DEV Community: Singaraja33

Babel or Jerusalem: The Pope's framework for developers.

Singaraja33 — Thu, 11 Jun 2026 07:23:14 +0000

During Pope Leo's presentation of his first Encyclical text last May at the Vatican, there was a word he used that made many people wonder.

Standing in front of Cardinals, diplomats, and AI top leaders (including Chris Olah, one of the founders of Anthropic), the Pope said that artificial intelligence needs to be "disarmed". He didn't mention it should be regulated, governed or supervised. He said "disarmed", and when talking about such topic and in this specific act, for sure this word was carefully chosen.

He chose the word deliberately, explaining that this moment "needs words capable of attracting attention, awakening consciences and indicating paths forward", and right after he said something else that deserves to sit with every developer and pay special attention to: the Church does not claim technical answers, but brings "a wisdom concerning the human that our present time desperately needs".

That sentence, from a man such as the Pope (a Chief of State and the leader of over 1,4 billion Catholics), addressing an audience that builds the systems in question, is not something we should take lightly. It is an invitation. And the full 245 paragraphs of his Magnifica Humanitas encyclical, on safeguarding the human person in the time of AI, are worth taking seriously not despite the fact that they come from a Pope, but in part because of it.

To this day, it's obvious that the Catholic Church has been thinking about what technology does to human beings for a very long time. It was doing it before the chips came into play, before the algorithms appeared and before the first line of code was ever written, and in fact the very same date Leo XIV chose to sign the encyclical was not accidental. It was May 15, 2026, exactly on the 135th anniversary of Rerum Novarum, the encyclical written by his predecessor and namesake Leo XIII in 1891 in response to the Industrial Revolution.

That previous Encyclical, Rerum Novarum, was surely one of the most consequential documents in the history of social thought. It was written at the exact moment when factory owners were treating workers as interchangeable units of production, when underaged children worked over 14 hours a day, when families were being uprooted from agricultural communities and deposited into urban slums with no safety net, no recourse, and no social or political voice.

Back then, the market, left entirely to its own logic, had begun optimizing for output at the expense of the people doing the work, and then Pope Leo XIII said: "...this is not acceptable, not because markets are evil, but because human beings are not inputs. They have inherent dignity. Their labor is an extension of that dignity, not a commodity. The state, employers, and institutions all carry obligations that the profit motive alone cannot be trusted to honor".

It took decades for those ideas to fully materialize in labor law, workers rights, and social policy, but they eventually did and the world is now measurably better for it. The wisdom basically arrived before the legislation, and the legislation needed the wisdom to know what it was trying to protect.

As it happened to the Pope back then, now Leo XIV looked at AI in 2026 and saw the same shape of problem, moving forward to elaborate a very well built and extremely symbolic Encyclical where the organizing metaphor at its heart is striking in its clarity.

The encyclical frames the choice before humanity not as "AI yes or no" but as Babel or Jerusalem, with Babel meaning centralized power, optimizing for its own expansion, indifferent to the human cost and building something that ultimately fragments rather than connects. Jerusalem, in the other hand, means a shared reconstruction centered on dignity and oriented toward the common good, building something that makes human community possible rather than undermining it.

In any case, the Encyclical does not say AI is Babel, but what it says is that AI can turn into either, and that which one it becomes depends entirely on the choices made by the people building it, funding it, deploying it and governing it right now. Because according to the Pope, the technology is not neutral and it carries the values (or the absence of values), of those who made it. This is the concern that runs through every chapter.

The Encyclical names aspects of AI like job insecurity, manipulation of information, privacy violations, ideological bias and autonomous weapons as specific risks, but maybe the deeper risk underneath all of them is that the acceleration of AI deployment could be outpacing the moral and social infrastructure needed to shape it toward human ends. That the decisions being made this year, mostly by a small number of people and at extraordinary speed, will have consequences that fall most heavily on people who had no voice in making them.

Leo XIV warned specifically that rapid automation could displace workers and reshape labor markets in ways that risk leaving many in "forced inactivity" undermining both human dignity and social stability, in a clear echo of Rerum Novarum that is hard to dismiss as just a coincidence.

So once this is explained, and looking deeper into what this actually means for those of us building with AI, it must be noticed that most developers working on AI products are not trying to build Babel in purpose but instead they are just trying to solve real problems by using the outstanding advantage AI brings. They care about what they make, but the question Magnifica Humanitas puts to them is not if they are good or bad guys, but something much harder: Are the systems you're building designed to answer to the people they affect?

That is the question and this question matters in practice, not just in theory, because an AI system that optimizes for engagement metrics without accounting for the psychological wellbeing of users is not necessarily malicious, but it's just indifferent to a variable it was never asked to consider.

An AI hiring tool that produces discriminatory outcomes isn't evil, it's just a model that learned patterns from data that encoded historical inequity, and nobody built in a correction. An AI system deployed in a low income country that was trained entirely on data from wealthy ones is not necessarily cruel, but it's just misaligned in ways that weren't caught before deployment, and the people who pay the price for that misalignment had no seat at the table when it was designed. And so on.

These are engineering and governance problems and according to what the Pope says they also have engineering and governance solutions. The Pope is not asking developers to pray over their pull requests. He's asking the industry as a whole to take seriously a question that market incentives alone cannot answer. He asks the industry to think about who are their AI's designed for, who might it harm and what would we need to build differently if those people's wellbeing actually constrained our design choices...

The Encyclical's framing is clear and it points out that the challenge is not simply to manage technology, but to shape it with conscience so that it truly serves life, not just the ambitions of a few, and that's not a theological position but it should be a design principle. And it's one that many developers who will be most proud of their work in twenty years time from now are already trying to live by.

Going back to the Industrial Revolution's Previous Encyclical, written in 1891, the wisdom in Rerum Novarum was considered idealistic by the people with the most power to ignore it. Brutal things like child labor laws came anyway, the eight hour workday came anyway, social safety nets came anyway, the market didn't voluntarily generate those protections and the moral argument had to be made clearly enough and persistently enough that it eventually became politically irresistible.

In our own current case, we are still early in that process with AI, the legislation is fragmented and behind the technology, the governance frameworks are still forming, the power is highly concentrated and the people most affected have the least influence. So in line with the previous experience with Rerum Novarum, what Magnifica Humanitas offers is basically a framework for thinking about what we're actually trying to protect before the specific rules get written. It's a reminder that human dignity is not a variable to be weighted against efficiency, that the common good is a real constraint (not a marketing strategy), and that technology shaped without conscience produces outcomes that conscience, arriving later, will spend decades trying to repair.

Developers are not just spectators to that process. They are, in fact, right now making the decisions that will determine which side of the Babel / Jerusalem choice we end up on.

The Pope showed up in person to say so. The co founder of one of the largest AI companies was in the room to hear it. Both of those things, happening simultaneously, feel like a moment worth paying attention to, and we should all create our true opinion.

