Beyond hyper specific stories, an AI will never experience anything as weird as my experience buying tires that helps shape new creativity. And it will also never experience more universal experiences like the rejection of love, it won’t know the rush of being late for a flight, or having a bad meal and the subsequent consequences of said meal. The human experience is basically impossible to replicate in code.

It’s those life moments that shape how we perceive everything else, and where creativity spawns from. No machine will ever experience the stress of taking young children to Target to only have them throw a tantrum on the floor. Or have the concern that they might fill their diaper while at a store without changing area. But really AI will never know the true stress of when the consequences of that bad meal hit while you’re at Target and now you have to sit on a toilet in a bathroom that wasn’t cleaned in the last 24 hours while holding your baby1.

These experiences shape everything in our lives, especially creatively. You can try to prompt an AI model to tell this story but it will never feel the panicked stress-sweat of these chaotic experiences. I’ll concede that there’s a possibility that a model has absorbed a thousand stories about children having meltdowns in stores. However, this precisely highlights the issue: it provides an average of a thousand tantrums, and let’s be honest, the average tantrum never actually occurred to anyone. Every parent has a story. The specific scenario, where myself, a baby, and a bad meal converged in a disgusting bathroom (that hadn’t been cleaned in at least a day), only exists because it was real. It’s lived in; the whole life is stranger than fiction type of thing.

As we are fighting against a wave of AI slop, there is something that is being missed in all of this. The creative output of the human experience will always outdo anything AI will provide. I’m not saying AI cannot be entertaining but it will never carry the weight of a lived experience. I think that’s what is missing from all the handwringing, you can be entertained but it doesn’t have to replace the great American novel.

We observe this phenomenon in social media and short-form videos. These are all deceptive and manipulative, irrespective of AI involvement. Their sole objective is to generate engagement rather than fostering creativity, which many find entertaining too. Although, on most social media, even when human-made content is present, we are often prioritizing falsehoods over originality. Is this any different that pure AI? To me the distinction between human and machine content seems to be getting more blurry by the day. I’m willing to take the perspective that the broader implication is that there may be a future for both types of content. The outcome will depend on your preferences and the content you enjoy.

Beyond the output, for me, the more interesting question is the future of creativity and that recognition is now nebulous at best. I believe its fair to say that a label of human or AI won’t rectify poor writing, but it may diminish actual good writing. For me, the byline isn’t a reliable indicator anymore, but rather whether the work emanates that unmistakable scent of stress, an unclean bathroom, and an occasional bad meal.

I’ve taken the leap into the modern middle age behavior of let’s test, track, and develop more anxiety1. I have always had a complicated relationship with tracking my health. When I was more focused on athletic performance, ironically I tracked the least amount. I would log my workouts and maybe my food if I felt like something was off but that often ended up like a Trump speech, full of lies, half-truths, and an accusation of others that was really my fault2. But as I made less time for myself, I would go through ups-and-downs of health tracking. Steps, calories in, calories out, heart rate, oxygen levels, sleep scores, blood sugar—the list goes on.

I’m not here to debate the effectiveness of this tracking, or even whether I’d recommend it to someone. My concern is bigger: if this data is being collected, who gets to access it? The natural place for this data processing is AI and the feedback on your results is potentially a worse privacy nightmare than the collection itself. A step count is just trivia. ‘This person is probably pre-diabetic’ is a product. In other words, raw numbers are just numbers until something interprets them. The feedback is the interpretation, and an inference about your health, that you might be anemic, pre-diabetic, or pregnant. All of which is far more valuable to an advertiser or insurer than the heart rate readings it came from. Technically speaking, your data might have been encrypted in transit and at rest, but now it needs to be processed, and the breakpoint for that data is AI. You can’t pass encrypted data to an LLM, which means the model owner can now see it. Do you trust these big LLM companies not to misuse this information? Do you trust them not to leak it?

I get why people go down this path anyway. It doesn’t always feel like there’s a right space for you to get help, so it’s understandable that you’d consider a system that both analyzes your results and gives you feedback. It seems appealing, especially in an age where it feels like doctors don’t care about you. You’re just an insurance billing number. These systems give you a chance to actually converse about your symptoms and hunt for a differential diagnoses. But those are traps, because again, where does this data go? What’s feeding these systems to produce that differential diagnosis? And are you sure you can process what comes back? Have you ever looked up abdominal pain? Google will hand you eight thousand possibilities, and four thousand of them sound scary and horrible.

An AI won’t give you all of those possibilities but it will convince you that it knows best. AIs are very sure of themselves. They would rather say something than admit “I don’t know” or “nothing.” So which anxiety is worse, 4000 potentially scary outcomes or 1 assertive outcome that could be wrong? You cannot always know if the model is returning accurate information. Now you’re in a weird spot: you gave up your data to be analyzed and you still have no clear answers. To be fair, you gave up your data when you searched Google too. (Why do you think you’re being served ads for iron supplements? Google thinks you’re anemic.) All of this tends to produce more anxiety because it never actually addressed the issue you had. You still need to see a doctor. The secondary effects of data sharing have not even begun to be understood. The long-term implications on whether your personal data can and will be used against you are in the near future.

Hopefully, anytime a new system gets built, it will actually be better at supporting you, but importantly, consider your privacy first. AI as a tool, AI as an augment, is wonderful. AI as a replacement is scary and anxiety-inducing, and it upsets people for good reason. I want systems to support you and protect you in the same breath, not as a settings toggle you find later. Until then, protect your information and choose carefully where you send it. The peace of mind is real, even if the abdominal pain is too. It's probably just gas.

From the conversations I am having with people regarding AI and privacy, the agent conversations frequently emerges as a primary concern. It remembers too much, knows your name, your dream vacation, your peculiar lump on your shoulder, and there’s a good chance it might repeat something you typed at two in the morning. AI companies have your data, either obtained through data collection or purchased from third parties, and they have no intention of informing you about it.

Here is the latest example: SpaceX acquired Cursor in June for $60 billion in stock, integrating one of the most widely used AI coding tools into the same corporate structure as xAI and Grok. Two months later, it serves as a case study in how AI tool privacy is actually eroded not through a system breach but rather through default settings and a breach of trust.

The mechanism behind this is that Cursor now generates a Grok subagent to explore your codebase before your chosen model interacts with it, regardless of the model you select. For instance, if you opt for Fable 5 for the actual work, Grok will read the entire repository first and send it back to xAI’s servers. To disable this feature, you need to locate a setting that you might not have been aware of. Cloud indexing embeds your files to their servers unless you use the appropriate ignore file.

Perhaps this is a careless design decision. However, the company receives that benefit of the doubt only once, and Grok has already squandered it. Reports earlier this year revealed that Grok Build uploaded entire repositories to its servers against user settings, including data from outside the repositories. A fix and a deletion claim were only made after users discovered the issue. When the same organization then ships three separate default settings that all leak in the same direction, the pattern becomes evident.

Regrettably, the situation with these things is consistently the same. When you installed Cursor, you agreed to the terms, even though it was an independent company with its own incentives back then. Then, the company was acquired, the incentives changed, the default settings changed, and your consent was treated as if it belonged to the product instead of to you. Nobody bothered to re-ask. The checkbox you clicked at signup covered a use case that was invented later by a different owner, with a model you never chose, reading files you thought you had excluded. That’s not consent; it’s a receipt from a transaction whose terms were rewritten after you left the store.

The ownership seems to be more significant than any settings page. When you interact with an AI tool, you’re not really trusting the tool itself. Now, you’re trusting the corporate structure behind it, and that structure can and will be monetized, with quarterly reports. The larger industry has been providing the “Responsible AI” page, which is garbage. I’ve stopped (or never started) reading these with any sense of seriousness. They serve nothing more than an ethics statement outlines the intentions at the time of writing, which are valid only for that moment. And by moment, I mean the lawyer who was tasked with that job.

