[{"content":"You\u0026rsquo;re mid-conversation with an AI assistant. It\u0026rsquo;s been helpful. Then, somewhere around message thirty, it starts contradicting itself. It forgets what you told it ten minutes ago. It confidently suggests the thing you already ruled out. You know what happened\u0026hellip; you\u0026rsquo;ve hit the context window.\nFrustrating, right? But here\u0026rsquo;s the thing. We treat this like it\u0026rsquo;s a uniquely AI problem, like \u0026ldquo;losing context\u0026rdquo; is a technical limitation we\u0026rsquo;ll eventually engineer away.\nI think it\u0026rsquo;s not. Companies do the same thing every day. They ship confidently into markets they haven\u0026rsquo;t actually looked at. Same failure mode: producing output without understanding what you\u0026rsquo;re responding to. We just didn\u0026rsquo;t call it \u0026ldquo;losing context\u0026rdquo;. We called it (and still do) \u0026ldquo;shipping fast\u0026rdquo;.\nYou thought it was a post about AI and context? Sorry, it\u0026rsquo;s not.\nShip It \u0026ldquo;We need to ship it, fast\u0026rdquo;. You\u0026rsquo;ve heard it. Probably said it. The logic is seductive and the examples are always the same: the first laptop shipped fast, the first smartphone shipped fast, the first search engine shipped fast. Be first, iterate later. Move fast, break things, fix things, do stuff.\nIt makes sense, right? The historical record seems to back it up. Speed wins.\nExcept people conveniently forget why those products succeeded. It wasn\u0026rsquo;t just that they shipped fast. It\u0026rsquo;s that they shipped into something specific, in a specific context. Let me explain.\nThe Context You know what? There was no laptop before the laptop. There was no smartphone before the smartphone, not really. AltaVista existed before Google, sure, but Google shipped something meaningfully different into a space where \u0026ldquo;meaningfully different\u0026rdquo; still had room to breathe.\nWhat I\u0026rsquo;m trying to say is that \u0026ldquo;ship fast\u0026rdquo; works when you\u0026rsquo;re filling a gap that genuinely exists. The gap doesn\u0026rsquo;t have to be enormous. It doesn\u0026rsquo;t have to be a blank market. But it has to be real.\nYour new todo app doesn\u0026rsquo;t have a gap to fill, there are thousands of them. Your AI project does no more have a gap to fill. Your note taking app doesn\u0026rsquo;t have a gap to fill, like at all.\nShipping any of these is the equivalent of suggesting the thing you already ruled out. Same confidence. Same lack of context. Speed doesn\u0026rsquo;t help when there\u0026rsquo;s nowhere to land. You can ship in two weeks or two days, and it won\u0026rsquo;t matter, because you\u0026rsquo;re not filling anything. You\u0026rsquo;re just adding noise to a crowded room.\nThis is what people (on purpose?) miss when they repeat the \u0026ldquo;ship fast\u0026rdquo; mantra over and over again. They\u0026rsquo;re quoting the outcome without understanding the conditions that made it possible. It\u0026rsquo;s like watching someone win the lottery and concluding that buying tickets is a solid retirement strategy. Would you think that?\nWait, That\u0026rsquo;s Not True I already see you yelling at the screen: \u0026ldquo;But, but, X launched when Y already existed!?!\u0026rdquo;\nI mean, that\u0026rsquo;s fair. That\u0026rsquo;s a real objection, right?\nLet\u0026rsquo;s take Instagram as an example. It was a two-person startup in 2010, Facebook acquired it in 2012. Yeah, and social media was already there. Right. But, we are not considering that Instagram did fill a gap. Just not an obvious one.\nFlickr was desktop-first. Hipstamatic had filters but no social network. There was no mobile-first, simple, social photo app that made anyone\u0026rsquo;s photos look professional. Isn\u0026rsquo;t that still a gap? Just not a \u0026ldquo;nothing exists yet\u0026rdquo; one. Probably, your objection will perfectly fit in the Instagram context. And that\u0026rsquo;s fine.\nWhat I\u0026rsquo;m trying to tell you is that the gap doesn\u0026rsquo;t have to mean \u0026ldquo;nothing exists at all\u0026rdquo;, It can mean \u0026ldquo;nothing does this specific thing, in this specific way, for this specific audience\u0026rdquo;. See the context change? That\u0026rsquo;s a nice gap, don\u0026rsquo;t you agree?\nLook, my point is not to \u0026ldquo;never ship fast\u0026rdquo;. I mean, yeah ship fast if you can. But my point is, please understand what you\u0026rsquo;re shipping into. Please, understand the context. Speed is a multiplier. Multiply zero by anything and you still get zero.