I am a consultant. My job is to understand my clients’ needs and help them build products to meet those needs. Over the last 3 years, the vast majority of my projects have been applied AI projects.
The truth is that AI projects often struggle to live up to expectations when they hit production. Poor performance, token costs, inverted metrics, and frustrated users are the reality of most AI deployments.
Our industry focuses too much on hype-driven architectures and flashy demos. Most of us ignore the boring part: cost guardrails, evaluations, and design choices that survive the next model release. And above all, knowing when to reach for AI and when not to.
On May 21, Andrei and I are running the second webinar in Jeavio’s series on AI. The focus is running AI workloads in production. We will cover: – Understanding and managing token costs – Mental models for effective AI adoption – Making architecture decisions when underlying model capabilities move so fast – Engineering implications when AI writes most of the code
Link to sign up is in the comments. We will be posting the recording of the first webinar shortly. Please DM me if you have any questions.
A lot of software engineers are quietly miserable right now. They probably include your senior tech lead and your engineering manager.
The issue is the speed and shape of adoption: agentic tools like Claude Code and Codex are changing engineering output faster than engineering culture can absorb. The resulting explosion in output is leading to magnificent metrics but stoking a mania among the engineers on the front lines.
The symptoms are by now familiar. Bloated PRs, scope creep, and engineers relying on AI tools to answer basic questions about how their code works.
I was speaking to an experienced developer today and we half-jokingly discussed running an “AI fast.” A week of no AI use. No coding agents, no email summarization, no AI-enabled exploratory spirals.
Fasting is an ancient way of building physical and spiritual strength through abstinence and self-denial.
We are all committing the cardinal sin of token gluttony, and the results aren’t pretty. Like other forms of indulgence, it may lead to significant long-term health issues for our code bases and our organizations.
Maybe it’s time to consider what a healthier relationship with AI tooling might look like. Starting, perhaps, with a fast.
At Jeavio, we were building AI applications back when Boris Cherny was known for his TypeScript book. We’ve had surprise OpenAI bills before “token-maxxing” was a thing. We had long internal arguments about embedding models, chunking strategies, and hybrid search before RAG became a commodity. So yes… we have scars, and we have stories.
On Thursday, Andrei Tchouvelev and I are turning those scars into a working framework for PE firms making AI investment decisions. He’s also tasked with keeping me from going fully off the deep end.
There is a skit from the early 2000s BBC show Little Britain called “computer says no.” A bank teller responds to a customer trying to get a loan with the same refrain, regardless of the request.
Product managers used to operate inside a similar dynamic when negotiating scope with engineering. Engineering time was a precious resource. Writing code was expensive and time-consuming. So the default answer when discussing new features was often “engineering says no.”
AI-assisted coding has been a series of inversions. This is yet another. Writing code is no longer the bottleneck, which means engineering’s “no” becomes less common and a little less credible. The friction that gave the negotiation its shape is gone. The consequence falls on product management.
“Taste” is the key skill in the AI era. For product managers, discipline is the flip side of taste. The discipline to say no to their clients without the cover of “engineering says no.”
The framing around product decisions shifts from “we should build this, but we can’t” to “we **could** build this, but **should** we?”
Developing a thesis about what a product shouldn’t be becomes as important as the one about what the product should be.
Experienced product managers already do this. But they were brought up in an age of scarcity. They knew what to suggest to engineering teams because they had hard-won mental models of capability, capacity, and constraints. Product managers coming up in the AI age face an era of plenty. They will need to develop restraint instead.
Product management used to be an additive discipline: more requirements, expanded stories, negotiating for more resources. Now it becomes subtractive.
AI has the potential to transform businesses, but figuring out where to invest and how to integrate AI can be challenging. On May 7, Andrei and I will be talking about Jeavio’s experience building AI applications over the last 3 years. We will discuss how to evaluate opportunities, identify pitfalls and set up your AI initiatives for success.
Over the last couple of years I’ve run into a curious adversary. A doppelgänger. He writes clean, grammatically correct, structurally sound articles. He claims to be me, but I do not recognize myself in what I read.
My LinkedIn feed has turned into an infinite scroll of doppelgänger-generated content. The rough edges of writing have been sanded down into pithy phrases, listicles, and corrective antithesis.
It’s not only not me, it’s not you either.
It’s tempting to let the doppelgänger do his thing. Write the smooth prose and hit publish. What starts off as a distorted reflection morphs as you publish more. The cadence, beats, arguments, and calls to action start to sound like everything else on your feed. From a digital twin to a clone army.