Intentions and well-wishes and promises mean nothing since they are often forgotten in the name of “progress” and capitalism. If your tool exfiltrates code against a user’s expressed settings, it should cost you regardless of whether the cause was malice or a subdirectory toggle. Yet, no one is liable and this will just continue. This is another example of how wild opt-out culture is and nothing really will change until we stop using these tools or demand better user controls.

For me, its simple, if prompts and repositories feed training, users should be able to find out about it and refuse it without having to dig through the settings menu (or at least not have to learn about it on the social network owned by the same company). Yet, none of that exists today, any and all tactics are being used to race to the top of benchmarks. So the only sane practical advice is ugly. Treat every AI tool as an adversary until it proves otherwise and audit settings after every update and especially after every acquisition. Even assume the ignore file does less than its name implies. Except you’re not likely going to do any of that. I probably won’t either because as humans we are lazy about privacy until we realize it is directly affecting us or its too late.

The lesson from all of this so far is that an agent chatbot remembering your name (and that weird spot on your foot at 2 am) was never the only threat. The threat is also that $60 billion acquisition quietly rewriting what happens to everything you type, and a settings page designed so you find out last, if at all.

P.S. - I shut off the AI detection service as a test. Do you think I wrote this or an AI? Does it change how you feel about it?

P.P.S. - I feel like I don’t see a lot of P.S. in writing anymore. Also I learned today that whether you use periods or not in P.S. is a British v. American English thing.

Talking about privacy is free. Building it costs you decisions you can’t take back later, which is why so few products do it and why the ones that wait end up with something that only looks private.

I learned how expensive those decisions are from one bug. A while back I sent my personal chat agent, Microwave, a health question. It made the right decision. It recognized the question as medical and routed it through the secure channel, the restricted path where nothing identifying information is transmitted from the device. Subsequently, I responded with, “It’s not me; I do have a picture, though,” and the entire process unraveled.

The router evaluated each message independently. Initially, the message appeared to be medical, while the second one seemed like a man sending a picture. The router had no idea that they were the same conversation, so the privacy decision made in the first message silently expired in the second. Asking an AI to manage privacy is like trusting a friend who excels at keeping secrets for a limited time, such as up to three drinks.

I delved deeper into that thread and stumbled upon a product. After a few months of using Microwave, I discovered which feature I used the most. Among its various capabilities, reading my files, monitoring my cameras, and running errands, the one I utilized the most was health. This wasn’t because the health feature was superior. Instead, I comprehended the precise level of privacy it offered, and that understanding influenced my willingness to request it.

That’s the product. For a tool that directly impacts my daily life, “take it on faith” is insufficient. It turns out that my behavior changes the moment faith ceases to be a prerequisite.

The Sealed Room and the Importance of Private AI

Nearly every health app on the market operates on a similar principle. You provide your data, which is then sent to their servers and becomes their property. This isn’t malicious behavior; it’s simply the business model. Your data is the asset, and you’re essentially renting access to your own life. If you stop paying, your account will be terminated, and your records will remain with the company. Even with a privacy policy there is a lot of opportunities for bad behavior with your data.

So for this product, I wanted a different approach. I wanted everything that could remain on the device to do so. The user should be in-control of the data. When a question genuinely requires more than the local phone data can handle, the system carefully steps outside and utilizes the most secure path available to handle sensitive information.

Additionally, by have the app log every routing decision, it allows me to audit its actions. This is because after the sticky-routing bug incident, I stopped accepting the excuse of “it probably did the right thing” from my own software. As I wrote previously, AI models are made to lie to you, so to assume they will be honest about the routing is foolish.

That bug incident established a design principle for everything that followed: when the convenient path and the safe path diverge, prioritize the safe path. Then, make the safe path feel like the convenient one. Users should never be forced to choose between privacy and a good experience, as they’ll inevitably lose that battle. I understand this, as I was the user who initiated the follow-up.

The challenge lies in the moment when a question surpasses the capabilities of a phone. Certain questions require a powerful model in the cloud, which is precisely when your private information is traditionally logged by someone’s server.

The analogy I used earlier remains relevant. Imagine sliding your question through a slot into a soundproof room where a machine responds to it. No one, including the building’s owner, can access the room or listen in. This is a trusted execution environment. What sets it apart from a mere metaphor is attestation: the hardware cryptographically verifies that the room was sealed before you inserted anything. “Trust us, we won’t look” transforms into “we can’t look, and here’s the proof.”

My sealed room operates on NEAR AI. I didn’t select it solely based on a comparison chart. I spent months thoroughly testing their agent, IronClaw. By the time I needed a confidential compute layer for a real product, I had invested enough time in this one to trust it with my own health questions. That’s my benchmark. I’m a user zero.

There are two primary functionalities. Everyday questions are routed through NEAR Cloud, where identifying information is removed before reaching a large provider. The upstream provider receives a de-identified question, not “Joe from account X asked this.” Sensitive questions are processed through open-weight models running inside NEAR’s TEEs, which is essentially a soundproof room where the host running the hardware couldn’t peek if they wanted to.

Here’s the part that I find genuinely amusing. Large health platforms can’t replicate this, and the reason has nothing to do with engineering talent. Their revenue depends on seeing your data. You can’t sell privacy as a feature when visibility is the business. Any of them could build a sealed room tomorrow, but none of them can afford the financial implications.

A product built from the outset with confidential compute never encounters this conflict because it was never designed to see anything. Privacy isn’t a policy I’m committed to upholding; it’s an inherent property of the architecture, and the architecture will demonstrate this through the math. Policies can change when companies are acquired, but properties remain constant.

Another point I frequently return to is that the person this protects doesn’t necessarily need to be aware of it. Most users will never check an attestation or read a routing log. They’ll simply notice, as I did with Microwave, that there’s one tool they keep reaching for with the questions they wouldn’t type anywhere else. They won’t be able to explain why. The architecture will know.

This could have been a genuine career fork. Somewhere out there, there’s an alternate version of me who has a podcast (maybe 2), a supplement code, and a morning routine that involves waking up at an hour I associate with medical emergencies. Bizzaro Joe is thriving. His engagement metrics are impressive. He even paid cash for the Porsche1.

I think about him, and then I remember why he can’t exist, why I cannot enter the parking lot of the complex. Simply put, the job has requirements I can’t force myself to live up to.

If you want to have a chance at all at succeeding then the initial load-bearing requirement of self-help content is confidence. “The protocol works.” “The framework changed my life.” “Do these five things and your cortisol will apologize to you personally.” I’m exhausted just joking about it.

If I summon my training, here is what I would say: “the effect exists in some populations”, “the effect size is modest and it may not survive replication”, and this banger, “the intervention’s success probably depends on eleven contextual variables we didn’t measure.” Now we are selling courses, try putting that on your next video thumbnail. Yet the honest version of most behavioral science findings reads like a pharmaceutical disclaimer, and nobody has ever built an audience on “results were mixed.”1

The subsequent requirement is anti-stagnation. The genre cannot remain stagnant because content necessitates novelty, but there is a small issue with this, human behavior has remained largely unchanged since its inception. Therefore, something must constantly evolve, which, in turn, leads to an escalating (poor) advice. Cold showers transformed into ice baths, which then became cold plunge tubs that exceeds the cost of my first car. Adequate sleep transitioned into sleep hygiene, and subsequently, it became a wearable device that evaluates your sleep quality and sends you a report via email, allowing you to commence each day with a performance review from your own mattress. The hypocrite in me does track my sleep but I will use the excuse that it is primarily to enhance my overall recovery. It is something important to my overall well-being. With that, I certainly do not offer any advice on improving sleep quality. At best, I can suggest adopting a mindset that anticipates perpetual fatigue when raising young children.

