description but that live in the code. Conventions invented where the spec was unclear. Edge cases left behind out of fatigue. Behaviors that silently diverge from the original system and surface months later, when a user reports that ",[13,348,349],{},"\"it works with Node, not with Rust.\"",[10,351,352,353,356],{},"That's Cunningham's ",[13,354,355],{},"accidental"," debt — the kind you accumulate without knowing, because you didn't understand the problem well enough. You can't quantify it. You can't pay it down. You don't even know where it is.",[10,358,359,360,47],{},"The way to avoid it, in a port, is to invert the instinct: ",[44,361,362],{},"declare first what you are not doing",[10,364,365],{},"Not at the bottom of the PR, in small print. In the description, in a dedicated section, with the precise names of the missing APIs and behaviors. Publicly, before anyone asks.",[10,367,368,369,372,373,376],{},"That transforms the exact same situation — ",[13,370,371],{},"\"a library with missing features\""," — from accidental debt into ",[44,374,375],{},"deliberate debt",": the legitimate tool Cunningham was actually talking about. I know what's missing. It's written down. It's tracked. The next PR knows where to land. Users of the library know what to expect and what not to.",[132,378,379],{},[10,380,381],{},"Deliberate debt is a scoping tool. Accidental debt is scope that manages you.",[52,383,385],{"id":384},"the-temptation-to-port-everything","The temptation to port everything",[10,387,388,389,392],{},"There's a specific temptation in any port or reimplementation — and it's the same one that AI-assisted development speed makes easier to indulge: trying to do ",[13,390,391],{},"everything"," in one shot, because technically you could.",[10,394,395],{},"The original system has a hundred features. The port can imitate all of them. The LOC scale. The PR balloons. Review becomes impossible. The decisions about the parts you never really understood end up being made by default — by generated code, by analogy with other systems, by whatever seemed reasonable at three in the morning.",[10,397,398,399,402,403,406,407,410],{},"And when that PR gets merged, the resulting system has the surface area of the original and a bug surface nobody has mapped. It ",[13,400,401],{},"looks"," complete. And as I've written elsewhere, ",[13,404,405],{},"looks complete"," and ",[13,408,409],{},"is correct"," have never been the same thing.",[10,412,413],{},"The alternative is less glamorous. Port what's structural first — the wire format, the runtime, the atoms — then add surface one piece at a time, additively, writing down in plain sight what's still missing and why.",[10,415,416],{},"Nothing is proven by porting more. Something is almost always lost by porting without declaring what's in and what's out.",[52,418,420],{"id":419},"the-interface-you-dont-see","The interface you don't see",[10,422,423],{},"This holds outside of ports, too.",[10,425,426],{},"In almost every system that shares state with something external — another application, another team, another runtime, another language — the interface that actually matters is not the one in a function signature. It's the one that crosses the boundary. A JSON payload. A row in a table. An event on a stream.",[10,428,429],{},"It's the one the caller observes. It's the one that, changed silently, breaks things far away from where you changed them, at an hour when you're not at the screen.",[10,431,432,433],{},"Inside a closed system you have the luxury of renaming, restructuring, refactoring at will. When the system has a boundary someone else observes, that luxury is gone. ",[44,434,435],{},"The language API is ephemeral. The real contract is whatever crosses the boundary.",[10,437,438],{},"For the BullMQ port, that boundary was Redis. Once the boundary is right, everything else is additive. If the boundary is wrong, every feature you add on top is one more lie to maintain.",[10,440,441],{},"And lies in systems don't expire. They wait.",[52,443,445],{"id":444},"references","References",[190,447,448,456,463],{},[163,449,450,451],{},"Merged PR: ",[22,452,455],{"href":453,"rel":454},"https://github.com/bogardt/bullmq-rs/pull/3",[34],"bogardt/bullmq-rs#3",[163,457,458,459],{},"The issue that started it: ",[22,460,462],{"href":80,"rel":461},[34],"bogardt/bullmq-rs#2",[163,464,465,466,471,472],{},"The library, by ",[22,467,470],{"href":468,"rel":469},"https://github.com/bogardt",[34],"@bogardt",": ",[22,473,38],{"href":32,"rel":474},[34],[10,476,477],{},[13,478,479],{},"The most honest port is the one that declares first what it is not porting.",{"title":481,"searchDepth":482,"depth":482,"links":483},"",2,[484,485,486,487,488,489,490,491,492],{"id":54,"depth":482,"text":55},{"id":89,"depth":482,"text":90},{"id":126,"depth":482,"text":127},{"id":184,"depth":482,"text":185},{"id":284,"depth":482,"text":285},{"id":339,"depth":482,"text":340},{"id":384,"depth":482,"text":385},{"id":419,"depth":482,"text":420},{"id":444,"depth":482,"text":445},[494,495],"dev","open-source","https://i2.wp.com/enricodeleo.s3.eu-south-1.amazonaws.com/images/linterfaccia-che-non-si-vede.png","2026-07-12T09:00:00.000Z","Porting BullMQ to Rust taught me that the real interface of a system is never its API. It's whatever crosses the boundary — and someone is always watching