\nThe AI Graveyard Here\u0026rsquo;s where it gets worse. AI has commoditized speed itself.\nA solo developer with a good prompt and a weekend can now ship something that would have taken a team three months in 2019. That\u0026rsquo;s genuinely remarkable. But it also means anyone can do it. Speed used to be a differentiator. Now it\u0026rsquo;s expected.\nAnd when speed stops being scarce, what\u0026rsquo;s left? You got it: the gap. Or better, the context. The actual understanding of who needs what and why nothing else is giving it to them. The one thing AI can\u0026rsquo;t commoditize (for now, at least) is knowing what you\u0026rsquo;re responding to.\nIn the last weeks, my LinkedIn feed is exploding with \u0026ldquo;be fast\u0026rdquo; and \u0026ldquo;ship it\u0026rdquo; promotional posts, everyone understood that the barrier to shipping dropped, so more people are shipping products fast, and most of them are shipping without asking the hard questions first. Just because AI makes it so easy. And what if your idea fails? Who cares, just spend some more tokens on a new one.\nThe result? Well, the result is a graveyard. You can see it for yourself, aigraveyard.org and other similar (probably AI-built) sites track dead AI products, and $14.2 billion worth of them would probably agree that moving faster in the wrong direction means just failing faster.\nA Fluenta report found that 9 out of 12 SaaS failures happened in markets that were already full. Not bad products. Not bad teams. Just no room. Speed without context isn\u0026rsquo;t a strategy. It\u0026rsquo;s just a more efficient way to be wrong.\nJust, Ship It Next time someone tells you to just ship it, ask them: ship it into what, exactly?\nContext mattered before AI. It matters even more now. The only thing that changed is how fast we can ignore it. And ignoring context faster doesn\u0026rsquo;t make you move fast. It makes you the AI at message thirty, confidently suggesting the thing everyone already ruled out.\n","permalink":"https://simonetavoletta.it/blog/context-matters/","summary":"\u003cp\u003eYou\u0026rsquo;re mid-conversation with an AI assistant. It\u0026rsquo;s been helpful. Then, somewhere around message thirty, it starts contradicting itself. It forgets what you told it ten minutes ago. It confidently suggests the thing you already ruled out. You know what happened\u0026hellip; you\u0026rsquo;ve hit the context window.\u003c/p\u003e\n\u003cp\u003eFrustrating, right? But here\u0026rsquo;s the thing. We treat this like it\u0026rsquo;s a uniquely AI problem, like \u0026ldquo;losing context\u0026rdquo; is a technical limitation we\u0026rsquo;ll eventually engineer away.\u003c/p\u003e","title":"Context Matters"},{"content":"There\u0026rsquo;s a song from Bo Burnham that I really like. It\u0026rsquo;s called Art is Dead and it starts like this:\n\u0026ldquo;Engineers like to seem complicated\nBut we\u0026rsquo;re not complicated\nI can explain it pretty easily\u0026hellip;\u0026rdquo;\nSorry, it was \u0026ldquo;Entertainers\u0026rdquo;, not \u0026ldquo;Engineers\u0026rdquo;, my bad.\nAnyway, in the last few weeks I\u0026rsquo;ve read a lot of posts and watched lots of videos from developers and software engineers all over the world, famous and not, being frustrated over AI. Frustrated, scared, dismissive. The whole spectrum.\nAs always in life, I\u0026rsquo;ve started to notice that there are mostly two sides, we could identify them with 0 and 1. A small amount of people seems to be in the middle (0.5, or \u0026ldquo;on the fence\u0026rdquo;, as Tim Minchin would sing). But most of them, they just pick one side.\n001101 Think of 0 people as the ones (pun intended) that are in the first/second stage of grief:\n\u0026ldquo;AI is useless.\u0026rdquo;\n\u0026ldquo;Opus 4.7 is stupid.\u0026rdquo;\n\u0026ldquo;I tested it and it couldn\u0026rsquo;t even do X.\u0026rdquo;\nThese are the developers who need AI to fail, because if it doesn\u0026rsquo;t, something uncomfortable becomes true about their craft.\nOn the other hand, 1 people are in the third (or fourth) stage.\n\u0026ldquo;AI will steal all our jobs.\u0026rdquo;\n\u0026ldquo;We\u0026rsquo;re all doomed.\u0026rdquo; \u0026ldquo;The robots are coming.\u0026rdquo;\nAnd, finally, there\u0026rsquo;s people like me, who have reached the last stage: Acceptance.\nNobody Misses Their Horse I\u0026rsquo;ll take a tangent, but maybe it will illustrate my point. When cars appeared, people were outraged. Horses were reliable, alive, they had character. Cars were dangerous, noisy, impractical. Everyone said so.