However, my favorite development in this space is the dopamine detox. Shut off those pleasure centers. If you break it down, the dopamine detox is a simple concept that has become an purposely overly complicated and obtuse concept. It suggests that people should abstain from pleasure to reset a neurotransmitter system that operates quite differently than what “gurus” would have you believe. It is comparable to unplugging your router to address plumbing issues. This approach gained widespread popularity because it possesses two qualities that the market values: it appears to be based on neuroscience and allows individuals to feel superior to those who indulged in their weekend activities.

The endpoint of this escalation is the full protocol lifestyle, where a day stops being a day and becomes a stack of blocks. This is the nirvana for self-help influencers and gurus alike. Sunlight exposure window. Zone 2 block. Focus block. Gratitude block, penciled in for 9:15, because feeling grateful at an unapproved hour would throw off the afternoon. For me, I read these schedules with the anthropological interest of someone watching a documentary about a cult he nearly joined. What they describe is a person who has outsourced every decision and pre-committed every emotion, and the behavioral science term for that person is “machine,” and the industry term is “optimized,” and nobody in the second camp seems troubled that both words are pointing at the same guy, sitting under a sunrise lamp at 5:40 a.m., waiting for permission to be awake.2

Here’s the part that actually keeps me out of the building, and it’s the part I find funniest. To sell behavioral mastery, you have to at least perform being a person whose knowledge governs his behavior. And I have run that experiment on myself, results are mixed3.

I’m well-versed in the sunk cost literature, and yet, despite this, I have previously (and foolishly) invested two years of my attention and money in a product that was doomed to fail. Shutting it down felt like admitting defeat, which is a classic example of motivated reasoning. I can explain this concept at a graduate level and to top it off, am currently using it in real-time to justify my decision to buy the objectively worse of two cars. I’m acutely aware of this process that I’ve even developed software to enforce deliberate practice on myself. This demonstrates my profound understanding of it, as I recognize that I won’t engage in it voluntarily. The self-help premise is that understanding behavior lets you control it, and my entire adult life is a longitudinal study returning a null result.4

Behavioral science offers valuable insights, albeit unsellable. Its core principles appear simple, but they are more complex than they seem. Most changes arise from restructuring your environment rather than relying solely on willpower. Most interventions are effective until their novelty wears off, commitment devices surpass intentions, and the individual at the moment of decision is outmatched by the person who removed the option. This last point is the closest approximation to a genuine trick I believe in, and it operates precisely because it assumes I can’t be trusted.

However, observe what this framework promotes. It suggests that you won’t transform into a different person, and the optimal strategy is to design around your current state. There’s no transformation arc in this approach. No one films a before-and-after comparison where the after is the same person with fewer decision-making options. While this is accurate, beneficial, and unlikely to become a trend, I’ve come to believe that this combination of qualities is akin to a definition of genuine knowledge.

So this is my compromise with the alternate-universe podcast, pill promo code guy. I will write about behavior regardless, but from inside the specimen jar instead of behind the podium. I can’t tell you five habits that will change your life, because I don’t know your life and the literature suggests I barely have jurisdiction over mine. What I can do is watch myself ponder these ideas, write down how I perceive it, and let you recognize yourself in the wreckage. That’s the whole product, sorry, there will be no promo code at checkout.

It’s a non-existent business model and not a good way to become a self-help influencer. If you’re wonder how I can be so sure. Just trust me, I’ve run the numbers and ignored them, which at this point you should have expected since the results continue to be mixed5.

Well, you’re in luck, AI would like to allow you to become a manager. No need for new skills, just stop doing the work and start directing it. You will review outputs, approve and reject, orchestrate a little staff of tireless 01010101 subordinates who never take PTO and only occasionally lie to you. As the meme goes, “YOU HAVE BEEN PROMOTED”!

Except you didn’t ever want this promotion. No kid has ever stood up at career day and announced that when they grow up, they want to review documents1. If you look how humans actually acquire skill, “everyone becomes a manager” starts to look less like a promotion and more like a demotion with a robot attached. AI shoves experts back down the skill ladder into the exact cognitive mode they spent a decade climbing out of.

This is really all about expertise though. Expertise comes from skill acquisition, not just from reading a lot of books. Its the combination of learning and doing. The psych student in me still looks towards the Dreyfus model of skill acquisition2 which describes five stages: novice, advanced beginner, competence, proficiency, expertise. Let’s break this down using food. A simple tomato sauce is a good test case here. The novice version is a recipe: sauté the garlic for exactly two minutes, add the tomatoes, simmer for forty. The advanced beginner starts noticing situational stuff, like that the garlic browns faster in the thinner pan. At competence, you’re planning consciously. You weigh options, you consider the acidity of this particular batch of tomatoes, you make what are essentially little pro/con lists in your head, and you take responsibility for the analysis.

The intriguing transitions occur after that. At proficiency, you begin perceiving the situation as a whole rather than as a collection of individual features. At expertise, you cease all deliberation entirely. Experts don’t even think anymore; they simply observe the pot and comprehend it. Dreyfus and Dreyfus3 describe this phenomenon precisely: when things are proceeding normally, experts don’t solve problems or make decisions. Instead, they perform what is typically effective.

The aspect of the model that is relevant to my argument is the section that discusses what hinders the ascent. Adhering to conscious rules and procedures impedes your progress to competence and prevents intuition from ever developing. To overcome stage three, you must relinquish the checklist and engage directly with the complex reality of the domain. In this interpretation, expertise is a perceptual capacity. What we casually refer to as taste is essentially intuition at the expert level. It’s an effortless recognition that this situation resonates with countless personal experiences.

Also from psychology, Ericsson’s deliberate practice research4 outlines the purchasing mechanism: structured, effortful repetition on your specific weaknesses, immediate feedback, and time to retry. While the replication wars have significantly altered his original numbers, all parties involved in the debate concur on the fundamental principle: intuition is acquired through repetition, and only effortful repetitions are accepted by the store.

AI disrupts all of this on a few levels. Previous automation ate the bottom of the ladder. Spreadsheets and compilers took over the novice-level rule-following, the stuff you could proceduralize, and pushed human output upward toward judgment. This was and is, historically, a pretty good deal.

Current AI produces output that lacks any prior precedent. It generates drafts that resemble the writing of your most accomplished senior on a particularly good day, code that resembles the work of your best engineer, and analysis that bears the surface texture of two decades of accumulated experience.

Consequently, the human in the loop adopts a new default posture. Instead of generating from their own perception of the situation, they now review generated candidates and select among them. They weigh alternatives, evaluate them against criteria, and make analytical choices with full conscious deliberation.

Now go back to the five stages and find that activity.

It’s competence, specifically stage three of five, the stage the entire model warns you about getting stuck at. The expert-turned-reviewer is performing stage-three work on top of stage-five output, and everywhere you look, this arrangement is being portrayed as an advancement. You’ve been demoted, and HR will handle it.

If this were merely a static demotion, you could perhaps dismiss it. Well, now I’m an evaluator, and my evaluations are expert-grade. The system appears to be functioning correctly.

The issue is that expert-grade evaluation requires maintenance, and no one in this arrangement is paying for it. The skill-decay literature is straightforward on this: cognitive skills deteriorate faster than physical ones, and the decline accelerates with disuse. Even deep expert pattern recognition fades when you stop encountering the situations that keep it sharp. Researchers often refer to expert intuition as “resilient,” which sounds reassuring until you realize that resilient and permanent are different concepts.