it.","md","en",false,{},true,"/en/a-port-is-not-a-translation",{"title":5,"description":498},"en/a-port-is-not-a-translation",[508,509,495,510,511],"rust","bullmq","porting","engineering",null,"isiv4UFXbBfciGmA76xkN66TKUoyVub3mL69ouwye9c",[515,873],{"id":516,"title":517,"body":518,"categories":860,"coverImage":862,"date":863,"description":864,"extension":499,"lang":500,"legacy":501,"meta":865,"navigation":503,"path":866,"seo":867,"stem":868,"sticky":501,"tags":869,"translation":531,"updated":512,"__hash__":872},"articles/en/speed-without-judgment.md","Speed without judgment",{"type":7,"value":519,"toc":851},[520,525,533,536,543,546,549,554,561,565,568,575,588,595,601,604,608,614,621,624,635,638,645,653,656,660,663,670,681,684,691,696,699,702,709,713,716,719,722,725,731,734,739,742,746,752,755,762,765,776,780,786,789,799,804,807,810,815,819,822,829,832,843,846],[10,521,522],{},[13,523,524],{},"Vibe coding didn't invent badly written software. It just removed the friction.",[10,526,527],{},[13,528,20,529,26],{},[22,530,532],{"href":531},"/velocita-senza-giudizio","Velocità senza giudizio",[10,534,535],{},"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.",[10,537,538,539,542],{},"I mention it because what's happening now ",[44,540,541],{},"isn't new",". It's the same thing, faster.",[10,544,545],{},"I've seen Google Sheets used as a database. It collapsed at four concurrent users. The founders were surprised.",[10,547,548],{},"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.",[132,550,551],{},[10,552,553],{},"A system that certifies itself is more dangerous than one that fails openly.",[10,555,556,557,560],{},"The problem isn't the AI. It's that the speed it offers can make you forget that building software is ",[44,558,559],{},"fundamentally a cognitive act"," — not a writing one.",[52,562,564],{"id":563},"software-is-not-text","Software is not text",[10,566,567],{},"This is the thing the enthusiasm around vibe coding systematically forgets.",[10,569,570,571,574],{},"Software is not text. It's a ",[44,572,573],{},"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.",[10,576,577,578,581,582,584,585,47],{},"Writing code was never the hard problem. The hard problem is understanding ",[13,579,580],{},"what"," to write — and ",[13,583,339],{},". 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 — ",[44,586,587],{},"because they always change",[10,589,590,591,594],{},"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 ",[13,592,593],{},"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.",[10,596,597,600],{},[44,598,599],{},"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.",[10,602,603],{},"Which is not the same thing.",[52,605,607],{"id":606},"the-trap-of-apparent-perfection","The trap of apparent perfection",[10,609,610,611,47],{},"There's something specific about AI that makes all of this worse: ",[44,612,613],{},"the code it produces always looks well written",[10,615,616,617,620],{},"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 ",[13,618,619],{},"exactly"," what they're doing.",[10,622,623],{},"That's the problem.",[10,625,626,627,630,631,634],{},"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. ",[44,628,629],{},"Well-written wrong code"," sails through code review without tripping a single alarm, passes the tests, ships to production. And when it breaks — ",[13,632,633],{},"because it breaks"," — its correct-looking form makes it harder, not easier, to locate the conceptual problem underneath.",[10,636,637],{},"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.",[10,639,640,641,644],{},"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 ",[44,642,643],{},"intrinsic conceptual complexity"," of the systems we build. Forty years later, AI has all but eliminated the accidental difficulty. The essential one is untouched.",[132,646,647],{},[10,648,649,652],{},[13,650,651],{},"\"No silver bullet.\"","\n— Fred Brooks, 1986",[10,654,655],{},"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.",[52,657,659],{"id":658},"the-model-of-reality","The model of reality",[10,661,662],{},"When you ask AI to build something, what you get back isn't a solution.",[10,664,665,666,669],{},"It's an ",[44,667,668],{},"interpretation"," of the problem you described.",[10,671,672,673,676,677,680],{},"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 ",[13,674,675],{},"in your head"," and the problem the AI ",[13,678,679],{},"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.",[10,682,683],{},"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.",[10,685,686,687,690],{},"This is why ",[44,688,689],{},"\"it works\" is never enough"," as an acceptance criterion.",[10,692,693],{},[13,694,695],{},"Works against what?",[10,697,698],{},"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?",[10,700,701],{},"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.",[10,703,704,705,708],{},"With AI, the problem radicalizes: ",[44,706,707],{},"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.",[52,710,712],{"id":711},"let-it-sink","Let it sink",[10,714,715],{},"There's a practice every engineer with enough scar tissue has internalized: sleep on it.",[10,717,718],{},"Not out of laziness. Out of cognitive necessity.",[10,720,721],{},"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.",[10,723,724],{},"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.",[10,726,727,728],{},"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. ",[44,729,730],{},"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.",[10,732,733],{},"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.",[132,735,736],{},[10,737,738],{},"Stopping isn't losing the advantage of speed. It's the moment you verify that speed took you in the right direction.",[10,740,741],{},"Without that moment, you're just going fast toward the wrong place — and you get there sooner.",[52,743,745],{"id":744},"code-review-isnt-enough","Code review isn't enough",[10,747,748,749],{},"The instinctive response to all of this is: ",[13,750,751],{},"\"that's what code review is for.