\nLook around now. When was the last time you rode a horse to work?\nManual gearboxes had the same treatment:\n\u0026ldquo;Automatics are for lazy drivers.\u0026rdquo;\n\u0026ldquo;You don\u0026rsquo;t really know how to drive unless you drive manual.\u0026rdquo;\nNowadays most cars sold globally are automatic, and a modern dual-clutch gearbox is faster than any human hand ever was.\nI could go ahead with so many more examples, but I think you got the point. In every case, the resistance was real, and the outcome was inevitable.\nAI, to me, seems like the next in line. The pattern is always the same: resistance, denial, adaptation, normalcy. The developer community is somewhere between the first two right now, and a little bit of the third.\nThank God I\u0026rsquo;m An Engineer But hey, here\u0026rsquo;s something I\u0026rsquo;ve known for a while: I was never a developer. I was always an engineer.\nThe difference matters. Why? Because developers care about the code they write. Engineers care much more about the result. And look, this is not a difference in semantics or definition. It\u0026rsquo;s a difference in the way of thinking, something you have in your heart. You could be an engineer without even knowing and, for sure, even without a degree.\nHere\u0026rsquo;s the deal. I never lost sleep over whether my code was elegant and followed the latest cool practices written on a book by some opinionated person (that would later be challenged by some other opinionated person). I lost sleep over whether it worked. Whether it was fast enough. Whether it did what it was supposed to do. And yes, sometimes that made me unpopular in the \u0026ldquo;code is like art\u0026rdquo; circles. I\u0026rsquo;ll live.\nHere\u0026rsquo;s the uncomfortable truth that every \u0026ldquo;code is art\u0026rdquo; person does not want to hear: the final user has never, ever cared about your code.\nTwitter became X in 2023. It is now 2026. The meta tags on every website still say \u0026ldquo;twitter\u0026rdquo; (open this page\u0026rsquo;s source and just search for \u0026ldquo;twitter:card\u0026rdquo;). Millions of users, billions of visits. Nobody notices. Nobody cares.\nMost codebases out there are a maze of spaghetti, legacy systems held together with duct tape and prayers. Tech debt layered on tech debt. Thousands of engineers and developers who all made different decisions at different times, with different priorities. They are still making billions. So, really, who cares?\nAnd before anyone misunderstands me, let\u0026rsquo;s be super clear.\nI am not saying or suggesting to write bad code. I have a dignity.\nBut obsessing over code beauty? Really? While your product has bugs everywhere, is not performant at all, and everyone else is shipping? That is just vanity.\nArt Is Dead Here is what I think the \u0026ldquo;code is art\u0026rdquo; crowd is really afraid of: if AI can write code, then the craft they built their identity around is suddenly worth a lot less.\nAnd they are half right.\nIf your identity is \u0026ldquo;I write beautiful code\u0026rdquo;, well then, my friend, AI is definitely your enemy. If your identity is \u0026ldquo;I build things that work\u0026rdquo;, AI is an awesome gift. Users, CEOs, people in general do not care if your code is pretty. What matters is that the output is correct, the product works, and everything does what it should in a reliable (and fast) way.\nShocking news: AI can produce that.\nYou know what\u0026rsquo;s even more funny? AI can also read spaghetti code, understand it, and work with it just fine, without complaining.\nCode Is Dead Let\u0026rsquo;s circle back to the provocative title of this post. Is coding really dead? Well, not really.\nIs designing, architecting, and finally implementing a good, robust, performant, reliable product something that AI can do alone? Not really. Not for now, at least.\nBut here\u0026rsquo;s my point: nobody misses their horse, either. Nobody is sitting around lamenting the golden age of travelling by horse. They just moved on. And the world got faster.\nAI is a tool. A remarkably powerful one. So if you\u0026rsquo;re an engineer (read: \u0026ldquo;someone who cares about the result\u0026rdquo;), it\u0026rsquo;s probably the best tool you\u0026rsquo;ve ever had.\nBecause users, especially nowadays, aren\u0026rsquo;t waiting. The products aren\u0026rsquo;t pausing. And the people who embrace this truth are already shipping while the rest are still arguing about it.