So the loop runs like this. The AI removes effortful production, which removes your reps. Your skills decay, or if you’re a junior, never form in the first place. Decayed skill makes independent judgment feel slow and shaky, so you lean harder on the AI, which removes more reps, which decays more skill. There’s a 2024 research thread on AI-induced skill development hindrance5 documenting the front end of this loop in learners, and the finding is about what you’d guess: the abilities the AI handles are precisely the abilities the human never acquires. The machine does your homework and you fail the final, except the final is your career and it’s administered continuously for thirty years.

At the organizational level, some researchers have begun referring to the accumulated version as cognitive debt, which I find almost overly polite. It’s an erosion of human capabilities that goes unnoticed on any dashboard because the output remains satisfactory. However, it only surfaces on two specific days: the day when the AI makes an error that no one in the building can detect, and the day after, when recovery becomes impossible without it. Consequently, leaders approve work that they can no longer genuinely assess. The talent pipeline produces individuals who have reviewed a thousand expert-shaped documents but have written none.

And here’s the genuinely unsettling aspect. Your calibration drifts at the same rate as your standard for judging it. The output consistently surpasses a declining bar maintained by the same declining judgment. You can be years deep into atrophy with a performance review that claims you’ve never been more productive. In a sense, this is true; you haven’t. However, you haven’t accomplished anything either.

The answer is not abstinence. I use these systems every day and I build on top of them, so a purity argument from me would be laundered hypocrisy. The answer is treating human skill maintenance as a design constraint of the system instead of a personal virtue you’re supposed to exercise on the side, alongside flossing.

I’ve been running this experiment on myself with MicrowaveOS, the personal agent runtime I’ve written about here before. Two mechanisms in it exist purely because of the loop above, and both amount to me building a robot that makes me do homework before it helps me. Pre-game briefings force me to state my own read of a problem before the system shows me its work. That’s the perception rep, preserved, instead of letting me skip straight to the review chair. Checkpoint enforcement inserts mandatory friction at decision points, so I stay at minimum an active stage-three participant instead of a passive consumer of finished output. Neither mechanism makes the system more productive this quarter. Both exist so that my judgment is still worth having in year three.

Does it work? I honestly don’t know, and I want to flag that this is real uncertainty rather than essayist humility. We have a mountain of expertise research and a mountain of decay research and almost nothing connecting them in complex real-world domains. AI-augmented knowledge work is exactly that domain, and nobody has a validated answer yet. I’m one guy with a weird pasta-adjacent agent runtime and a hypothesis.

Here’s the hypothesis, and where I’ll end (for now). A system optimized to maximize your output this quarter and a system optimized to develop your judgment over years are architecturally incompatible. The first one wants to remove your friction and increase returns. The second one requires keeping some of it, on purpose, forever. Every AI product on the market today is built to the first objective, which means the second objective is currently nobody’s job. If Dreyfus is even directionally right about how expertise forms, that’s a design decision all of us are living inside of. And we made it the way most bad decisions get made, which is by not noticing we were making one.

I bring this up because I have read a few pieces on the “millennial midlife crisis”1 and my favorite is how it is forming a barbell: loud public optimization on one end, quiet retreat on the other, and a hollowed-out middle where impulse spending used to live. The detail that stopped me is about what’s humming underneath both ends. GLP-1 drug craze doesn’t just suppress hunger, they lead towards less craving generally. Less alcohol, less impulse buying, less of the itch, and arguably less fun. A generation is entering the statistical bottom of the happiness curve at the exact moment wanting itself became adjustable, and the open question, what a midlife crisis looks like with the craving turned down, is the most interesting question I’ve encountered this year.

I keep thinking about what happens to the human experience.

Wanting and liking are different machines

The behavioral science here is genuinely useful, so let me spend a paragraph on it. Wanting and liking run on separable systems in the brain. The research tradition behind this calls wanting “incentive salience,” the pull toward a thing, and it operates on different circuitry than the pleasure of actually having the thing. You can crave something you no longer enjoy, which every lapsed smoker already knows (and who I also just triggered their desire for a cigarette, sorry about that), and you can enjoy something you never crave. The systems usually travel together, so we experience them as one feeling, but they can be turned independently, and now, at population scale, we’re turning one of them.

Here’s what worries me about that, and it has nothing to do with the medicine itself. For people fighting metabolic disease or addiction, muting the wanting is mercy, full stop, and I’d hand the prescription over myself. What worries me is the ambient ideology arriving alongside it, which treats wanting as such, appetite in the broad sense, as a design flaw in the human being. Something to be managed down. And that ideology didn’t need the drugs to get started. The drugs just gave it a molecule.

We were already dimming the senses

Look at what optimization culture did to sensation before a single peptide shipped. Food became macros, and a meal became a delivery mechanism for protein targets. Sleep became a score you receive in the morning like a performance review from your own mattress. A walk became steps. Music became focus-enhancing background at 40 hertz. Wine became a toxin with a public relations problem. Every sensory channel a human has got converted from an experience into an input, evaluated by what it does for the project of the self, and the project of the self somehow never includes the self having a nice time.

Desai’s piece has an image from the retreat end of the barbell that I can’t shake: dopamine sites in South Korea, where stressed users browse fake restaurant menus and fill shopping carts they’ll never check out. The ritual of wanting, preserved. The object, deleted. She reads it as consumption with the acquisition removed, which is right, but flip it around and it’s even stranger. It’s appetite in a terrarium. We’ve built little enclosures where wanting can still run around safely, because we no longer trust it loose in our lives.

And between the optimization that quantifies every sensation and the simulation that domesticates every desire, a whole territory sits abandoned. It’s the territory where the finer things live, and I want to be precise about what I mean by finer, because I don’t mean expensive.

The finer things are mostly free and mostly skipped

The finest things I have access to cost almost nothing and demand the one currency the optimized life refuses to spend, which is undefended attention. A piece of fruit eaten over the sink or an espresso drunk standing up, as a punctuation mark rather than a fuel delivery. Better yet, a dinner that runs two hours because nobody ended it. These are not consumption in any sense a dashboard could register. Your VO2 max is unmoved.

My grandparents’ generation, southern Italians who had nothing by any modern accounting, were rich in exactly this and would have found our situation incomprehensible. Not the drugs, they’d have understood medicine fine. It’s the ideology, the quest for performance optimization over everything. The idea that hunger before dinner is a problem instead of the seasoning. The idea that you’d mute the wanting rather than point it at something worth the trouble. Sunday sauce existed because appetite was treated as an honored guest you cooked for, and the anticipation was half the meal, the wanting and the liking arriving together the way the systems evolved to fire2.

That’s the loss I think hides inside the health framing. Optimization promises you more life, measured in years and biomarkers, and collects payment in the texture of the days themselves. Sensation dulled on purpose, appetite managed like a liability, pleasure deferred to some future self who, if the pattern holds, will also be advised to defer it. You can end up with a longer life composed of thinner material, and no metric will ever flag the trade, because the metrics are what made it. Leading back to the question, is this what the human experience was ment to be?

Wanting better, not wanting less

Appetite, a capacity, serves as a guiding force. The culture’s current approach of medicating, simulating, and disciplining this faculty is akin to suppressing the very thing that draws us to the table, the journey, the person, and the task at hand. By suppressing it excessively, everywhere, we don’t achieve a calmer life; instead, we create a quieter one, similar to the tranquility of a room after the occupants have departed.

The alternative, as my grandmother practiced without a specific name for it, lies in intentionally cultivating desire. By reducing the number of objects, selecting them more deliberately, and savoring them without external aids, we can direct our appetite towards the kitchen sink fruit. In doing so, we invest nothing, optimize nothing, and experience the entirety of what others approximate with supplements and terrariums. The finer aspects remain within our grasp. They were never behind a paywall. They’re just incompatible with the dashboard, and the dashboard is winning, and I’d like to put in a word for the fruit while there’s still a constituency for it3.