The doppelgänger is the product of billions of dollars and millions of hours spent teaching LLMs to sound helpful, polished, and sane. The result is writing that feels safe, agreeable, and eerily familiar. Writing with all the now notorious hallmarks of AI-generated content.
Retreat is an option. Step away from the keyboard. Leave LinkedIn to the listicles. Buy an artisanal notebook and a fountain pen. Retire to the countryside to write poetry.
But there’s another way.
It’s time to put the doppelgänger to work. The AI tools have options. Claude and OpenAI have skills. Gemini has Gems. xAI may have an anime character or something.
I built a writing workflow with one goal: take back control. I start with a brain dump, usually a long voice note. Then three skills. editing-prose pressure-tests the argument and cleans up bad grammar (English isn’t my first language, it’s just how I hallucinate). my-voice keeps the doppelgänger in his place. crayon-simple makes sure the core idea comes through and doesn’t get lost in asides and in-jokes. Then I rewrite, line by line.
The doppelgänger doesn’t go away. Sometimes I catch him. Sometimes he sneaks in an em dash – he’s sneaky. I just try and make sure he doesn’t get the final word.
It doesn’t matter whether you are Anthropic, OpenAI, Google, or xAI. Every frontier lab and hyperscaler is now committed to spending billions on infrastructure, data, and talent to build and scale large language models. This gamble is propping up the US economy. It is also being sold as the path to solving the world’s hardest problems. No matter where these labs started, they are now on the same road, and they are taking us with them.
I’ve been thinking about this while reading Sebastian Mallaby’s The Infinity Machine, his biography of Demis Hassabis and history of DeepMind. The book is also a snapshot of the current AI moment.
Hassabis is an extraordinary figure: chess prodigy, game developer, neuroscientist, now head of Google DeepMind. AlphaGo beat the world’s best Go player. AlphaFold solved protein folding and won him a Nobel Prize. Mallaby paints a sympathetic picture. Hassabis sees himself as Turing’s champion – using inductive reasoning to solve the world’s hardest problems.
And yet Hassabis and his peers are all running the same race. Altman, Amodei, Musk, each with his own higher calling: saving humanity, transcending the body, understanding the universe. In practice they are all spending billions to replicate each other’s work. The race has a winner-take-all logic, and that logic allows only one strategy: get there first.
So we get Dario Amodei warning about the destruction of white-collar work while Anthropic ships Claude Design, a tool aimed straight at automating design work. We get Hassabis talking about hadron colliders in space while pushing Gemini to catch OpenAI.
This is what path dependence looks like. Mallaby’s book shows that even a figure as sympathetic as Hassabis has a pragmatic, competitive side that will do what it takes to win.
I find Hassabis genuinely inspiring, and that is what makes the book unsettling. If the most thoughtful figure at the frontier cannot escape the spiral, then the spiral is the story, not the people inside it.
If you want to understand despair, try building a book tracking app and then searching the App Store to see what the competition looks like. Hint: it’s brutal out there.
My book tracker is called QuietReads. It has an AI assistant, a nifty note taking feature, and I get a lot of value out of it. But let’s be honest: QuietReads is not going to fund my retirement.
The reason is simple. It’s easy to understand what a book tracker does. You search for books, track what you are reading, maybe have some note taking functionality. Anyone can build a fairly comprehensive book tracker with some rudimentary knowledge and a weekend of token-maxxing with Claude Code.
AI has removed the friction between ideas and code and that has resulted in an absolute explosion of apps in crowded domains like, well, book tracking.
Pondering my predicament made me think of an article I read about Indian leopards. Habitat loss and competition with humans has led leopards in marginal areas of India to become more nocturnal and more specialized. More opportunistic. They survive by knowing their specific territory so intimately that no newcomer can navigate it.
Maybe there is a lesson here. AI has turned the software ecosystem into a dark forest. Brutal competition. If you expose yourself too much, you get copied. Features replicated, vulnerabilities exploited as soon as the are exposed.
The leopard has an answer: adaptation and specialization. AI can copy what your product does. It struggles with why you built it that way: the customer workflows you spent years understanding, the edge cases that only surface at scale, the tradeoffs you made because you know this domain from the inside. It may be knowing that using floating point numbers for money is a terrible idea, or why user engagement may be a dangerous metric for an AI app. It’s the scars that earn their keep.
That knowledge is your ecological niche. And AI doesn’t erode it. It lets you compound your advantages. Domain experts who use AI to build move faster inside a territory only they can navigate.
The time for the generalist is ending. The specialists are taking over. QuietReads may not make me rich, but I have enough scars that I don’t mind navigating the dark forest.