\"",[10,753,754],{},"But code review, as practiced on most teams, doesn't solve the problem we're describing.",[10,756,757,758,761],{},"Traditional code review is great at catching syntax bugs, convention violations, obvious security issues. It's much less effective at evaluating whether the ",[44,759,760],{},"architectural decisions are correct",", whether the domain model reflects reality, whether the implicit assumptions in the code are valid.",[10,763,764],{},"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.",[10,766,767,768,771,772,775],{},"Useful code review in this context isn't ",[13,769,770],{},"\"is this code well written?\""," — that answer is almost always yes. It's ",[13,773,774],{},"\"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.",[52,777,779],{"id":778},"the-debt-you-dont-see","The debt you don't see",[10,781,782,783,47],{},"Technical debt generated without judgment has one specific trait: ",[44,784,785],{},"it's silent",[10,787,788],{},"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.",[10,790,791,792,795,796,798],{},"Ward Cunningham, who coined the term \"technical debt,\" meant something specific: ",[13,793,794],{},"deliberate"," debt, consciously taken on to move faster now with the intention of paying it back later. Not ",[13,797,355],{}," debt — the kind you accumulate without knowing it because you didn't understand the problem well enough.",[10,800,801],{},[44,802,803],{},"Deliberate debt is a legitimate tool. Accidental debt is capitalized ignorance.",[10,805,806],{},"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.",[10,808,809],{},"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.",[132,811,812],{},[10,813,814],{},"The bill always arrives. Speed only determines how fast you run it up.",[52,816,818],{"id":817},"speed-is-a-means","Speed is a means",[10,820,821],{},"AI in software development is extraordinary. I use it every day.",[10,823,824,825,828],{},"But speed is not the goal — it's a means. The goal is to build something that ",[44,826,827],{},"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.",[10,830,831],{},"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.",[10,833,834,835,838,839,842],{},"The difference between an engineer who uses AI well and one who uses it badly isn't technical. ",[44,836,837],{},"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 ",[13,840,841],{},"looks like"," what I wanted?",[10,844,845],{},"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.",[10,847,848],{},[13,849,850],{},"Ask them anyway.",{"title":481,"searchDepth":482,"depth":482,"links":852},[853,854,855,856,857,858,859],{"id":563,"depth":482,"text":564},{"id":606,"depth":482,"text":607},{"id":658,"depth":482,"text":659},{"id":711,"depth":482,"text":712},{"id":744,"depth":482,"text":745},{"id":778,"depth":482,"text":779},{"id":817,"depth":482,"text":818},[494,861],"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":517,"description":864},"en/speed-without-judgment",[861,870,511,871],"vibe-coding","build-in-public","1EiIosWChyFyr3gLffNLNa3AnQzQaH3u9ym3ejZxupc",{"id":874,"title":875,"body":876,"categories":942,"coverImage":943,"date":944,"description":945,"extension":499,"lang":500,"legacy":501,"meta":946,"navigation":503,"path":947,"seo":948,"stem":949,"sticky":503,"tags":950,"translation":884,"updated":512,"__hash__":954},"articles/en/a-human-if-needed.md","A human, if needed",{"type":7,"value":877,"toc":937},[878,886,889,892,895,898,902,905,908,914,918,921,924,928,931,934],[10,879,880],{},[13,881,20,882,26],{},[22,883,885],{"href":884},"/un-umano-se-serve","Un umano, se serve",[10,887,888],{},"Every time a new model ships, the feed fills up with reviews. \"I tried it. Here are my impressions.\" Screenshots, homegrown benchmarks, verdicts.",[10,890,891],{},"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.",[10,893,894],{},"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.",[10,896,897],{},"The variable is what you build around it.",[52,899,901],{"id":900},"the-question-i-care-about","The question I care about",[10,903,904],{},"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.",[10,906,907],{},"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.",[10,909,910,911],{},"The AI-native company of tomorrow will do the exact opposite: ",[44,912,913],{},"an agent for every role, and a human if needed.",[52,915,917],{"id":916},"the-inverted-org-chart","The inverted org chart",[10,919,920],{},"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.",[10,922,923],{},"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.",[52,925,927],{"id":926},"the-flip-side","The flip side",[10,929,930],{},"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.",[10,932,933],{},"The agent executes. The human responds — in both senses: they respond when needed, and they answer for the consequences.",[10,935,936],{},"Everything else — which model, which benchmark, which review — is noise.",{"title":481,"searchDepth":482,"depth":482,"links":938},[939,940,941],{"id":900,"depth":482,"text":901},{"id":916,"depth":482,"text":917},{"id":926,"depth":482,"text":927},[861],"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":875,"description":945},"en/a-human-if-needed",[861,951,952,953,511],"agents","organization","ai-native","PhZuSu3jlmuZ6PtFT1vI-fLdztj_T-hgZV3LarO_VCg",1783814584652]