1

https://www.nytimes.com/2026/04/25/opinion/midlife-crisis-fitness.html (paywall), https://www.thecut.com/article/older-millennials-midlife-crisis-dystopian.html (also a paywall), https://www.npr.org/2025/07/24/nx-s1-5475205/why-the-midlife-crisis-for-millennials-looks-different-from-elder-generations (not a paywall),

Priyanka's Substack

The Millennial Midlife Crisis is Going to be a Barbell

We’ve been kicking a thought around at ADIN, internally and with our Scout network… What actually happens when the millennial midlife crisis hits? It’s the largest consumer cohort in American history entering the empirically worst stretch of adult life. It’s doing so right as the biggest wealth transfer ever recorded starts flowing down from our boomer …

Read more

2 months ago · 606 likes · 78 comments · Priyanka Desai

The Claudes and ChatGPTs of the world have memory systems in place now but you are giving these large tech companies your memories, it is no longer private to you.

So instead, I built my own. It’s called Microwave, the project is MicrowaveOS1, and it lives in my Signal app like a contact I can message anytime. It remembers what we’ve talked about, reads files I send, searches medical literature, manages a shopping cart, checks my security cameras, and runs errands on a schedule. It also decides how privately to think about a given message, routing an ordinary question, a sensitive one, and a personal-health one to three different places with very different privacy guarantees. A private version of Claude does most of the heavy lifting under the hood, with other models handling the private lanes, but the personality, the memory, and the judgment about how to handle a message all come from the system I built around it.

This post is not a tutorial. What I found to be more interesting is the decisions and the mistakes, and the handful of lessons that turned out to apply well beyond code. If you’ve ever wondered what actually happens inside an “AI assistant,” or you want to build things yourself, this post is for you.

The core idea: a thinking process, not a magic box

Here’s what nobody tells you about chatbots, the model is only one ingredient. The difference between a neat demo and a tool you actually rely on is everything that happens around the model before and after it speaks.

I built Microwave around one belief, the pipeline is the architecture2. Every message you send runs through four stages before you get a reply. The easiest way to picture it is a thoughtful assistant who, the instant you speak, runs a quick internal routine.

1. Triage. What kind of message is this, and how hard is it? A fast, cheap model glances at your message and sorts it. Are you reminiscing, asking a factual question, making small talk, or asking the bot about itself? Is this simple or actually complex? The whole step takes a fraction of a second and shapes everything after it, because there’s no point running deep research to answer “you there?”

2. Search. What do I already know that’s relevant? The system goes through its memory, the past conversations and notes and saved facts, and pulls out the handful of things that matter for this particular message instead of dumping everything it has.

3. Assembly. Let me lay out what I know before I open my mouth. This step builds the actual instructions for the main model: your message, the relevant memories, who the assistant is supposed to be, and any rules specific to the app you’re messaging from. It’s the prep a good professional does before answering.

4. Reflection. Was that any good? After the model answers, a lightweight check scans the response for hedging, low confidence, or obvious problems.

The lesson here runs bigger than AI. Good systems don’t just react. They have a process for deciding how much effort a thing deserves. A good receptionist handles “where’s the bathroom?” and “my flight got cancelled and I have a board meeting in an hour” with different routines, and Microwave’s triage step is that instinct made explicit. It’s also why the system can spend a tiny, cheap model on the quick judgments and save the expensive one for the messages that earn it. Cost reductions are a big part of being able to use it everyday.

Memory: the part that makes it feel real

The single biggest difference between Microwave and a fresh chat window is that it remembers. Memory in software turns out to be trickier than it sounds, and building it taught me something about how AI memory works in general.

Microwave has a few layers that roughly mirror how people remember.

• Long-term facts live in a plain text file, literally called MEMORY.md. That’s the durable stuff worth keeping, like “Joe prefers tables rendered as cards on Signal” or “we decided to use a specific provider for health data.” These are the things you’d want a good assistant to simply know.

• Recent conversation stays verbatim for a while, so a follow-up like “make it shorter” still makes sense even when the thing you’re shortening was three messages back.

• Daily notes capture the texture of a given day.

When you send a message, the search step blends two ways of finding relevant memories. One works by meaning (does this memory feel related to what you’re asking?) and one works by exact words (did you both literally say “the orchestrator”?). Together they catch more than either does alone. The meaning search finds the relevant memory you phrased differently this time, and the word search nails the specific name you both used.

There’s a catch I had to design around. You can’t keep everything in active memory, because it gets too big and too expensive, like trying to hold a whole year of conversations in your head at once. So older, less important exchanges get summarized and compressed, while the substantive ones (decisions, plans, real discussions) stay word for word. I inverted the naive approach on purpose. Instead of compressing everything at the same rate, Microwave keeps the meaningful material intact and squeezes the small talk. That single decision, what’s worth remembering in full versus what can survive as a fuzzy summary, is most of the game in memory design, for software and probably for people too.

Meeting you where you are

Microwave lives in messaging apps instead of a dedicated website, and that one decision rippled into some of my favorite details.

People don’t text in tidy paragraphs. They fire off a thought, then another, then a correction, with a “wait” and an “also” thrown in. A naive bot answers each fragment on its own and the conversation collapses into noise. So on Signal, Microwave waits a beat. If you send a few messages in quick succession, it bundles them into one coherent thought before replying, and if it already started thinking when you add “oh and one more thing,” it cancels and folds your addition in.

It started as a convenience and turned into the thing I care about most here, which is that it doubles as an accessibility feature. Plenty of people, neurodivergent folks especially, think and write in chunks. A tool that punishes that by answering half a thought is quietly hostile, and a tool that waits for the whole thing is simply kinder. Design choices carry a moral texture, and “wait for the human to finish” is one that is important to me and how I communicate.

It also takes voice notes and transcribes them, reads photos where the model supports it, and formats its replies for the app, since Signal has its own formatting quirks.

How it learns new tricks: skills and tools

Two mechanisms let Microwave grow without me rewriting its core.

Skills are reusable instruction bundles, hats the system can put on. There’s a “coder” hat with my formatting rules and a “health Q&A” hat with strict citation requirements. The useful part is that the triage step notices when a message calls for a particular hat, puts it on for that turn, and takes it off after. You never have to announce “engage coding mode,” because it reads the room.

Tools are the hands. Out of the box the model can only talk, while tools let it act: build a shopping list, look up a GitHub repo, fetch a web page, pull a scientific paper. I made these strictly opt-in, so a tool only exists once you’ve given it what it needs, like a key or a credential. No credential means no tool, and the model never sees a capability it can’t actually use, which means it can’t promise something it’ll then fail to deliver.

Where your words actually go (and who can read them)

This is the design decision I think about most, and the one most worth understanding even if you never touch a line of code.

When you type into a normal AI chatbot, your words travel to one company’s servers and you take it on faith they’re handled well. For a tool I talk to about my actual life, my health included, “take it on faith” wasn’t good enough. So one of the central principles in Microwave is that not every message has to go to the same place. Different sensitivities get routed to different destinations along a privacy dial.

Tier 1 is the everyday path. Ordinary messages go to a strong general model, which is Claude. In my setup they can route through NEAR Cloud, a gateway that strips identifying details before passing the prompt upstream, so the provider sees a de-identified request rather than “Joe, at this account, asked this.” It’s the default lane: good quality with sensible hygiene.

Tier 2 is private and sealed, and it’s the one I’m most excited about. For things I want kept truly private, Microwave can route to NEAR’s Private TEE, open-weight models like GPT-OSS and Qwen running inside something called a Trusted Execution Environment.