Sources:

Full Magnifica Humanitas Encyclical text:
Vatican.va
https://www.vatican.va/content/leo-xiv/en/encyclicals/documents/20260515-magnifica-humanitas.html
Analysis for engineers and builders
ExplainX.ai
https://www.explainx.ai/blog/magnifica-humanitas-pope-leo-xiv-ai-encyclical-2026
Historical context towards Rerum Novarum Encyclical Vatican News
https://www.vaticannews.va/en/pope/news/2026-05/pope-leo-xiv-encyclical-magnifica-humanitas-ai.html
AI Regulation and facts
https://luisyanguas22.medium.com/the-truth-about-ai-regulation-power-and-the-real-winners-behind-all-of-it-da9c0367c3d5

www.translockit.com

The guy managing a 2,1 trillion USD fund thinks AI might be a bubble, and he can be partially right.

Singaraja33 — Wed, 03 Jun 2026 05:40:49 +0000

Nicolai Tangen is the man behind Norway's sovereign wealth fund, a kind of mega fund owned by the small but oil ultra rich country and which owns approx 1,5% of every publicly listed company on this planet. To put this in perspective, we talk about approximately 7.200 leading companies across 60 countries.

When a guy like this wakes up in any given morning, before he had his first coffee, the needle of his 2,1 trillion might be moving up or down by a number of geopolitical tensions, currency fluctuations, interest rate decisions or earning surprises in industries he doesn't even particularly follow.

His job at the wheel of this giant is basically to see things coming before they arrive, and calculate the outcomes of any chosen investment or industry the fund is invested in, so when Tangen jumps into an interview saying that an AI bubble is one of the two greatest threats to global markets right now (the other according to him is geopolitical fragmentation), it is worth taking seriously.

About a month and a half ago, his fund formally identified that a potential AI bubble would be a major risk scenario that could potentially erase 35% off the fund's value in a worst scenario case, an along interviews where he participated between last year and this one, Tangen has been consistently saying that the concentration of capital into AI related stocks is creating, in his own words, "a risk we have never seen before".

He thinks that basically the market may be running ahead of what the technology can actually deliver in the near term, and by checking his funds investments anyone can see that he is definitely not a technophobe making this argument from ignorance. In the contrary, he has spent years (in his own words) "running around like a maniac" to get his 700 staff to use AI in their daily work.

His fund actually right now uses large language models, including Claude, to analyse every new equity investment on the day it enters the portfolio, generating daily risk assessments in real time, and a very big number of his employees code their own AI tools. So Tangen is a genuine believer in what the technology does, but his concern goes way further and is about what the market is actually pricing in. That differentiation (between the technology and the valuation), is for him the key of everything.

If we are to give a true analysis, the concentration argument is quite hard to dismiss when we see things like seven single companies now representing roughly a third of the entire SP500. Another indicator is the Shiller CAPE ratio, a inflation oriented measure that has historically preceded major corrections and that now sits around 38 to 40, levels only exceeded at the dotcom peak of 44.
We can clearly see that the market is pricing in a future where AI delivers transformational returns at a speed and scale that, looking at enterprise adoption data, is not yet materializing.

And maybe the part of Tangen's argument that is most interesting is what he describes as a structural trap, with him explicitly saying that even if he believed AI valuations were stretched, he could not simply sell out of the AI companies because they are already too large a share of the market that for a fund his size, reducing exposure to giants like Nvidia, Microsoft or Apple would mean moving markets against himself (basically selling into a decline he helped starting)

He is, according to his own words, an investor with "skin in the game" in a way that it even affects his own investing behavior. That is an uncommon and honest admission from someone in his position, and it also points to a systemic fragility that most optimistic arguments quietly diminished.

The enterprise adoption data gives Tangen's caution some realistic sense too, as despite extraordinary capital investment, most of the productivity gains from AI are still showing up in individual measurements rather than in corporate earnings at the scale the current impressive valuations imply. The MIT study finding that 95% of enterprise AI pilots fail to scale is not a detail that trillion dollar valuations should be able to ignore indefinitely.

In any case, there are also points on Tangen's speech that might be incorrect or at least incomplete, because there is a problem with the bubble analogy that we already explored in a previous article a few weeks ago: The companies at the center of the AI rally cannot be compared with those of the Dotcom era, simply because they are generating cash at a scale that has no historical precedent, not even close.

Well known companies like Apple, Microsoft, Alphabet, Amazon and Meta combined for roughly 350 billion USD in free cash flow in their most recent fiscal years, with Nvidia alone marking a net income exceeding 120 billion USD in 2026. These are not speculative bets on future revenue that may never materialize, as it clearly was in the case of previous tech crashes. They are simple audited financial statements from the most profitable enterprises in the history of capitalism.

But the infrastructure argument is also something that the bubble narrative struggles to fully explain, when we see companies like Microsoft independently committing more than 80 billion USD to AI infrastructure, Alphabet 85 and Meta between 115 and 135 billion USD. These are accounting decisions funded by existing free cash flow, not leveraged bets financed by faithful or blind investors.

The companies building AI capacity and infrastructure have among the strongest balance sheets in the entire equity market, and are setting their decisions not by guessing but by building with a financial cushion that no Dotcom company ever had.

And finally there is the adoption curve itself, a factor where Tangen is also conservative but where we should be confortably positive about when we see that enterprise AI deployment is still, as of now in mid 2026, in measurable very early stages, with companies like Gartner projecting that 40% of enterprise applications will integrate AI agents by end of 2026, up from less than 5% in 2025. If that curve materializes, even if it's on a 50% basis of what Gartner expects, the earnings growth that would justify current valuations hasn't happened yet. It's just coming.

In any case, what makes Tangen a more interesting voice than most market commentators (apart from all his professional validations) is that he doesn't claim certainty. Actually, when he presented his fund's results back in early 2026, he said openly that it was "very difficult to gauge right now whether there was an AI bubble". So this means he is basically not anticipating the crash but instead he is flagging the risk, managing the uncertainty and being open about the limits of his own visibility. So his vision as a whole might make sense and is actually the correct response to the situation, leaving aside other superbullish or crash anticipating arguments.

The most clear version of Tangen's argument is not just a vague "AI is a bubble", but he is more meaning that the concentration of capital in AI related companies, combined with a mismatch between current valuations and earnings realization, creates a correction risk that is larger than most people are pricing in. That is a serious and defensible position that doesn't come as a prediction but as a risk assessment from someone who, by the nature of his job, is paid to think about risk.

And to defend his opinion, a 30-35% correction in AI related equities would not be crazy because the history of transformative technologies is full with periods of genuine overvaluation followed by painful corrections followed, eventually, by the technology delivering on most of its original promise.

Internet did change everything but when we look back we realise that it took a decade longer than 1999 suggested it would, and most of the companies that were going to lead that change didn't exist yet when the bubble exploted.

AI probably will change everything too. Whether the companies currently priced to do so will be the ones that deliver it, and whether that delivery will happen on the timeline the market is currently pricing is a question that even the man managinf one of the largest funds in the world says he really cannot answer, which means the rest of us should probably be thinking about that too.