\n","permalink":"https://simonetavoletta.it/blog/code-is-dead/","summary":"\u003cp\u003eThere\u0026rsquo;s a song from Bo Burnham that I really like. It\u0026rsquo;s called \u003ca href=\"https://open.spotify.com/track/5KqdkuWE9AkTtoAiPpdD0E\" target=\"_blank\" rel=\"noopener noreferrer\"\u003eArt is Dead\u003c/a\u003e and it starts like this:\u003c/p\u003e\n\u003cblockquote\u003e\n\u003cp\u003e\u003cem\u003e\u0026ldquo;Engineers like to seem complicated\u003cbr\u003e\nBut we\u0026rsquo;re not complicated\u003cbr\u003e\nI can explain it pretty easily\u0026hellip;\u0026rdquo;\u003c/em\u003e\u003c/p\u003e\u003c/blockquote\u003e\n\u003cp\u003eSorry, it was \u0026ldquo;Entertainers\u0026rdquo;, not \u0026ldquo;Engineers\u0026rdquo;, my bad.\u003c/p\u003e\n\u003cp\u003eAnyway, in the last few weeks I\u0026rsquo;ve read a lot of posts and watched lots of videos from developers and software engineers all over the world, famous and not, being frustrated over AI. Frustrated, scared, dismissive. The whole spectrum.\u003c/p\u003e","title":"Code is Dead, Long Live the Code!"},{"content":"I run a website on Cloudflare Workers, paid plan. I also happen to care (a lot) about performance, so I always pay attention to metrics like Worker CPU Time and D1 database reads, not just Core Web Vitals.\nUsing Cloudflare Workers for the website is a constraint I chose, and I like it because it forces me to think about performance at every level. Most months, I stay well below every threshold defined in the official documentation.\nSo when my bill came in at $5.03 instead of $5.00 flat, I noticed. I mean, who notices three cents? I do.\nThe Bug Something was off, but I didn\u0026rsquo;t know what. After reading the bill, I found the culprit. I use Workers AI occasionally, a feature that uses \u0026ldquo;Neurons\u0026rdquo; as its unit of measure.\nThe official pricing docs are pretty clear:\n\u0026ldquo;Our free allocation allows anyone to use a total of 10,000 Neurons per day at no charge.\u0026rdquo;\nThat month, I used roughly 2,800 Neurons total. In a full month. Not even close to 10,000 in a single day. I should have paid exactly $0.\nBut I paid $0.03. So I opened a support ticket, and this is where the fun part begins.\nThe (missing) Support I want to be clear: I\u0026rsquo;m not writing this to drag Cloudflare. I like their product, in fact I use it and would recommend it. This post is about something else.\nBut yeah, I waited two weeks. Two weeks of silence from a multi-billion dollar company, for a billing question. I pinged them multiple times. Nothing.\nOut of frustration, I posted in the Cloudflare Community forums. And something interesting happened: I got a response within hours.\nA Cloudflare team member apologized and said they\u0026rsquo;d ping the billing team to follow up on my original ticket. One hour later, I had a reply.\nThe reply, though, was wrong. The billing agent told me I had used 2,800 Neurons and that, multiplied by the rate, it came to $0.03. Correct math, wrong result. The entire point of my ticket was that those 2,800 Neurons should have been free, because they were well below the 10,000 per day threshold.\nI pointed this out and after some back and forth, they agreed, apologized, and issued a refund.\nGreat. Issue resolved, right?\nThe Punchline The next month\u0026rsquo;s bill: $5.02. And I\u0026rsquo;m like: \u0026ldquo;What?\u0026rdquo;\nLong story short, checking the bill again showed that the bug was never fixed. They knew about it, refunded me, and a full month later, nothing changed. They wrongly charged me again, this time with $0.02 more.\nI didn\u0026rsquo;t (and won\u0026rsquo;t) open another ticket for two cents, honestly. But that\u0026rsquo;s not the point.\nThe Point Here\u0026rsquo;s where I want to zoom out, because blaming Cloudflare would be too easy and also unfair. Why? Because this is a structural problem. It\u0026rsquo;s the kind of thing that happens when:\nThe team that writes the docs is not the team that writes the billing logic The team that handles support tickets is not the team that can fix the underlying bug The team that responds to community posts is not the team that owns the ticket queue No single person is wrong here. The problem is the gap between them.\nThere\u0026rsquo;s a principle in software engineering that describes this almost too perfectly.\nConway\u0026rsquo;s Law In 1967, Melvin Conway observed something that has aged like fine wine:\n\u0026ldquo;Organizations which design systems are constrained to produce designs which are copies of the communication structures of those organizations.