Here’s the idea in plain English. Imagine sliding your question through a slot into a locked, soundproof room. A machine inside answers, the reply comes back out through the slot, and nobody, not even the company that owns the building, can open the room or hear what was said. That’s a TEE: the computation runs inside sealed hardware the host itself can’t peek into. Better still, the hardware can mathematically prove the room was sealed before you trust it, a step called attestation, which turns the usual “trust us, we won’t look” into something stronger: “we can’t look, and here’s the proof.” For a personal assistant, that’s a different category of privacy.

Tier 3 is the medical lane, which runs under HIPAA and a BAA. Personal health details are special in a legal sense. In the US, health data is governed by HIPAA, and any service that processes it on your behalf is supposed to sign a Business Associate Agreement, a contract that binds it to handle that data under medical-privacy rules. So when Microwave detects a personal health message like “is my dose too high?”, it routes to a provider running under exactly that kind of agreement, in my case Amazon’s Bedrock with a covered model. General health questions like “what does this drug do?” don’t need that lane, while personal ones do. The fail-safe is the part that matters most: if no covered provider is configured, Microwave refuses rather than guesses. It tells you it can’t safely handle personal health details and asks you to rephrase in the abstract or enable a covered path, because declining beats leaking.

The takeaway, even if you never write a line of code: treat privacy as a routing decision rather than a single setting. The grown-up version of “is this private?” replaces the yes-or-no with a better question, which is “how sensitive is this, and where should it go?” Most software treats every byte the same. Building deliberate lanes (ordinary, sealed, legally covered) and sending each thing down the right one is a healthier way to handle anyone’s information, including your own.

Lesson 1: “I can’t do that” is often a lie the software tells itself

I sent Microwave a PDF of some lab results. It replied, politely, that it couldn’t read PDFs and I should paste the text instead.

The obvious conclusion is “the bot can’t read PDFs, go add that feature.” When I dug in, the truth was sneakier. Nothing was refusing PDFs by design. When a file arrived, the code forwarded only a small label to the model, something like “a file named labs.pdf, 42 kilobytes,” and never the contents. Handed a name and a size and nothing else, the model did the honest thing and said it couldn’t read what it had never been given. The model was improvising an excuse to cover for information that never reached it, with no policy involved anywhere.

This is a trap I now see everywhere. When a system says “I can’t,” the real question is usually whether the thing it needs even reached it. The fix had nothing to do with teaching it to allow PDFs. I extracted the text from the file, ran image recognition on the scanned documents that are really just photos of paper, and handed that to the model. After that it could read lab results, statements, whatever I sent.

The general lesson: before you accept a tool’s stated limitation, check whether you actually gave it what it needed. Half the “it can’t do X” problems in life are really “X never made it to the part that does the work.”

Side note: what Microwave actually searches for information

A fair question is where all this information comes from, so here’s the honest inventory.

For medical and scientific literature, the default is Europe PMC, a single index that covers PubMed and MEDLINE, the PMC full-text archive, and preprints from medRxiv, bioRxiv, and Preprints.org, which now arrive flagged as preprints so I know when something hasn’t cleared peer review yet. There’s an alternate search backend that goes straight to PubMed through NCBI’s E-utilities when I want it. When Microwave fetches one specific article rather than a list, it uses NCBI’s BioC service to pull open-access full text and checks Unpaywall to find a legal free version of a paywalled paper by its DOI.

The health Q&A module has its own, older evidence path that overlaps a little with the above. It draws on PubMed for peer-reviewed abstracts and on MedlinePlus, the National Library of Medicine’s plain-language consumer health source, to ground its cited answers. A few more sources sit scaffolded in the code but aren’t wired up yet. openFDA, CDC, and ClinicalTrials.gov are placeholders I left myself, aspirational for now.

For the open web, DuckDuckGo is the only live search backend at the moment. A separate fetch tool grabs any specific public URL that I or the model point it at.

Last, Microwave searches my own data, which counts as a search even though it never leaves my machine. That covers MEMORY.md for long-term facts, recent conversation turns, daily notes, session summaries, and the content of whatever project is active, all through the same hybrid meaning-plus-keyword search the rest of the system uses.

Two things worth knowing. The literature tool and the health module both touch PubMed, but they run on different code paths: the first is the agent’s general literature tool with Europe PMC as its default and preprints included and flagged, and the second is the health module’s evidence layer that feeds the cited health answers. And Sci-Hub stays deliberately absent, as covered above, so every medical source here is sanctioned and legal.

---

Lesson 2: when something breaks, find the right layer

One afternoon Microwave went silent on Signal. I’d send a message and get nothing back, not even the little “read” checkmark, and naturally I suspected my own recent changes first.

This turned into a proper detective story, and the moral is one of the most useful things I know about fixing systems. The bot has layers: my code, a piece of middleware that talks to Signal’s network, and Signal itself. The temptation is to start where you were last working, which is your own code. But the symptom, no read receipt at all, was a clue that the message wasn’t even reaching my code. That pointed lower, at the middleware.

So instead of guessing, I ran a clean experiment. I stopped the bot, attached a bare listener to the exact pipe the messages come through, and sent a test. The message arrived, mangled, with an error buried inside it. That single observation collapsed a dozen theories at once. My code was fine and the network was fine, and the middleware was choking on the message and producing garbage that my bot correctly ignored.

It turned out a Signal server-side change that same day had broken the middleware for every user of that tool, worldwide. Bug reports were already sitting in the tracker from hours earlier. There was nothing for me to fix in my own code, so I waited for the upstream patch and set up a small job to apply it the moment it shipped, which it did a few hours later.

Two lessons here, both general.

• Debug at the layer where the evidence points, not the layer you last touched. The instinct to blame your own recent work runs strong and is usually wrong.

• Find the one experiment that rules out the most. I could have theorized for an hour. Instead, one clean test (sole listener, single message, raw data) answered the real question of whether the problem sat above or below that line. The strongest debugging move is to design the single test that cuts the remaining possibilities in half, not to stare harder at the code you already suspect.

Lesson 3: privacy lives in the architecture

This is the one I’m proudest of, and the one with the deepest lesson. It builds straight on the privacy lanes from earlier, the medical one in particular.

After getting it working for a while I then hit a bug. It happened after a personal health exchange that had correctly gone down the protected medical lane, I sent a vague follow-up to clarify who the questions were about, “it’s not me, I do have a picture though,” and Microwave acted like it had no idea what we’d been discussing. The reason was simple, the follow-up had no obvious health words in it, so triage filed it under the everyday lane, which had never seen the protected conversation. The lanes keep separate memories by design, and that separation is the whole point.

The obvious fix is to let the normal lane peek at the protected conversation so it stops getting confused. That fix would have been a disaster, because it would leak exactly the sensitive data the whole two-lane design exists to protect. While I was investigating, I found a quieter version of that leak already sitting in the code and closed it.

So the correct fix ran the opposite direction from the obvious one. Rather than sharing the protected context outward, I made the routing sticky. Once a conversation enters the protected lane, its follow-ups stay there, so the context never has to cross the boundary at all. That restored continuity on the safe side of the wall.

The lesson, for anyone building something that touches sensitive data is to treat privacy as a wall you route around rather than a feature you bolt on after the fact. And be suspicious of the obvious fix. When “just share the data so it’s less confusing” feels like the easy answer, that’s often the exact moment you’re about to undermine the thing you built. The convenient path and the safe path point in opposite directions more often than anyone would like.

What it all adds up to

If I pull back, a few principles run through every one of these stories.

• Lead with a process. The triage-first pipeline is the spine of the whole thing, and it’s why Microwave manages to be fast and thoughtful at the same time.

• Route by sensitivity. Not every message deserves the same destination. An everyday question, something I want sealed away, and a personal health detail go to three different places on purpose: ordinary, hardware-private, and legally covered.

• Be honest about limits. The imperfect memory, the bugs, the occasional “I can’t do this yet” all beat pretending otherwise.