Sources:

AI bubble as a risk scenario (35% fund loss)

Bloomberg, March 2026:
https://www.bloomberg.com/news/articles/2026-03-18/norway-s-wealth-fund-warns-of-ai-bubble-and-geopolitical-risks

Bloomberg video, April 28, 2026 (markets, AI and China):
https://www.bloomberg.com/news/videos/2026-04-28/norway-s-tangen-on-markets-real-estate-ai-and-china-video

"Managing $2 Trillion: AI Bubbles & Contrarian Investing", February 2026:
https://www.youtube.com/watch?v=zyvuM3J9QqQ

Seven IT giants owning 1 third of the SP500:
https://luisyanguas22.medium.com/seven-it-companies-now-own-a-third-of-the-sp500-heres-what-this-actually-means-for-developers-2803c43cf093

IPE — testimony to Norwegian parliament on "unprecedented concentration risk", May 2026:
https://www.ipe.com/news/nbims-tangen-warns-lawmakers-of-swfs-unprecedented-big-tech-concentration/10136514.article

CNBC — fund using Claude AI to screen investments + AI is "deflationary and positive" quote, April 2026:
https://www.cnbc.com/2026/04/28/norway-sovereign-wealth-fund-oil-iran-war-anthropic-claude-ai-invest-stocks.html

The incredible transformation of modern software by AI:
https://translockit.com/en/article/the-incredible-transformation-of-modern-software-development-by-artificial-intelligence

www.translockit.com
Luis Carlos Yanguas Gómez de la Serna

Magnifica Humanitas: How the Pope walked into the room full of AI engineers and said what few else dared to say

Singaraja33 — Wed, 27 May 2026 01:12:06 +0000

Check our short article in Future Forem on Pope Leo XIV and his last encyclical "Magnifica Humanitas" 👇🏻

Just two days ago, in the morning of May 25th, something absolutely unusual happened in the Vatican's Synod Hall. HH. Pope Leo XIV walked into a room filled with cardinals, diplomats and also some top guys from the AI industry, and personally presented his first encyclical. That fact alone was historically unprecedented as it was the first time in history when a Pope himself attended the launch of his own documents. The encyclical was called Magnifica Humanitas ( Latin for "Magnificent Humanity"), and he said was addressed not just to Catholics but to "every person of goodwill"

The timing was most probably not accidental when we realise that, despite not having been widely commented, Leo XIV signed it on May 15th, the very exact 135 anniversary of Rerum Novarum, the landmark 1891 encyclical in which his predecessor and namesake Leo XIII responded to the dehumanization caused by the Industrial Revolution.
The message embedded in that date is impossible to miss, and drove the minds of everyone to the fact that we are right now living through another revolution of equal magnitude, and the Church has something urgent to say about it.

The document was written originally in English and to read it in that language was specially interesting as the powerful message was better perceived, as it happens when we, non native english speakers, watch an American movie in its original version. It's opening words set the tone with strong clarity, saying the following: "Humanity, created by God in all its grandeur, is today facing a pivotal choice, either to construct a new Tower of Babel or to build the city in which God and humanity dwell together"

That is not a metaphor chosen carelessly, because what Leo XIV most clearly argues throughout Magnifica Humanitas is that technology is not our enemy (the encyclical is explicit that AI is neither "a force antagonistic to humanity" nor "inherently evil"), but it is equally explicit that technology is never neutral. It takes on the characteristics of those who devise it, finance it, regulate it and finally use it, which means that the question is not whether AI is good or bad in the abstract.

The question is what vision of the human person is embedded in the data, the models and the decisions being made right now, mostly by a very small number of people and at extraordinary speed.
This is where the encyclical becomes specifically important for those of us who build technology for a living.

The Pope argues that "a more moral AI" is not enough if that morality is determined only by a few. He calls for active political and social involvement capable of "slowing things down when everything is accelerating". Not to stop progress but to ensure that communities still have the chance to participate, ask questions and shape the future that is being built in their name.

Sitting in that room at the Vatican, listening to the Pope, was Christopher Olah, one of the founders of Anthropic which as we all know is one of the most powerful AI companies on earth, and a man who describes himself as not a believer. After the presentation, Christopher said he was grateful to the Church for "taking this work of discernment seriously" and even issued his own call: "We need moral voices that the incentives cannot bend"
That sentence, from an AI engineer at a Papal encyclical launch, says something profound about where we are.

For anyone building software, developing AI systems and making the daily decisions about what to optimize for and what to leave behind, Magnifica Humanitas is a reminder that every technical choice carries a moral weight because the human being on the other end of the product is not just a user metric. They are, in the Catholic tradition, made in the image of God. Unrepeatable and irreducible to simple data.

Despite its probable errors, nobody can deny that the Church has been thinking about human dignity for two thousand years. It was thinking about it way before the algorithm, before the chip, before the first line of code was ever written. That continuity of thought, and the insistence that no amount of efficiency justifies the erosion of what makes us human, is exactly what AI development needs more of right now.

And the Pope showed up in person to say it. That, only in itself, is very worth paying attention to.

Sources:

Full text of the encyclical:
https://www.ncregister.com/cna/full-text-magnifica-humanitas

Vatican News
https://www.vaticannews.va/en/pope/news/2026-05/pope-leo-xiv-encyclical-magnifica-humanitas-ai.html

NPR:
https://www.npr.org/2026/05/25/nx-s1-5828375/pope-leo-to-weigh-in-on-the-perils-and-promises-of-artificial-intelligence

EWTN:
https://ewtnvatican.com/articles/pope-leo-xiv-unveils-magnifica-humanitas-ai

www.translockit.com

Magnifica Humanitas: How the Pope walked into the room full of AI engineers and said what few else dared to say.

Singaraja33 — Wed, 27 May 2026 01:07:24 +0000

And the Pope showed up in person to say it. That, only in itself, is very worth paying attention to.

Sources:

Full text of the encyclical:
https://www.ncregister.com/cna/full-text-magnifica-humanitas

Vatican News
https://www.vaticannews.va/en/pope/news/2026-05/pope-leo-xiv-encyclical-magnifica-humanitas-ai.html

NPR:
https://www.npr.org/2026/05/25/nx-s1-5828375/pope-leo-to-weigh-in-on-the-perils-and-promises-of-artificial-intelligence

EWTN:
https://ewtnvatican.com/articles/pope-leo-xiv-unveils-magnifica-humanitas-ai

www.translockit.com

We reviewed many AI project failures, and this is the pattern most of them clearly show.

Singaraja33 — Mon, 25 May 2026 03:02:42 +0000

Check out our Dev.to article on the reason why a majority of AI initiatives stall, and the reason to do things properly 👇🏻

In one of our previous posts we talked about the importance of the experienced developer behind any serious AI project, and now we would like to dig more on the reason why most AI project failures happen.

To put it simple, today anyone with enough experience can confirm that by trusting the development of our ideas and projects fully into AI autonomous systems and models, what we will probably obtain as a result is a cascade of automated fixes, each one making the underlying problem slightly worse and none of them understanding the dependencies that connected everything together. The main reason behind that situation is just because AI models are designed for perfect conditions, but production is very rarely done under those "perfect conditions".

Just in 2025, the freshest year we can see statistics from, global companies invested hundreds of billions in AI initiatives. By the end of that year, estimations show that approx 80% of all that huge investment had produced no measurable results (not just low returns or disappointing returns, but literally no results), and in an analysis by The RAND Corporation, it was seen that across more than 2.400 enterprise AI initiatives, 80,3% of them failed to deliver their intended business value. And what is maybe more alarming is that these numbers have barely moved in three years, despite better models, better tooling and dramatically more organizational awareness of the problem.

Everyone in the industry knows the failure rate is high, but the question is that very few people can tell you exactly why in a way that's actually useful. So here is what the data actually shows and what the minority that succeeds is consistently doing differently.