\u0026rdquo;\nThis is known as Conway\u0026rsquo;s Law. Simply put: your software will look like your org chart. If your teams don\u0026rsquo;t talk to each other, your systems won\u0026rsquo;t talk to each other either.\nIn the Cloudflare billing case, it\u0026rsquo;s almost textbook. The docs team, the billing system, the support team, and the engineering team are clearly separate silos (or at least, they seem to be). They don\u0026rsquo;t share enough context to close the loop on something as simple as \u0026ldquo;the billing code doesn\u0026rsquo;t match what the docs say\u0026rdquo;.\nThis isn\u0026rsquo;t unique to Cloudflare. It\u0026rsquo;s how most large tech companies actually work.\nAI as a Multiplier Now here\u0026rsquo;s where it gets even more interesting, and where I think the industry has a real blind spot nowadays.\nAI is a force multiplier. It makes fast teams faster, productive engineers more productive, and good processes more efficient.\nWell, it also makes slow teams slower, siloed organizations more siloed, and broken processes break harder.\nThe Harvard Business Review captured it well: AI is reinforcing functional silos rather than breaking them down, with departments adopting tools independently and generating fragmented gains that don\u0026rsquo;t add up to real impact. The World Economic Forum echoed the same point: enterprise AI works through integration, not silos.\nEveryone is rushing to go \u0026ldquo;AI-First\u0026rdquo;. But if your organization already has communication problems, if teams are siloed, if Conway\u0026rsquo;s Law fingerprints are all over your systems, then in my opinion going AI-First just means you\u0026rsquo;ll move faster in the wrong direction.\nThe Real Bottleneck I\u0026rsquo;ve seen this in practice everywhere. The AI tools are ready. The models are capable. The infrastructure is there.\nWhat slows everything down is the org. The handoff between teams. The ticket that sits for two weeks because no one owns it end-to-end. The bug that isn\u0026rsquo;t fixed because the person who understood the context is in a different department from the person who can deploy the fix.\nConway\u0026rsquo;s Law is more visible than ever precisely because AI amplifies everything. As WunderGraph explored in a piece on Conway\u0026rsquo;s Law and AI: if AI agents join the team as participants rather than tools, the organization\u0026rsquo;s communication structure will shape how they collaborate too. The law doesn\u0026rsquo;t go away. It scales up.\nSo listen carefully, you C-level Executives out there. The cracks in your communication structure will show up faster, louder, and more expensively when you add AI to the mix.\nFrom AI-First to Fix-Your-Org-First I got my three cents back. I\u0026rsquo;ll probably keep getting charged some cents a month for Neurons I shouldn\u0026rsquo;t be billed for, because the alternative is spending more time on tickets than the charge is worth. Which is, ironically, how most broken processes survive. The cost of fixing them feels higher than the cost of living with them.\nUntil it doesn\u0026rsquo;t.\nIf you\u0026rsquo;re planning an AI-First transformation, great. AI genuinely changes what\u0026rsquo;s possible. But before rolling out agents, automation, and LLM pipelines, ask yourself: does my team communicate well enough that AI will amplify the good stuff? Or will it just make it faster at going the wrong way?\nConway\u0026rsquo;s Law doesn\u0026rsquo;t care about your AI strategy. It just reflects your org back at you. So, please, fix your org first.\n","permalink":"https://simonetavoletta.it/blog/moving-faster-in-the-wrong-direction/","summary":"\u003cp\u003eI run a website on Cloudflare Workers, paid plan. I also happen to care (a lot) about performance, so I always pay attention to metrics like Worker CPU Time and D1 database reads, not just Core Web Vitals.\u003c/p\u003e\n\u003cp\u003eUsing Cloudflare Workers for the website is a constraint I chose, and I like it because it forces me to think about performance at \u003cstrong\u003eevery\u003c/strong\u003e level. Most months, I stay well below every threshold defined in the official documentation.\u003c/p\u003e","title":"Moving Faster in the Wrong Direction: Why AI Won't Fix What's Already Broken"},{"content":"I\u0026rsquo;m not a finance guy. I want to be upfront about that.\nBut those who know me also know that I\u0026rsquo;m very interested in economics and finance.