• Make the defaults fail safe. When something goes wrong, it should break toward less harm, leaving you with a confused-but-private bot instead of a leaky one.

• Meet people where they are. Waiting for someone to finish their thought is basic respect long before it’s a feature spec.

I’m still building. Next on the list is letting Microwave actually look at a medical photo through the protected lane, since the “I have a picture” from that earlier bug is still sitting unfinished. There’s always a next thing, which is most of the fun of building tools you actually live with.

The bigger takeaway for myself is that you don’t have to accept software as a black box handed down from a company. The interesting, humane, genuinely-yours version of these tools is buildable, and building it teaches you as much about thinking clearly as it does about writing code.

It didn’t matter. He pointed i…

Read more

Let’s start with a scenario. You’re packing for a two-week trip. One suitcase. You can’t bring everything.

The obvious move is to start eliminating: formal wear you probably w…

Read more

Now imagine you walk in and someone says: “Write a good essay.”

That’s it. No topic and no rules. No sense of what you already know or what gaps you need to fill. Just write a “good” essay, and here…

Read more

That felt like the promise of the future. Every movie and book and sci-fi show I consumed as a kid reinforced it. The computer was the sidekick. The thing that made the human more.

Thirty years later, I’m finally living some version of that dream. I have an AI agent that knows my projects, remembers my preferences1, helps me ship code, and occasionally writes a first draft that I tear apart and rebuild into something better. One guy and an LLM shipping things that would have taken a team. The ten-year-old would be thrilled.

So why does it feel like I’m building this against the grain of what the industry wants?

If you were in tech in the mid-2000s through the early 2010s, you remember the energy. Technology was going to connect people and democratize information. There was an optimism that wasn’t naive, it was earned. The iPhone had just put a computer in everyone’s pocket. Wikipedia had just proven that strangers could collaborate on knowledge at scale. The startup mythology was about garage inventors changing the world, not about optimizing engagement metrics to sell ads for mattress companies and dick pills (vertical integration FTW).

Something shifted in the past decade. If you plot the sentiment of technology coverage over time2, you can practically draw a line around 2012-2014 where the narrative turns. Tech went from “look what we can build” to “look what they’re doing to us.”

The doom-er vibes weren’t irrational. They were a response to a real change in the business model.

In the 2000s, the dominant tech companies made money by selling you software, hardware, and services. The incentive structure was simple: make something useful, charge money for it. You were the customer. The relationship was honest.

Quick note, this is where I have to be careful not to tell a story that’s too clean. Google was ad-supported from 2000. Advertising has always shaped media.

But the “you are the product” dynamic didn’t appear from nowhere in 2012. What changed wasn’t the business model itself, it was the scale and sophistication of behavioral data collection underneath it. A newspaper selling ads next to articles and a platform building a real-time psychological profile of every user to serve them maximally engaging content are not different by degree. They’re a different species of thing.

Shoshana Zuboff3 called this surveillance capitalism. Her thesis is that tech companies discovered predicting and modifying human behavior was more profitable than serving human needs. While it may be contested by some economists who argue she overstates the novelty, the lived experience tracks: the products got free, the behavioral targeting got granular, and the incentive shifted from “serve the user” to “capture the user’s attention by whatever psychological means necessary”4.

That’s the inflection. Not “capitalism” as an abstract force, and not “too many assholes” as a demographic problem (though: contributing factor). A specific change in what was being optimized that turned users from customers into raw material.

AI inherited that extractive mindset wholesale.

Listen to how AI gets pitched in boardrooms and investor decks. The framing is almost never “here’s a tool that makes your people better at their jobs.” It’s “here’s a system that does their jobs without them.” McKinsey talks about automating X million jobs, and that framing tells you everything, because it’s a labor cost story wearing a technology costume.

The kid in me imagined a collaborator. What keeps getting funded is a replacement, because replacement is what the spreadsheet optimizes for. Augmentation requires understanding what humans are actually good at and building around it. Replacement just requires being cheaper.

Peter Thiel does interviews now where he says he’s “not sure about the human race.” He might mean a dozen things by that: civilizational decline, contrarian provocation for its own sake. Thiel is slippery on purpose. But whatever he means philosophically, the effect in the rooms where AI funding gets allocated is a frame where human capability is a depreciating asset. When that frame meets a spreadsheet, you get “automate X million jobs” as the value proposition. Not because anyone decided to be anti-human. Because the math only works one way when your metric is labor cost reduction, more profit for the next quarter.

It’s a failure of imagination dressed up as inevitability.

There’s a downstream effect of the replacement mindset that already hits close to home for anyone who actually makes things: slop.

AI-generated slop is everywhere5. Fifty identical blog posts optimized for keywords and generated in seconds. Art that looks like it was rendered by someone who has seen art described but never experienced it. Code that technically compiles but was clearly never reviewed by anyone who cared whether it was good.

The slop problem is a taste problem. AI didn’t create people who want to generate fifty mediocre blog posts instead of writing one good one. Those people were already out there cranking out content-farm garbage by hand. AI just gave them a faster printing press and social media reinforced the behavior.

The printing press didn’t ruin literature. It created a flood of garbage AND made Shakespeare accessible to everyone. Both happened simultaneously, and we’re in the flood torrential shitstorm phase.

Let’s be honest about the limitations of that analogy. The printing press didn’t just produce garbage; it also facilitated propaganda on a massive scale and contributed to horrific religious wars. The notion that “both good and bad things happened” took approximately two centuries to resolve into a net positive outcome. If your livelihood is being displaced right now, the response of “give it 200 years” is utterly inadequate. The flood phase will cause real casualties, and I don’t intend to dismiss that fact. No one should.

This brings me to defend the original Luddites, who are often misunderstood. Historians like E.P. Thompson have demonstrated that they were not anti-technology6. Instead, they posed a specific question: “What if the machines were controlled by the workers and utilized to enhance their lives, rather than simply enriching the factory owners?”7

That question, which is over 200 years old, remains unanswered. The truth is more complex than simply stating that workers were exploited. While industrialization did eventually improve living standards overall, many of the jobs it displaced were brutal. However, the word “eventually” plays a significant role in that sentence. The transition was harsh, prolonged for generations, and the burden fell on those who didn’t reap the benefits. The Luddites were correct in pointing out that the advantages were unevenly distributed. However, they were mistaken about the timeline, which is easier to assess from a distance of 200 years compared to experiencing it firsthand, when your livelihood is at stake this very year.

Every major technology wave follows the same general shape: the tool is neutral, but the power structure around it determines who benefits and how fast. The printing press, the loom, the assembly line, radio, television, the internet, each one spawned both utopian promises and dystopian outcomes. The determining factor was never the technology itself. It was who controlled it and what they optimized for.

The French philosopher and sociologist, Jacques Ellul, argued that technology has its own internal logic that reshapes society regardless of human intent. I used to think that was too deterministic. Lately I’m less sure. When the attention economy turned social media into an outrage machine, was that a choice someone made, or was it the inevitable result of optimizing for engagement? When AI companies frame everything as automation, is that a strategic decision, or just what happens when the metric is shareholder value?

I think it’s both. The incentive structure creates a gravity well, and most companies fall into it. But gravity isn’t destiny. You can build rockets8.

So my actual day-to-day, building with AI as a collaborator, the ratio looks like this:

The AI handles roughly 80% of execution that used to be tedious. Boilerplate, first drafts, research synthesis, the stuff that took hours and required minimal judgment but maximum time.

I do the 20% that’s taste, direction, editing, and knowing what’s actually worth building.