Maybe the pattern that keeps repeating the most is simply how we think about the intrinsical risks on AI projects themselves. In a study done among 140 company AI implementations, it was seen that only 23% of failures were caused by model performance, data quality problems or integration complexity. The rest (77% of them), came down to simple strategy and organizational decisions that had nothing to do with the technology itself...Read that again: 3 out of 4 AI project failures are not technical failures but purely organizational ones.
This means the model worked, the data was good enough and the integration was achievable, but despite of it all the project itself still failed because nobody had agreed on what success looked like or because the team behind the project treated AI as an IT project rather than a business transformation, or because the team behind was not actually prepared to change how it operated around the technology it had just deployed.

This is crucially important, because it means that the most common response to AI project failure (picking a better model, hiring more data guys, switching outsourcing) is basically solving the wrong problem. The issue was never primarily in the code.

The specific failure that appears in the data with strong regularity deserves its own name, and is called "demo to production" collapse and means that many AI systems fail mainly during the transition from pilot to production. The model might perform very well in a controlled environment, impressing everyone in the room and with budgets easily approved. But then when the rollout begins and the real world conditions arrive is when inconsistent data come up from systems that don't talk to each other cleanly, edge cases the demo never encountered appear and the whole thing stalls.

S&P Global found that only 48% of AI projects make it into production at all. Of those that do, the average journey from prototype to production takes eight months. Big companies abandoned an average of 2 AI initiatives in 2025, at an average cost of 7,2 million USD million per abandoned initiative.
Gartner puts a specific number on the data problem that sits underneath most of these failures, and is that 60% of AI projects that lack AI ready data will be abandoned by the end of this year 2026. Maybe more interestingly, McKinsey's 2025 research found that companies achieving significant AI returns were twice as likely to have invested in data workflow redesign before model selection. Not after, before. This simply means that companies that succeed build the foundation first and choose the model second, and the companies that fail do it the other way round, because the model is the exciting part, and foundation work is not.

To solve all this and get to clear optimal results, leadership is fundamental and plays a key role, and according to other findings 84% of AI project failures are basically leadership driven. Not engineering driven and not data driven, but just provoked by a failure in the leadership of the project itself. And this is a symptom that repeats not only in our industry but across other industries as well. To say it clear, most projects lack clear and measurable success metrics from the start because they are normally approved on the basis of strategic intent rather than defined outcomes. Teams tend to treat AI as a technology project when it is actually a business transformation, which means the people with the authority to change workflows and incentives need to be involved before it becomes too late.

Companies that consistently succeed share a specific characteristic, and this is that they generally define what success looks like in clear and measurable terms before a single line of code is written. Instead of concluding that "we want to use AI to improve customer service", they normally say "we want to reduce average customer query resolution time from 8 minutes to 3 minutes, with a customer satisfaction score above 4,5, within 90 days of deployment" That way of deciding and that leadership generates several things simultaneously: it builds a clear alignment on what is actually being built, and it means that when something goes wrong in production (and something always goes wrong) the team knows exactly what they're trying to get back to.

The previous RAND analysis we mentioned also identifies a specific profile on the minority of AI projects that deliver their intended value, and the pattern is consistent enough to be useful.
They build observable systems from day one. The successful teams log inputs, outputs, latencies and metadata from the beginning, turning what would otherwise be a black box into an clear system they can analyse in full. And even if doing this might feel like overworking in the early stages, it is actually the only thing that makes debugging in production manageable when problems arrive, and as we said, problems always arrive.

The teams that skip that previous efforts spend then months trying to reconstruct failures they could have diagnosed in minutes if initial phases were done properly. Those teams don't get to understand that maybe the most reliable AI systems in the data are human AI collaborations, and that while the AI handles volume, humans handle exceptions. This is not just a compromise or a temporary measure until the AI gets better, but it is the architecture that works in practice, across industries and consistently. Fully automated AI systems without explicit human review points fail at strongly higher rates than hybrid systems, because they can't recognize when they've missed something important.

Another recent study on March 2026, just two months ago, found an engineer using Claude to fix a condition in a payment processing platform. The AI's solution looked elegant to his eyes and passed all initial tests. It introduced 12 new bugs, leading to severe system failure. The AI didn't understand the concurrency models or production load patterns that made the original code fragile. The fix cost a lot of money in lost revenue and engineering time...AI is genuinely extraordinary at generating code that works in isolation, but as we mentioned in previous articles, it is not good at understanding the historical context, the outages, the edge cases, the undeclared dependencies and all those things that shapes what a system actually needs to survive in production.

Teams that succeed with AI share a characteristic that sounds almost insultingly simple: they choose their AI application based on where it fits a genuine, measurable business problem, not based on what the technology is theoretically capable of. "Let's use AI" is not a strategy, what looks like a strategy is "Let's automate the specific part of our customer onboarding process that currently takes 4 days and costs us 30% of customers before they reach activation".

The main risk of doing things wrong is what analysts are calling AI Capital Risk, basically meaning the exposure created when significant capital is committed to AI initiatives before structural readiness is validated. And the structural readiness question is not primarily technical but is instead organizational, architectural and strategic. That means that the most valuable resource for an AI project is not, despite what the vendor landscape might suggest, a better model or a faster compute cluster. It is experienced analysis about which problems are worth solving, how to structure a system that will survive production conditions, how to define success in ways that survive organizational changes and how to build the foundation that the technology actually requires before the technology gets selected.
The teams that have seen the failure modes and learned what the data took three years and a lot of spending to confirm, bring something that no tool, no model and no amount of internal enthusiasm can substitute for: the accumulated knowledge of what actually breaks and how to design around it before it breaks on you.

The gap between a project that impresses everyone in the demo and a project that delivers measurable value twelve months later is not simply a gap in technology but a gap in the analysis applied to every decision made in the first couple of days or weeks, and this gap is where only the 19,7% of cases live.

If you're in the early stages of an AI project and want an honest assessment of where your structural risks actually are (before the expensive part begins) at Translock IT we'd be glad to talk. The conversation costs nothing but the alternative might cost considerably more.

Sources:

RAND Corporation — analysis across 2.400+ AI initiatives
https://www.pertamapartners.com/insights/ai-project-failure-statistics-2026

MIT — 95% of GenAI pilots fail to scale
https://www.wiseback.com/why-ai-projects-failed-2025-and-2026-cx-strategy/

RAND + MIT + McKinsey + Gartner — comprehensive synthesis
https://labor411.org/411-blog/report-80-of-ai-projects-fail-overall-with-84-of-the-failures-caused-by-leadership/

S&P Global — 42% of companies scrapped most AI initiatives in 2025
https://www.folio3.ai/blog/ai-project-failure-rate-stats

Gartner — 60% of AI projects without AI-ready data abandoned
https://talyx.ai/insights/enterprise-ai-implementation-failure

McKinsey 2025 — organizations with AI returns invested in data first
https://quicklaunchanalytics.com/bi-blog/why-80-of-ai-projects-fail-before-they-start-its-your-data-foundation/

Stratify Capital 2026 — AI Capital Risk framework + structural failure analysis
https://www.stratifycapital.ai/ai-project-failure-rate

Case $47k AWS outage + case $13,540 payment endpoint (Claude race condition)
https://altersquare.io/ai-not-suited-for-architecture-decisions-no-knowledge-of-past-failures/

McKinsey Global AI Survey 2026 — 73% ROI failure rate
https://www.aigovernancetoday.com/news/enterprise-ai-spending-crisis-2026

Valuebound 2026 — architectural and organizational failure patterns
https://www.valuebound.com/resources/blog/ai-projects-fail-enterprises-2026-reality-check

www.translockit.com
Author: Luis Carlos Yanguas Gómez de la Serna

Very interesting analysis on why most AI projects simply fail, and the key to avoid that 👇🏻

Singaraja33 — Mon, 25 May 2026 02:57:06 +0000

Singaraja33

May 25

We reviewed many AI project failures, and this is the pattern most of them clearly show.