\nSo, lately I\u0026rsquo;ve been paying more attention to what\u0026rsquo;s happening in markets, partly out of curiosity and partly because it affects everyone whether you like it or not. So bear with me while I try to make sense of something that caught my attention this week.\nBlackRock is limiting withdrawals from one of its private credit funds.\nYou can read more here or on any other news website. Not just BlackRock. Morgan Stanley and Blue Owl Capital are doing the same thing with similar products. And when three of the biggest names in finance make the same move at the same time, it\u0026rsquo;s probably worth paying attention.\nWhat even is private credit? Okay, quick detour, because I had to look this up too.\nPrivate credit is basically lending that happens outside of banks and public markets. A company needs money, but instead of going to a bank or issuing public bonds, it borrows directly from a specialized fund. That fund raised money from investors and now lends it out to companies, collecting interest in return.\nThis whole sector exploded after 2008. Banks got burned by the subprime mess and became way more conservative about who they\u0026rsquo;d lend to. Private credit funds saw the gap and jumped in: \u0026ldquo;Banks won\u0026rsquo;t give you money? We will.\u0026rdquo;\nFast forward to today and it\u0026rsquo;s a multi-trillion dollar market.\nSo far so good. The problem is how these investments actually work.\nThe illiquidity trap When you buy a stock or a public bond, you can sell it whenever you want. There\u0026rsquo;s a market, there are buyers, you click a button and you\u0026rsquo;re done.\nIn private credit, there\u0026rsquo;s no secondary market. You can\u0026rsquo;t just exit when you feel like it. The fund manager decides when redemptions are allowed, on their own schedule. And the \u0026ldquo;value\u0026rdquo; of your investment is estimated, not discovered by the market.\nThe weird side effect of this is that the price of your investment doesn\u0026rsquo;t fluctuate daily, so it feels stable (like house prices). So it feels safe. But the risk is still there, just hidden. You can\u0026rsquo;t see it until everyone tries to exit at once.\nHere\u0026rsquo;s where it gets bad Default rates on these private loans have been climbing. We\u0026rsquo;re talking from roughly 2.6% to nearly 5% over the past few months. And when defaults rise that fast, investor confidence collapses even faster.\nBlackRock reportedly saw withdrawal requests hitting around 10% of the fund\u0026rsquo;s value in a short window. The problem is that the money was already lent out. You can\u0026rsquo;t just make up liquidity. So they did what funds do in this situation: they slowed withdrawals down.\nYeah, it\u0026rsquo;s basically a bank run, but in private fund.\nWhat about the Stock Market? Let\u0026rsquo;s say that everything is fine tho, so even if the private credit market is not stonks at all, the stock market is fine as it is completely unrelated. Right?\nThis is where it gets interesting, and where I think everything is connected (or will be).\nDid you notice that in the last days the S\u0026amp;P 500 has been struggling? Well, we know that part of that is geopolitical for sure. The Iran conflict has pushed oil prices up a lot. Big jump in a short time, that\u0026rsquo;s true for sure.\nSo, like I said, it might be completely unrelated to private credit. But I think it\u0026rsquo;s actually not.\nLet me break down what I\u0026rsquo;m thinking (maybe I\u0026rsquo;m wrong):\nHigher oil price means higher inflation (more or less). Higher inflation means central banks keep rates elevated instead of cutting them. Higher rates means the companies that borrowed through private credit funds are paying more to service that debt. More of them default. Which means more investors panic about their private credit funds. Which means more withdrawal requests. See a pattern there?\nWhen investors can\u0026rsquo;t get their money out of illiquid funds, they sell what they can sell. Stocks. ETFs. Liquid bonds. Everything that has a button to click. That wave of forced selling could drag the S\u0026amp;P down even more.\nIt\u0026rsquo;s not one crisis. It\u0026rsquo;s the same crisis slowly showing up in different places at once.\nThe takeaway Is this 2008? Honestly, I don\u0026rsquo;t know. I guess nobody does right now. But the warning signs are there:\nwithdrawal freezes panic redemptions forced selling (possibly) cascading into public markets That\u0026rsquo;s all I can say. I\u0026rsquo;m just a software engineer trying to understand why my index funds are doing weird things.