My kids are three and five. I have been in this industry for twenty two years. You would think that somewhere along the way I would have put together a coherent mental model of how the technology industry works, and make sense of what is happening. Something that I could wrap up as wisdom, or advice, or maybe just a pointer as my kids think about their future in a few years time.
Instead, I have been reacting. Lurching from one cycle to another. Mobile, Cloud, Crypto, AI and whatever comes next. As I navigate each cycle, like a befuddled tourist, the scale gets bigger. Each time, it seems that fewer people are thinking and the explanations for what is happening and why are more confused.
So I have been reading. Postman. Karen Hao’s Empire of AI on Sam Altman and OpenAI. Sebastian Mallaby’s The Infinity Machine on Demis Hassabis and DeepMind. Kyla Scanlon on the strange economics of the moment. And I have been trying to put together a through-line. The outline of a semi-coherent narrative that gestures towards why we end up here – arguing about data centers, fretting about AI, as the world literally burns.
I think it starts with a banner ad.
October 27, 1994
Bill Clinton was President, and Marc Andreessen had just released the Mosaic web browser just a year earlier. On that date, Wired Magazine’s digital spinoff, HotWired, launched its website with about a dozen paid advertisements. One of them was for AT&T. It was 476 by 56 pixels. It asked: “Have you ever clicked your mouse right HERE? You will.” Clicking it took you on a virtual tour of the world’s great museums.
The world’s first banner ad
By today’s standards, it was quaint. I suspect it was quaint even by the standards of 1994. But it was the beginning of something that would birth an enormous range of products and services: from social media to AI slop. But at the time, nobody involved thought they were the catalyst to the creation of a giant industry.
Within a year of HotWired’s launch, it was followed by Lycos, Excite, InfoSeek, and others – each company created with advertising as their primary business model. By the time Google created AdWords in 2000, the model was established. By 2014, Google alone was generating $60 billion a year in ad revenue. In 2025, Google brought in nearly $295 billion.
Nobody planned this. Andrew Anker, the former investment banker who wrote HotWired’s business plan, settled on advertising because it was, as an oral history of the era describes the only logical revenue stream he could envision. So, the humble beginnings of the business model that today underpins Meta,Google, and one that upstarts like OpenAI are staking their future on. An incidental decision, that became possibly the most profitable business model ever invented.
The same decision was also the inflection point from where technology went from being the domain of a handful of nerds and dreamers in a sleepy suburb of San Jose to the dominating social and economic force it is today. It is also where the trouble starts.
The Spigot Opens
Digital advertising became the canonical business model as the world went online. Technology meant zero marginal costs, global reach, and rapid dissemination. It generated staggering amounts of surplus capital concentrated in a handful of companies. Google, Facebook, Yahoo before them and many others to follow. The venture capitalists who were early investors in these companies generated mind boggling returns. Serving targeted ads at scale became a money printer.
All that surplus capital needed somewhere to go.
Capital chases returns. When you have that much money looking for a home, the bar for “this might work” drops dramatically. The entire Venture Capital business model is predicated on finding that one unicorn that would generate massive returns. You can see multi-billion dollar bets on businesses leasing office space for a loss while wearing the accoutrements of a technology company were justified.
For a VC, sitting on uninvested funds means your fund underperforms, your investors look to greener pastures. So you fund whatever has the ceremonial language of disruption. You fund NFTs, you fund fancy juice makers, because the alternative would be irrelevance.
The Juicero (RIP) – disrupting the fruit squeezing industry
The post-pandemic crypto boom seems like a fever dream now. Billions raised on the promise that pixelated pictures of digital apes would one day sublimate into collectible and unique art. Investors poured money into automated money-making machines based on complicated DeFi (Decentralized Finance) protocols that required a PhD in applied mathematics and the willingness to suspend disbelief.
I wasn’t immune. I spent way too much time trying to understand how Bitcoin works (and wrote a terrible science fiction story trying). I spent my hard-earned money on staking out some space on the blockchain. Money that could have gone to my kids’ 529 plans. Sorry girls. I wasn’t the outlier though. People way smarter than me poured millions into something that was, when you stripped away the jargon and the vibes, the world’s slowest and most expensive database.
How did this happen? How did so much capital get allocated (and is still allocated!) on something that is so obviously broken?
It is downstream of the spigot. There is so much money in the system, generated by that initial accident of online advertising, that it has to go somewhere. And the job of any ambitious entrepreneur is to provide it a plausible channel. Once the money starts flowing, it becomes self-reinforcing. More money validates the narrative. The narrative attracts more believers. More believers attract more capital. FOMO reigns supreme. The cycle runs until something breaks the spell.
Technopoly
Neil Postman wrote Technopoly in 1992, before the Internet was really a thing. I finished it a few weeks ago and it’s been rattling around in my head ever since.