#aiprojects #ai #softwaredevelopment #translockit

8 min read

We reviewed many AI project failures, and this is the pattern most of them clearly show.

Singaraja33 — Mon, 25 May 2026 02:54:48 +0000

Sources:

RAND Corporation — analysis across 2.400+ AI initiatives
https://www.pertamapartners.com/insights/ai-project-failure-statistics-2026

MIT — 95% of GenAI pilots fail to scale
https://www.wiseback.com/why-ai-projects-failed-2025-and-2026-cx-strategy/

RAND + MIT + McKinsey + Gartner — comprehensive synthesis
https://labor411.org/411-blog/report-80-of-ai-projects-fail-overall-with-84-of-the-failures-caused-by-leadership/

S&P Global — 42% of companies scrapped most AI initiatives in 2025
https://www.folio3.ai/blog/ai-project-failure-rate-stats

Gartner — 60% of AI projects without AI-ready data abandoned
https://talyx.ai/insights/enterprise-ai-implementation-failure

McKinsey 2025 — organizations with AI returns invested in data first
https://quicklaunchanalytics.com/bi-blog/why-80-of-ai-projects-fail-before-they-start-its-your-data-foundation/

Stratify Capital 2026 — AI Capital Risk framework + structural failure analysis
https://www.stratifycapital.ai/ai-project-failure-rate

Case $47k AWS outage + case $13,540 payment endpoint (Claude race condition)
https://altersquare.io/ai-not-suited-for-architecture-decisions-no-knowledge-of-past-failures/

McKinsey Global AI Survey 2026 — 73% ROI failure rate
https://www.aigovernancetoday.com/news/enterprise-ai-spending-crisis-2026

Valuebound 2026 — architectural and organizational failure patterns
https://www.valuebound.com/resources/blog/ai-projects-fail-enterprises-2026-reality-check

www.translockit.com
Author: Luis Carlos Yanguas Gómez de la Serna

Almost anyone can build an App in 2026, but here's the part nobody mentions

Singaraja33 — Fri, 22 May 2026 02:48:33 +0000

Our article on Dev.to on Vibe Coding, setting up Apps and the need of experienced developers today 👇🏻

When anyone starts trying to vibe code on a project for the first time, it's quite typical that after typing something like "build me a project management app with a nice dashboard and user authentication" they become amazed on how with just such a simple prompt they can see in front of their eyes a fully functional application assembling itself. And not only that, but also the app buttons work, the overall layout looks clean and they can actually click around in it quite properly. The whole thing took just a couple of minutes and without the need of writing a single line of code.

That feeling is now happening, it's real and the market has rewarded it if we just look at the numbers and realise that vibe coding tools have attracted over a billion USD in venture capital in 2025 alone. Lovable, a well known platform we've written about before and that was funded by a simple Nordic young guy, hit 200 million USD in annual recurring revenue. Cursor's parent company was valued at 9,2B USD, with Bolt hitting 2,1B. With this data at hand, we can clearly see that these are not just experimental tools with a handful of geeky users.

As we approach the end of May 26, in this exact moment approx 60% of vibe coding users are not developers. So yes, as mentioned in the title, almost anyone can build an app now, but the question that we hear less often is: What kind of app exactly? What happens at the parts the demos never show?
Let's start by giving credit to the sides that deserve it, because the capabilities are impressive in ways that of course matter a lot.

To start with Bolt.new, this platform runs entirely in the browser through WebContainers technology that requires no installation, no local setup and even no terminal. You simply describe an application in plain English or Spanish and it generates and runs the code live. Designers who have never opened a terminal are building prototypes in it right now.

v0 from Vercel had 2 million users generating React components and full landing pages from just text descriptions by the first quarter of this year alone. Lovable targets full stack web application generation with Supabase as the default backend, and has become the mandatory tool for people who want to test an immensely valuable thing as whether an idea is worth pursuing before spending money on a development team.

The productivity numbers for experienced developers using these tools are also real, with some statistics saying that AI coding tools have boosted individual developer output by almost 80% on average, measured by lines of code shipped. GitHub Copilot now has around 2 million paid subscribers and 20 million total users, and approximately 40% of all code written in the world today has become AI generated.

The workflow used by many in the industry is also quite common, with most starting with Bolt or Lovable to prototype fast, and then moving to Cursor or Claude Code for production level refinement. And this is absolutely changing how software gets built today, in ways unimaginable just a few years ago.

All those tools and ways of coding have brought to a way easyer, much quicker and incredibly cheaper reality things like rapid prototyping, idea validation and internal tools. And that is a meaningful democratization and definitely not hype.

Now, having said all of that, we should also explain the part that the product walkthroughs reliably skip, and we should understand that getting from 0 to 90% of an app might be quite easy with vibe coding, while getting from 90% to 100% (handling edge cases, authentication that doesn't have vulnerabilities, payment processing, real deployment, production database design, or error states for every scenario a real user will eventually stumble into) is where things get complicated in ways that simple prompts don't easily resolve.

Karpathy himself, the man who actually invented the term "vibe coding" and is maybe among the most technically capable person you could imagine using these tools, discovered this very early when right after building his own app with vibe coding, he wrote that it was "exhilarating and fun as a local demo but a bit of a painful slog as a deployed, real app". So if Andrej Karpathy himself finds the last 10% a challenging part, it's worth sitting with that for a moment.

The security picture is also worth looking at, and just about a year ago, in May 2025, a study found security vulnerabilities in 170 out of 1.645 apps built with Lovable (apps that real users were actually using and trusting with their data), and some critical security flaws were also identified on Lovable's generated code. These are not random and lonely cases, but are more of a structural consequence of using tools that optimize for getting something working quickly rather than getting something secure reliably, and we should be aware of that.

Several developers who have tested these platforms on stress and at scale have also noted the same pattern, reaching to the conclusion that vibe coded apps work fantastically well for prototypes but "have patterns you will regret at scale" They basically experienced that the code that gets you to a demo often makes your life way harder when you try to grow beyond it. Not because the app is bad, but because it was simply not designed with growth in mind, it was designed to exist.

Very interestingly, a study published mid last year found that while AI coding tools boosted average developer output by 76%, experienced developers using AI assistance were paradoxically 19% slower than when they worked without it (even though those same developers believed they were working 20% faster, according to the study)
The explanation for this is what actually matters, because it was found out that those experienced developers knew when the AI had gotten something essentially wrong. When that happened, they just stopped, backtracked, rethinked the structure and catched the vulnerability before it shiped. And it was that process of oversight and correction that was taking a lot of time, a time that didnt show up as productive in metrics but absolutely shows up in whether the application works correctly in production.