\n","permalink":"https://simonetavoletta.it/blog/private-credit-crisis/","summary":"\u003cp\u003eI\u0026rsquo;m not a finance guy. I want to be upfront about that.\u003c/p\u003e\n\u003cp\u003eBut those who know me also know that I\u0026rsquo;m very interested in economics and finance.\u003c/p\u003e\n\u003cp\u003eSo, lately I\u0026rsquo;ve been paying more attention to what\u0026rsquo;s happening in markets, partly out of curiosity and partly because it affects everyone whether you like it or not. So bear with me while I try to make sense of something that caught my attention this week.\u003c/p\u003e","title":"What's the deal with private credit funds in the US?"},{"content":"Let\u0026rsquo;s be honest: nowadays, everybody says they are AI-First.\nAI-First Engineer. AI-First Designer. AI-First this. AI-First that. I mean, I get it, it\u0026rsquo;s the trend, it\u0026rsquo;s the job market speaking. AI is everywhere (what do you read in the above image now?).\nHere\u0026rsquo;s the thing: I\u0026rsquo;ve been using AI since 2022. That\u0026rsquo;s four years now. And I never once thought of it as \u0026ldquo;AI-First\u0026rdquo;. To me, it was just\u0026hellip; a tool. A very good autocomplete. A brainstorming partner. A second pair of eyes for code review. I wrote my own code, and AI helped me refine it.\nSo, when everyone started slapping \u0026ldquo;AI-First\u0026rdquo; on their LinkedIn title, I didn\u0026rsquo;t feel like I fit in that bucket. I wasn\u0026rsquo;t orchestrating agentic AI fleets to conquer the world. I was just using AI as a smart assistant on the side.\nIf You Can\u0026rsquo;t Beat \u0026lsquo;Em, Join \u0026lsquo;Em A few months ago, something shifted.\nI started noticing it in the job postings at first. Everyone wants someone who can orchestrate AI agents, not just use them as an autocomplete engine. The market is literally screaming for AI.\nSo I thought\u0026hellip; fine. If you can\u0026rsquo;t beat \u0026rsquo;em, join \u0026rsquo;em. Let\u0026rsquo;s see what this AI-first approach actually looks like in practice.\nI started experimenting. Multiple tools, multiple approaches. Most of them were\u0026hellip; fine. Useful, but nothing that really changed my life or how I worked.\nThen I tried OpenCode with the Oh-My-OpenCode plugin.\nAnd I was astonished.\nWell, small disclaimer here: I was astonished when using it with good AI models like Claude Opus. Otherwise, it\u0026rsquo;s pretty similar to all the other tools.\nAn Agentic AI Fleet What makes this combination special isn\u0026rsquo;t just one feature, it\u0026rsquo;s the overall architecture. You\u0026rsquo;re not talking to a single AI. You\u0026rsquo;re working with a fleet: multiple agents, multiple sub-agents, multiple workers all coordinated to tackle whatever task you throw at them.\nI\u0026rsquo;ll not go into the details, but I\u0026rsquo;ve been using it with Claude Opus 4.6, and it works like a charm. I give it a prompt, like for example \u0026ldquo;rebuild my website following this design\u0026rdquo; (please do notice the \u0026ldquo;following this design\u0026rdquo; part, it will come in handy later on in the post), and the fleet divides the work, executes in parallel, coordinates, reviews, and ships.\nLeft to my own devices, rebuilding this site would have taken me days. With the AI? A few hours.\nThe orchestration layer that Oh-My-OpenCode provides on top of OpenCode is what makes it click. It\u0026rsquo;s not just \u0026ldquo;AI writes code\u0026rdquo;. It\u0026rsquo;s \u0026ldquo;AI manages a team that writes, reviews, and ships code\u0026rdquo;.\nAnd that\u0026rsquo;s where it gets really interesting.\nWhat This Means for My Job Yeah, I mean, let\u0026rsquo;s get philosophical.\nI now genuinely believe that simple coding tasks like boilerplate, standard CRUD, straightforward features, etc. will be automated by AI in a few years. No question. Like, no question at all.\nBut wait, before firing everyone in your company (yes, I\u0026rsquo;m looking at you, C-level executives), let\u0026rsquo;s just add that I don\u0026rsquo;t think engineers are going away. At least, not the ones who can think, architect, and guide.\nWhy? Because AI still can\u0026rsquo;t do good reasoning. Not really. It can simulate it, generate convincing outputs, but when it comes to truly understanding why something should be built a certain way, why a trade-off matters in a specific context, why this architecture will scale and that one won\u0026rsquo;t, that\u0026rsquo;s still human territory (remember the \u0026ldquo;following this design\u0026rdquo; part before?).