Postman, a famously skeptical and uncompromising media critic, defined a technopoly as a society that has surrendered its decision-making to technology. Not in a “I welcome our robot overlords” way, but through a blind submission to metrics and statistics of dubious value. His examples – IQ tests, a man “drowning in a river that is, on average, four feet deep”. A society where technology becomes dominant through a self-perpetuating loop where it drives investment in itself, regardless of the impact on human-well being. Does this sound familiar?
Postman didn’t believe resistance was futile
I am also reading Sebastian Mallaby’s The Infinity machine. What comes through in his account of Demis Hassabis, Elon Musk, Larry Page and other industry titans is just how each of them operates with a profoundly different view on Artificial Intelligence – set to be the dominant technology of the 2020s and beyond. Hassabis comes across almost monk-like – viewing AI as the means to understand the nature of existence. Page as a transhumanist who sees the body as a shell to be discarded and the machine as the vessel for eternal life. Musk as a self-appointed guardian of a very particular vision of humanity – presumably with him as some sort of God-Emperor. These are the people who are deciding how and where the most consequential technology in a generation is deployed.
Consider the data center buildout. Trillions of dollars of private capital are being deployed to construct AI infrastructure at a pace that makes the railroad boom look modest. Even adjusted for inflation.
The rationale behind the investment seems to follow a circular logic. AI requires compute for training and inference, so you build data centers and fill them with compute. You fill them with compute which makes rapid deployment of AI possible and sparks massive competition between foundation model companies and hyperscalers to build, fundamentally, the same things. That in turn drives further demand for compute, and so on.
And AI is not just another speculative bet sitting alongside the economy. In 2026, it is the economy. Kyla Scanlon put it plainly: as AI swallows more and more capital, it has become the stock market and the economy simultaneously. The same companies – Microsoft, Google, Amazon, NVIDIA, etc. bankrolling the AI infrastructure buildout are also its biggest customers. Data center construction and investment in AI is driving GDP growth without growing jobs. Scanlon calls it a “jobless expansion“.
The prosperity exists in balance sheets and in the giant data centers sprouting up along state highways in places like Northern Virginia and Tennessee. It does not exist in communities that have been hollowed out as manufacturing evaporated and a career meant becoming the meat interface of a faceless algorithm directing you to the next gig.
Postman would have recognized this instantly. Nobody is in the driver’s seat. No single actor decided this was a good idea for society. No democratic process approved it. The technology cycle itself is driving societal change without any significant discussion. We have abdicated to the technology itself.
I wrote this in my notes on the book: “FOMO-driven investment in AI data centers with little to no prospect of broad societal benefits. We could have spent this money on climate change remediation or education or universal healthcare but here we are.”
Here we are.
The Arbitrage Trap
So there is a clear sense of a backlash brewing against technology. Try posting a pro-AI take on Threads or Bluesky and see the reaction. It’s not just the keyboard warriors who are ready to fight. A young man threw a molotov cocktail at OpenAI CEO Sam Altman’s house last weekend. Politicians who oppose moratoriums on data center constructions are threatened. There is a sense of rage as a new generation emerges into the post-ChatGPT landscape of disappearing knowledge work and the deep sense of careers and meaning being stolen by algorithms and trillion-parameter models.
But there is a key point that is missing in the public discourse. When Bernie Sanders “debates” Claude, it isn’t a meeting of equals. Bernie is an influential Senator and can call for moratoriums, hearings, and regulation. But Anthropic just raised billions of dollars in private capital. Claude runs not on the public dime, but on VC dollars. If Sanders pushed for a ban on data centers in Vermont, they will just build them in North Dakota. It doesn’t really matter where the data center sits. If the capacity is built out, compute will be deployed, and that compute will reshape the day-to-day work of people whether they want it or not.
Hope Bernie has a Claude Max plan
Jasmine Sun, who writes one of the sharpest Substacks on AI and Silicon Valley culture, spent time in DC and San Francisco earlier this year tracking what she calls “AI populism.” Her observation is stark: there is a widening chasm between the people who are building and funding this technology and the people who will live with its consequences. Politicians gearing up for the 2026 midterms are scrambling to design their AI agendas. Labor unions, environmentalists, social conservatives are all rushing to come up with a position.
But the backlash lacks a mechanism to break the capital flow. Moratoriums work only if they are coordinated and enforced. Otherwise you just get regulatory arbitrage. Capital exits to friendlier jurisdictions, and the places that resisted end up with neither the investment nor the jobs. That is the trap.