This paradox captures something essential about what expertise actually does in software development, because it clearly shows that it is not primarily about being able to write code, but about knowing when the code is wrong, why it is wrong, what the consequences of that wrongness will be six months from now and how to fix it in a way that doesn't create three new problems. A non technical user prompting Lovable has no mechanism to do that check because he lacks expertise, and when they see something that appears to work they just ship it. The problems arrives later in security audits, in production failures, in scaling walls and in technical debt that accumulates invisibly until it becomes impossible to ignore.

The honest synthesis and the conclusion we can extract from all the above is that vibe coding tools have created a true new capability tier that didn't exist only three years ago. Today, a small team or even a single person with limited technical experience can build and validate a product idea at a speed and cost that previously required a full engineering team, and that matters a lot for founders, for product teams or for internal tooling at companies that can't justify a dedicated developer or a team of developers. But that capability tier has a ceiling, and that ceiling arrives at the moment when the product starts to matter, when real users are depending on it, when security vulnerabilities have real consequences and when the architecture decisions made in the first minutes sprint start constraining everything that follows.

We live in a tech phase where the companies that use vibe coding tools most effectively treat them for just exactly what they are, basically extraordinary tools for speed and validation, but not replacements for engineering basis. Those companies know that the prototype gets built fast, but then it must be the professionals the ones who arrive to evaluate whether the foundation is worth building on, refactor what needs refactoring, harden what needs hardening and design the system that will actually scale.

This is also, frankly, why specialized software development companies have never been more relevant rather than less. The market is now full of beautifully looking products built by excited non technical teams who got to 90% faster than ever before and are now staring at the 10% that requires actual expertise. And the demand for that expertise at the moment, applied to real production systems, is actually higher than it has ever been.

For development firms that know what they're doing, the era of vibe coding is not a threat but a pipeline, and many creative entrepreneurs can today just set up very quick and cheap companies to develop apps at very low costs and relying on experienced subcontractors for the more technical side of their initiatives.

As a brief of the above explained tools, we could brief as follows:

1- For non technical guys or teams out there validating an idea, Lovable and Bolt.new are the clearest starting points. Both run in the browser, require no setup and can produce functional full stack applications from natural language descriptions within minutes. Lovable handles more complex app generation and Bolt prioritizes raw speed and is excellent for proof of concept work.

2- For teams with some technical experience who want AI assistance in a real development environment, then Cursor is the most interesting tool, valued at around 9 billion for reasons that are obvious the first time you use its Composer feature on a complex codebase.

3- GitHub Copilot remains the most widely adopted AI coding tool overall, with so many people around the globe as paid subscribers and integration across every major IDE.

4- For frontend and UI generation specifically, v0 from Vercel produces React components of a quality that impresses even experienced frontend developers.

And for anything that will carry real user data, handle payments, operate at scale or be built to last beyond the initial prototype phase, our strong recommendation is to keep bringing in people who have done it before, because the most clear thing the vibe coding era has clarified is not that developers are becoming obsolete, but that getting something working and getting something right are still two meaningfully different things. One of them is now much faster than it used to be, and the other still takes what it always took. Time and knowledge.

Vibe Coding, la nueva forma de desarrollar software

GPT Images, designing without designers
Translock IT
Luis Yanguas Gomez de la Serna

Almost anyone can build an App in 2026, but here's the part nobody mentions

Singaraja33 — Fri, 22 May 2026 02:42:51 +0000

As a brief of the above explained tools, we could brief as follows:

3- GitHub Copilot remains the most widely adopted AI coding tool overall, with so many people around the globe as paid subscribers and integration across every major IDE.

4- For frontend and UI generation specifically, v0 from Vercel produces React components of a quality that impresses even experienced frontend developers.

Vibe Coding, la nueva forma de desarrollar software

GPT Images, designing without designers

Translock IT
Luis Yanguas Gomez de la Serna

My kids are 2, 6 and 7, and the school they’ll graduate from doesn’t exist yet (and that’s actually good news!)

Singaraja33 — Mon, 18 May 2026 06:44:25 +0000

Here is our article on Medium about EdTech and how education will change in the near future 👇🏻

I have three children. The youngest is just 2yo and currently considers throwing food around a form of creative self expression. The next one just turned recently 6yo and questions almost everything , a habit I’ve tried to keep as I think is fantastic. And then I have a 7yo who is way more reflective and is starting to enter this kind of fun “pre-teenage” stage, with new feelings coming out constantly and a humor that changes from 0 to 100 without clear explanations.

As I guess happens to any of you in a similar situation, I think about their future a lot, not in an anxious way but in the sense that anyone paying attention to what’s happening in education technology right now cannot help but feel that something truly big is coming. And this something is not coming in decades, but in our opinion it will happen along the next five to ten years, literally within the actual school careers of children who are now under ten years old.

As a parent who also watches this change and this space professionally, and who actually enjoys history in general, I figure out things that can make anyone stop and think. To go to a historical example, I think it’s worth sitting with one potentially uncomfortable fact. France’s baccalauréat (still one of the most used university entrance qualifications in the world ) was created by Napoleon back in 1808. This standardized school model that so many western countries run on was designed literally during the Industrial Revolution with a very specific purpose: to produce reliable, consistent workers for factories and administrative roles. It was a system that basically told kids: Sit still. Follow instructions. Reproduce what you were told. Pass the test.

Even in our days, nobody hesitates that his model worked extraordinarily well for over a century, and it is actually right now, in 2026, producing students who are being evaluated for jobs that don’t exist yet, by testing skills that are increasingly automated, using methods designed for a world that ended roughly fifteen years ago.

Even the World Economic Forum has put it plainly quite recently , stating that most secondary education systems are still built to optimize for standardized metrics that reward memorization, individual performance and technical accuracy , which are precisely the skills being automated fastest.

The scary thing is that no single child alive today will spend their working life doing the thing their school system was primarily designed to train them for, and that is the baseline from which everything interesting follows, and a crazy number that stopped me when I first read it is that according to a 2025 Microsoft report, 86% of education organizations worldwide now use generative AI , being the highest adoption rate of any industry. A sector that historically has been one of the slowest to adopt new technology has, in the space of about three years, become the leading industry for AI integration. Insane.

But in any case, it’s also true that the scale at which this is already happening is easy to underestimate. Khan Academy’s AI tutor, Khanmigo, went from roughly 68.000 users in partner school districts in 2024 to more than 700.000 in 2024/25, expanding from 45 to more than 380 district partners in a single academic year. This is definitely not a pilot program scaling tentatively but a technology crossing the adoption threshold and accelerating really fast. Khan Academy ran a strong 6 month testing program from October 2025 to April 2026 measuring Khanmigo’s effectiveness, iterating on outcomes in real time in ways no textbook publisher has ever been able to do.

What makes Khanmigo specially interesting is its design philosophy, because it was actually and deliberately built to never give students direct answers. Instead, his system uses the Socratic method of guiding learners through questions and hints until they reach the answer themselves. This is not a trivial design choice, but is the plain and clear difference between a tool that replaces thinking and a tool that trains it, and it is the model that the most thoughtful edtech companies are converging on.