\nMy job title hasn\u0026rsquo;t changed. I\u0026rsquo;m still a Software Engineer. It\u0026rsquo;s what that means day-to-day that\u0026rsquo;s shifting.\nI used to write code. Now I architect AI to write code for me.\nI used to be the builder, getting my hands dirty. Now, I\u0026rsquo;m the orchestrator of the actual builders.\nEver played Overlord? That\u0026rsquo;s how I feel now (without the evil part of course).\nThinking About the Industrial Revolution Analogy Old topic, everyone wrote about this I think. I\u0026rsquo;ll keep it short.\nAs you know, during the industrial revolution, craftsmanship didn\u0026rsquo;t disappear: it evolved. The skilled artisan who made every chair by hand was displaced, yes. But the people who understood chairs, who could design them, who knew materials and ergonomics and aesthetics, they adapted. They became designers, engineers, factory planners.\nThe chair was still there. The craft of making it was industrialized.\nSoftware is heading the same way. The \u0026ldquo;chair\u0026rdquo;, i.e. the working software product, will still exist. But the craft of writing every line of code will be industrialized. Automated. AI-Driven.\nEngineers who adapt will become the designers and architects of this new industrial era. The ones who can\u0026rsquo;t or won\u0026rsquo;t adapt\u0026hellip;\nI don\u0026rsquo;t think I\u0026rsquo;m being dramatic when I say: in five years, the title \u0026ldquo;Software Engineer\u0026rdquo; as we know it will mean something fundamentally different. Maybe it already does.\nThis Website as a Proof of Concept Right now, AI cannot build a website by its own, that\u0026rsquo;s been obvious for a while. Why? Because it\u0026rsquo;s struggling with UI, UX, behavior and more in general, it\u0026rsquo;s stuck on empty templates without a soul or idea behind them.\nSo, it\u0026rsquo;s obvious that AI didn\u0026rsquo;t do everything for me. I re-built this website with AI. Remember, the orchestration layer, the human brain directing the fleet. This is the real skill now.\n99% of this site was generated by AI. I gave prompts. I reviewed. I guided. I made the decisions that mattered. The AI did the rest.\nAnd honestly? It looks better than what I would have built on my own. More consistent. More polished. More designed, because I could focus on what I wanted rather than getting lost in the weeds of CSS and HTML.\nSo, About This Post Of course I wrote this post with the help of AI!\nNot because I\u0026rsquo;m lazy, though I\u0026rsquo;ll admit it\u0026rsquo;s nice to offload the actual typing, but because this is the point. I thought it, I guided it, I reviewed it. The words on the page are 99% AI-generated, but the ideas, the structure, the reason you\u0026rsquo;re reading this in the first place. That\u0026rsquo;s all me.\nCall it AI-first if you want. I call it the natural evolution of being a software engineer in 2026.\nThe industrial revolution didn\u0026rsquo;t kill craftsmanship. It elevated it.\nWell, it also displaced a lot of workers and cost many lives. Don’t forget that.\nLet\u0026rsquo;s see where this takes us.\n","permalink":"https://simonetavoletta.it/blog/ai-rebuilding-website-and-career/","summary":"\u003cp\u003eLet\u0026rsquo;s be honest: nowadays, everybody says they are \u003cstrong\u003eAI-First\u003c/strong\u003e.\u003c/p\u003e\n\u003cp\u003eAI-First Engineer. AI-First Designer. AI-First this. AI-First that. I mean, I get it, it\u0026rsquo;s the trend, it\u0026rsquo;s the job market speaking. AI is everywhere (what do you read in the above image now?).\u003c/p\u003e\n\u003cp\u003eHere\u0026rsquo;s the thing: I\u0026rsquo;ve been using AI since \u003cstrong\u003e2022\u003c/strong\u003e. That\u0026rsquo;s four years now. And I never once thought of it as \u0026ldquo;AI-First\u0026rdquo;. To me, it was just\u0026hellip; a tool. A very good autocomplete. A brainstorming partner. A second pair of eyes for code review. I wrote my own code, and AI helped me refine it.\u003c/p\u003e","title":"I Let AI Rebuild My Website (And My Career)"},{"content":"Here you can check out (almost) everything I\u0026rsquo;ve tinkered with in my spare time.\n","permalink":"https://simonetavoletta.it/portfolio/","summary":"\u003cp\u003eHere you can check out (almost) everything I\u0026rsquo;ve tinkered with in my spare time.\u003c/p\u003e","title":"Portfolio"}]