And if you want to see the arbitrage logic taken to its absurd conclusion, look no further than Elon Musk’s push for building data centers in space. As they say, in space, no one can hear your strident demands for a datacenter moratorium.
Spending a trillion dollars to deploy millions of GPUs in space seems insane, but it is also internally consistent with the incentive structure.
The Ratchet
The AI investment cycle can be best described as a ratchet. It moves only in one direction before locking in place. We seem to be committed to seeing where this cycle plays out – even if it ends in tears.
Sun points out the worst case scenario in a recent post – “One nightmare is a future where we get AI that’s good enough to wreak social and economic havoc, but not yet good enough to cure cancer / solve climate change / deliver 10% GDP growth. In that world… who pays?”
I work with AI every day. I run teams that build with it. I can see the utility. I have watched it compress weeks of work into hours and deliver real value for my clients. I am no old man shaking my fist at the clouds. But I also know that when the backlash arrives in full force, when the torches are lit and the pitchforks come out, the distinction between “I used AI thoughtfully” and “I profited from AI” will not matter. I worked in investment banking in 2008, I know what it means to be a social pariah.
People like me, who have bet careers on this technology being useful, will be caught in the same sweep as the people who bet billions on it being transformative. Everyone on the ratchet moves in the same direction. And yet, apart from the odd paper, there are little to no concrete suggestions from the same billionaires about how to make technology work for everyone apart from some vague gesturing to super-intelligence and to abundance.
And what feels like willful delusion rather than mere miscalculation is that the ratchet keeps clicking forward even as the world around it deteriorates. The United States, Israel, and Iran are in an active military conflict with direct strikes and counter-strikes. The Strait of Hormuz, through which a quarter of the world’s traded oil passes, is under threat. Russia’s invasion of Ukraine grinds on. These are the kinds of events that should be sending capital fleeing to safety. Instead, the markets shrug and carry just .. carry on?
Trying and Failing to Understand the World
I started this year realizing that I did not have a coherent mental model to explain what was happening. I read Postman, Hao, Sun and Scanlon to try and see how others made sense of a world so utterly dominated by technology and the eccentric billionaires who control it.
I ended up with a series of explanations that seem to involve some sort of hardware. Spigots, ratchets, and data centers in space. But I do not think I have a mental model. I do not think I can predict what comes next except a vague feeling that we will continue to spin faster until the whole edifice comes crashing down or we ascend to the singularity.
Postman’s views on the subservience of humanity to technology appeal to me because they seem to be manifest everywhere I look. From people scrolling aimlessly on their phones to pouring out their darkest secrets and deepest fears into the maw of a trillion-parameter language model. It is grim stuff.
But Postman also offers a solution. He calls it a “thoughtful rebellion.” Maybe the movement to touch grass, the surging sales of physical books and vinyl are signs that there is a genuine desire to disengage from digital technology. But how much of that movement is itself driven by mimetic desires pushed by algorithms, through BookTok and the like?
Perhaps technology provides its own means of meaningful disengagement. And maybe that is the only, if unsatisfactory, answer.
My kids will inherit a world shaped by decisions nobody consciously made, funded by a torrent of money nobody voted to spend, run on infrastructure nobody asked for. The best I can do is to show them the machinery that drives the world. And hope that they can find a way.
I run a personal agent called Saarthi. It is built on OpenClaw, an open-source agent framework. I have configured Saarthi around my own information processing workflows: research synthesis, content management, and journaling. OpenClaw comes with a set of underlying primitives such as scheduling, memory management, file access and so on. I assembled these primitives to help with my use cases.
OpenClaw uses a similar architecture to what powers Claude Code and Codex: tool calling, persistent context, planning loops, sub-agent spawning. These are the capabilities that turn a frontier model into something that can do useful work.
But Claude Code, Codex, and OpenClaw are all general-purpose tools. They are powerful because the architecture is well engineered. They are limited because they are general-purpose. They can call tools and manage context, but they can’t tell you which workflow to automate or where an agent creates more value than it consumes in review overhead.
That limitation is getting easier to solve. LangChain’s Deep Agents library ships the same kind of primitives as open building blocks on top of LangGraph. The scaffolding that makes Claude Code effective can be assembled by any developer.
The next wave of useful agents won’t come from better frontier models and general-purpose harnesses. They will come from the people who understand a domain deeply enough to know exactly where agentic capabilities can help, what workflows to target, which edge cases matter, and where the real complexity lies.
If you’re a senior engineer who has spent years building that understanding, Claude Code isn’t going to replace you. It is the demo. Deep Agents and frameworks like it are the toolkit. Your knowledge of the systems, the failure modes, and the corner cases: that’s the part nobody else can supply.