The global AI in education market sit at approx 7 billion USD in 2025, with not exaggerated projections putting it at close to 136 billion by 2035, growing at a compound annual growth rate of over 34%. To put that another way, we can say that the market is expected to be roughly twenty times its current size within the school careers of children who are starting primary school right now.

Think about what private tutoring does that a classroom of thirty children cannot. It adjusts pace to the individual, it identifies the specific misconception a child has and addresses exactly that, rather than moving the whole class forward regardless. It responds to engagement , noticing when a student’s attention has wandered and changing approach accordingly. It celebrates progress in the specific areas where that student needed to grow, not just the areas the curriculum deemed important this term.

Private tutoring that does these things has historically cost big sums of money per hour. It has been, overwhelmingly, the preserve of families who can afford it and a mechanism that compounds educational inequality rather than reducing it, but AI changes this equation structurally. Khanmigo, for example, costs just 4 USD per month, offering a meaningful fraction of the value of private tutoring at roughly 1% of the price. Scaled across an entire school system, that access gap (which has distorted educational outcomes for generations ) begins to close.

But the deeper change is what becomes possible when personalization moves beyond tutoring into the design of education itself. Today’s most ambitious edtech platforms don’t just adjust pace but instead they identify learning styles. They notice that one child engages more deeply when math problems are framed around football rather than abstract numbers. They track which type of explanation worked and which didn’t, and adjust automatically. They connect topics across disciplines in ways a single teacher managing thirty different children cannot.

The OECD’s Learning Compass 2030 framework describes the direction this is heading. They describe education systems that equip students not just to acquire knowledge, but to think critically, act judiciously and navigate real world challenges with a genuine opinion. Not what to think but HOW to think.

But despite of all of this, and looking at the old system, we should also remember that the things that made great teachers great were never primarily about information delivery. A great teacher noticed when a child was struggling emotionally, not just academically. They created the psychological safety to make mistakes without shame. They modeled intellectual curiosity by being genuinely excited about ideas, and they connected with children as humans first, as students second , creating a relationship that was often what made learning feel safe enough to actually happen.

No AI system is good at empathy. Machines are genuinely bad at navigating messy, unstructured social situations, at inspiring a child who has given up and at reading the room in a way that changes what happens next. Emotional awareness, ethical reasoning or collaborative problem solving all remain distinctly human capabilities, and they are, not coincidentally, exactly the capabilities that will be most economically valuable in a world where AI handles the rest.

Project based learning, which the Harvard Graduate School of Education has shown increases student engagement and produces deeper understanding across disciplines, is not primarily a technology story but is much more of a pedagogy story. A lemonade stand, as one education researcher memorably put it, teaches entrepreneurship, customer service and resilience better than any classroom worksheet. These things need to be protected, amplified and placed at the center of what schools do , not squeezed to the margins as AI handles the academic content more efficiently.

The children who will thrive in 2040 will not be the ones who know the most facts. They will be the ones who can connect disparate ideas, understand and motivate people, navigate uncertainty with confidence and create things that didn’t exist before. The curriculum of the future needs to build for that outcome, and the technology can help get there only if it is used to augment human development rather than replace it.

And from the business perspective, for anyone building or investing in education technology, you all must know that the gap between what is now technically possible and what is actually deployed in schools is enormous, and that gap represents one of the most significant market opportunities of the next decade.

The edtech market is projected to exceed the gigantic figure of 1 trillion by 2030. Gartner projects that 40% of enterprise applications will be integrated with task specific AI agents by the end of 2026 , and education is not exempt from that trajectory. Microsoft committed more than 4 billion to AI education initiatives in July 2025 alone, targeting schools, community colleges and nonprofits through its newly launched “Microsoft Elevate Academy”.

But the companies that will matter most in this space are not the ones that bolt AI onto existing systems and call it innovation. They are the ones that understand pedagogy deeply enough to use AI in service of genuine learning outcomes. The ones that know the difference between a tool that tests comprehension and a tool that builds it, and the ones that can serve both the child in a well funded private school in Madrid and the child in an underserved public school in rural India, because equitable access is not just an ethical imperative, it is simply the market.

Specialized education technology companies (he ones that sit at the intersection of learning science, data engineering and product design ) are positioned to do something genuinely important here. The technology is mature enough to deploy. The institutional willingness to adopt has arrived faster than anyone expected. The regulatory environment is beginning to develop frameworks. What remains is the translation layer, basically the work of turning powerful general purpose AI into education products that produce outcomes that parents, teachers, regulators and most importantly children can trust.

My 2yo daughter is currently just learning that cause and effect is a real thing . She will start school around 2028 and she will likely be in the workforce by around 2045. The jobs she will do in 2045, with big probability do not yet exist, the technology she will use routinely was not imaginable five years ago, the skills that will matter for her (creativity, adaptability, critical thinking, emotional intelligence or the ability to collaborate across cultures and disciplines ) are not well measured by any standardized test currently in widespread use.
The education system that serves her best will be one that uses the most powerful tools available to personalize her learning, free her teachers to do the human things only humans can do and keep genuine curiosity at the center of everything.

That whole system is being built right now. It is not finished and is not evenly distributed, but the pieces are moving faster than most people realize.
And we should realise that parents have always wanted the same thing: an education that sees their child as an individual, not a student number in a big classroom.

And for the first time in the history of mass education, technology is making that actually possible at scale, something that is worth getting excited about.

AI in school bus transportation. The Indonesian case

My kids are 2, 6 and 7, and the school they’ll graduate from doesn’t exist yet (and that’s actually good news!)

Singaraja33 — Mon, 18 May 2026 05:10:40 +0000

Here is our article on Medium about EdTech and how education will change in the near future 👇🏻

And for the first time in the history of mass education, technology is making that actually possible at scale, something that is worth getting excited about.

AI in school bus transportation. The Indonesian case

Seven IT companies now own a third of the SP500, and this is what this actually means for developers

Singaraja33 — Mon, 11 May 2026 05:33:39 +0000

Our article on Medium about the gigantic hype of AI capital on the SP500 and the unbeatable opportunity it brings for developers 👇🏻

If you are at least 45, you will probably remember those hectic times of the “dot.com” boom back in the days, when tech concentration became something nobody could ever have predicted.

As it happened then, today we are again living what is without a doubt the next big IT bubble, with just seven companies ( Apple, Microsoft, Nvidia, Amazon, Alphabet, Meta and Tesla) , now accounting for an insane 33,7% of the entire SP500 as of April 2026.

This is not just one more fact, but instead it’s a massive one, when we realise that one third of the US flagship stock index, the benchmark that pension funds, retirement accounts and institutional investors worldwide track religiously, is controlled by fewer companies than fit around a boardroom table.

That number gets even more surreal when you analyse it and see that only by 2016, those same seven companies represented just 12,5% of the same SP500 index, and that the concentration has nearly tripled in just a decade. The top ten companies combined now account for 40% of the index’s total market cap , and that is almost double what the top ten represented in 2015. That’s not something that happens frequently or even every decade or two…In terms used by financial guys, this is an “all time high” in a dataset going back 155 years. Over a century and a half years.

When we look at all this, the obvious question we normally hear from people with a purely financial brain is whether this looks or not like the dot.com bubble, and maybe the proper answer is that while it clearly looks like it on the surface, it doesn’t look like it where it actually counts.

If we look back to the early 2000’s and the specific case of Cisco Systems, for example, we will see that at the peak of the bubble in those years, Cisco was not less than the most valuable company on earth, briefly worth more than 500 billion USD of the time. It traded at 200 times earnings (yes, not 200 times revenue , 200 times earnings) If you bought Cisco stock at its peak in March 2000, you would have waited until 2019 (a whole nineteen year period) just to do break even.

Pets.com, a company many will remember, raised 82 million USD in its IPO of February 2000 and was completely liquidated by November of that same year. It just took nine months, from start to finish.

The dot.com era was built on a very specific kind of magical thinking where people assumed that any business touching the internet deserved an astronomical valuation regardless of whether it made money or even had a coherent path to making money, but most companies actually didn’t need to make a penny.

Website traffic was treated as a proxy for value, revenue was just optional and profits were almost considered unsophisticated and just as a thing small minded companies worried about before they truly understood the internet.

When all this chapter finally came to an end, the wreckage of that collapsed illusion was historic, the Nasdaq fell 78% from it’s peak, and literally trillions in paper wealth were automatically evaporated.

But today we live a completely different reality, at least in our opinion, basically because this time the fundamentals are genuinely different and a careful comparison will help most see the clear difference between the two historical episodes.

The average 2 year forward price to earnings ratio for the biggest AI companies (like Microsoft, Alphabet, Amazon or Meta) currently sit at around 26 times, while at the dot.com peak, that same metric for the largest tech leaders of the time was nearly 70 times, close to three times more. Also, the Nasdaq is currently trading at about 33 times trailing earnings, versus roughly 60 in March 2000, and while no one can deny that these are elevated valuations in both cases, they also exist in a completely different universe from the numbers that defined the bubble era.

To understand this, maybe one of the most evident facts is that the companies driving the current AI “bubble” are, by almost any measure and as simple as that, among the most profitable enterprises in the history of capitalism.

Companies like Apple, Microsoft, Alphabet, Amazon and Meta (alone) generated a combined 350 billion USD in free cash flow and only in their most recent fiscal years, and even if today we are all overwhelmed by great big figures all around us, just think this for a moment: Three hundred and fifty billion dollars in free cash flow.

To name just another big AI monster, Nvidia alone reported a crazy net income of over 120 billion USD for year 2026, and its data center revenue grew 279% year over year in 2024. And the big difference with the dot.com era is that these figures are not speculative future promises but just seriously audited financial independent statements.

If we look back, during the dot.com boom, the big IT players driving the market were mainly just destroying capital, while today the companies driving the market are among the most capital efficient enterprises ever to exist, with just the average top 10 on the SP500 posting a return on capital of 73% versus 18% at the start of the decade.

The infrastructure spending is also structurally and dramatically different from the dot.com era, and here we see again Microsoft planning to invest approx 80 billion just in AI infrastructure alone, or Alphabet raising this figure to target 85 billion, and Meta spending between 115 and 135 billion (nearly double what it spent the year before)

And the most important aspect to notice here is that this massive infrastructure spending is not something speculative or financed by cheap debt or easy to convince investors, but instead it’s backed by a hugely strong balance sheet funded by existing free cash flow, and that is actually a big difference.

The companies building now AI capacity have among the highest free cash flow and strongest balance sheets in the entire equity market. They are also, and this matters, experts in compute logistics. They have managed enormous data center operations for a decade and they are definitely not guessing or experimenting by basing their decisions on hypothesis.

On the other hand, looking aside all these magnificent numbers and maybe with more conservative glasses, we also think that none of this means the bulls are entirely right, and to honour the truth we should also acknowledge the parts of the argument that still remain unresolved, because if we are to be cautious and look for example at the famous Shiller CAPE ratio (a measure that has predicted past bubbles with reasonable accuracy) we would see that it currently sits around 38 to 40, and the scary thing is maybe that the only time it’s been higher in 155 years of data was precisely at the dot.com peak of 44,19.

Also, SP500 top ten concentration now exceeds dot.com levels by nearly 50%, with a recent MIT study founding that 95% of enterprise AI pilot projects have produced zero meaningful return.

Those are also facts we should not forget, and maybe the most correct way of taking in all this is not just saying "everything is fine”, but understanding that the companies at the center of this new story are truly making gigantic amounts of money, that AI use is only starting to reach enterprise scale, and that the bubble pops not when some ratios are high but when the earnings stop growing, and whether the earnings stop growing is the only question that maybe actually matters.

With regards to us in the developing industry, what is interesting from a software development perspective is probably that the AI race happening right now is, in historical terms, the equivalent of when cloud computing went from an interesting concept to infrastructure that every software team was building on top of.

The companies at the center of that moment (Azure, Google Cloud, etc) didn’t just benefit from the transition, but instead they created an entire ecosystem of opportunity that made millions of other companies and developers wealthier and more productive simultaneously, and that’s the pattern that we think will be repeated and that is more interesting.

The numbers today suggest the developer opportunity is truly enormous in ways that haven’t fully shown up yet, with a global software market that reached 823.92 billion USD in 2025 and is projected to reach the stunning figure of 2.24 trillion by 2034.

The custom software development segment, one that for us is maybe highest in the podium, is expanding at a 22,71% annual growth rate (nearly double the general software market) Also, global IT spending on software is projected to exceed 6 trillion in 2026, the largest spending category increase across all segments.

Many also project that around 40% of enterprise applications will be integrated with task specific AI agents by the end of 2026, up from less than 5% in 2025, with the application software market looking at growing to approx 780 billion by 2030 on that trajectory alone.

The developer adoption numbers are also striking, and if we listen to the JetBrains 2025 Developer Ecosystem Survey (with over 24.000 professionals), 85% of developers now regularly use AI tools for coding and software design. Over 51% of all code committed just to GitHub in early 2026 was either generated or substantially assisted by an AI tool. The AI coding tools market alone reached 12.8 billion in 2026, with a 150% increase with respect to the previous year.

The productivity implication is what actually changes all, and this is what is dictating what a small team can actually build. AI coding assistants can now save tiny teams 3 to 5 hours per developer and per week, according to many researchers, and in practical terms that means a team of five developers today has the productive output of a team that would have required seven or eight people just two years ago.

For small startups, for freelancers or for boutique software agencies, the above is actually a structural competitive advantage that compounds.

So to brief, the valuations of the main AI monsters are indeed elevated, but the critical thing to understand about that statistics is what it implies rather than what it states. It clearly does not mean AI doesn’t work , but instead it means that most enterprises are still figuring out how to deploy it effectively. They are just in an already very profitable experimental phase, and the companies, developers and consultants who can bridge that gap are sitting in front of a demand curve that is only beginning to ramp in order to change this world forever.

We think that this huge AI market concentration is ultimately a reflection of where the people and the market believes the next decade of economic value creation will be concentrated. That belief may be somewhat overstated in valuation terms, but the adoption curve with AI doesn’t look similar under any circumstances to what we saw during the dot.com era.

Cisco at 200 times earnings was a story about hype meeting accounting. Nvidia at 41 times earnings, generating 120 billion USD in net income in a single year is an absolutely different story.

Developers must understand that distinction and simply build the layer of software that sits on top of these platforms . They should not just act as spectators to the biggest capital allocation event of the history of technology. They must be participants.