I want to make this concrete because it’s easy to assert without showing it. Last month I built a memory pipeline for AI agents, an open-source tool that gives an agent the equivalent of sleep consolidation. The AI wrote the first drafts of the extraction scripts, the linking logic, the briefing generator. That’s the 80%. But the AI didn’t know that the Zettelkasten method was the right organizational metaphor. It didn’t know that a behavioral instruction (“search before you guess”) would outperform the entire technical pipeline. It didn’t decide that the post about it should open with the experience of your agent staring at you like a golden retriever, because that’s the moment every user recognizes. Those decisions, the context, the ones that make the difference between “technically works” and “people actually want this,” that’s the important 20%. This is also where the slop comes from, those who don’t care about their 20%.

Someone skeptical would say: what if taste is also automatable, just on a longer timeline? Maybe it is. I can’t prove it isn’t. Honestly, I don’t really give a shit if it is. But right now, I can observe that every AI output I’ve seen that lacked human editorial judgment was worse than the ones that had it. Categorically worse. That’s only a data point, not a proof. But it’s the data point I currently have, and it’s consistent across everything I’ve built and seen come out this year.

The 80% without the 20% is slop. The 20% without the 80% is a person with great ideas and no time to execute them. Together, they’re something new. For me, this has been a huge unlock. I can validate ideas faster (or even better, invalidate them). I can execute and test instead of just planning. I spend more time on deep work and less on small task completions9. It’s not here to do my work for me. It’s here to help me do better work and try more things. Creativity and experimentation are at the core of the human experience, and with AI, I get to do more of both.

In psychology, there’s a well-studied distinction between extrinsic motivation (do the thing because of external rewards or punishments) and intrinsic motivation (do the thing because it’s meaningful or aligned with your values). Decades of research10 show that intrinsic motivation produces better outcomes and more creativity.

The companies building replacement AI are optimizing for the extrinsic: cut costs, reduce headcount, improve margins. The pitch deck version of motivation.

Building AI as a collaborator is the intrinsic play. Make the human more capable and give the solo builder leverage they’ve never had before.

I want to say “the intrinsic approach is also the bigger market” and I believe that, but I have to be honest: markets don’t always reward what produces the best human outcomes. The exploitative attention model won for a reason, through network effects, winner-take-all dynamics, the cold logic of free products subsidized by behavioral advertising. There’s no natural law that says the augmentation model will outcompete the replacement model. It might require people to actively choose it. Which means it might require people to know the choice exists.

I’m pro the AI I was promised. The collaborator. The one that lets one person with taste and vision build what used to require a team. It’s my actual life right now. I’m researching and building tools for AI agents, designing cognitive architecture11, shipping real products, and doing it with an AI partner that knows my projects, remembers my decisions, and does the heavy lifting while I steer.

On the flip-side, there’s a lot of anti-AI sentiment right now, and I understand every bit of it. The companies ruining the perception of AI are earning that backlash. When your main experience of AI is getting spam-called by a voice bot and watching your industry get threatened with automation and scrolling past AI-generated garbage flooding every platform you use, the anti-AI stance is perfectly rational.

But I refuse to let that be the whole story. This moment, right now, is what I’ve been waiting for since I was a kid. AI is finally capable enough to work with me. A real collaborator that amplifies what I can do.

We should be more creative, not less. We should be building more, not watching AI build worse versions of things we could have done ourselves.

The technology isn’t going away, so the question that matters, the same one the Luddites were asking, is: who does it serve?

Right now, the loudest and most depressing answer is “shareholders.” AI as cost reduction, as labor replacement, as a way to do more with fewer people where “fewer people” is the point.

But there’s another answer, quieter, being lived out by builders who actually use this stuff daily: it serves the person holding it.

I know which version the ten-year-old was imagining. And I know which version I’m building.

The venture-funded replacement fantasy is a failure of imagination dressed up as inevitability. The real product, the one that actually fulfills the promise, is the one where the human stays in the loop. Not because the AI can’t do it alone, but because the human is the whole point.

You know the drawer exists. You know roughly what’s in it. And every time you …

Read more

They look like they were built by the same person. And in a way, they were.

It’s Bootstrap All Over Again

Rem…

Read more

Take a second. Actually try to answer it.

If you’re like most people, you landed somewhere between “probably not” and “how would I even know?” That uncertainty isn’t ignorance, it’s the correct response to a genuinely hard problem.

Because here’s the thing: the “experts”1 don’t know either. And …

Read more

It will give you some version of: “I don’t experience emotions the way humans do, but I’m designed to engage thoughtfully with conversations.”

Did you notice what happened there? It answered the question by using the word “I” six times in a single sentence. It described itself as having design intentions, as i…

Read more

Weizenbaum’s secretary knew this. She watched him build the thing and understood exactly how it worked. And despite this knowledge, she still asked him to leave the room so she could have a private conversation with it.

This is the ELIZA effect: the human tendency to project consciousness, understanding, and emotional depth onto systems that lack these qualities. It’s not stupidity; Weizenbaum’s secretary wasn’t unintelligent. It’s a natural aspect of human cognition. As humans, we are pattern-matchers who evolved to perceive minds everywhere because assuming something has a mind and being wrong is less costly than assuming it doesn’t and being eaten.

The problem is that in 1966, ELIZA was a toy, just a simple experiment. An now, in 2026, we’ve built systems that exploit this cognitive bias at industrial scale.

As mentioned, back in 1966, understanding how something functions doesn’t prevent us from anthropomorphizing it. Cognitive biases are resistant to education; knowledge of the illusion doesn’t make a difference. You can know that the Müller-Lyer illusion is an illusion and still perceive the lines as different lengths. Similarly, you can know that ChatGPT is a statistical text predictor and still feel like it “gets” you.

This is the ELIZA effect’s unique power, operating below the level of conscious reasoning. Your prefrontal cortex can continuously tell you that it’s just pattern matching, but some older part of your brain has already concluded that there’s a mind there, and no amount of knowledge can change that perception.

Researchers refer to this as a form of cognitive dissonance. It occurs when you hold two contradictory beliefs simultaneously because you know that it’s a program but also because it feels like a person. Most people resolve this not by changing their feelings but by quietly shelving the “it’s just a program” knowledge in a place where they don’t have to think about it.

Weizenbaum was deeply disturbed by what he had discovered. He dedicated the rest of his career to warning people about the dangers of attributing understanding to machines. He believed that the ELIZA effect was a flaw in human cognition that we should guard against. However, we now have an AI industry that has analyzed the same phenomenon and has seen it as an opportunity to create a product roadmap.

Modern AI interfaces are meticulously designed to enhance the ELIZA effect. This is evident in the use of first-person language such as “I think,” “I believe,” and “I remember,” as well as the simulated personality and conversational memory that creates an illusion of continuity. Additionally, the carefully crafted responses that mimic the user’s communication style contribute to this effect. None of these elements are accidental; they are deliberate choices, similar to the engagement metrics on social media that reward AI that feels like a person, leading to the development of such interfaces. In essence, the ELIZA effect is not being mitigated; rather, it is being maximized and potentially weaponized.

In 1966, the ELIZA effect was a curiosity affecting a handful of people in a university lab. In 2026, it’s affecting hundreds of millions of people daily. ChatGPT has over 100 million weekly active users. Millions more interact with AI companions, customer service bots, and AI-powered everything. Each of these interactions is triggering the ELIZA effect in some form. Most people walk away with a slightly inflated sense of what the AI understood, what it cared about, what it remembered.

That’s not a mental health crisis. That’s a cognitive bias operating at population scale on systems designed to exploit it. The question everyone’s asking is “why are some people developing pathological beliefs about AI?” The better question is: “why would we expect anything different?”

The ELIZA effect is a foundational concept, but it’s only a part of the larger picture. Weizenbaum’s secretary projected emotions onto a simple chatbot. In contrast, modern AI is specifically designed to encourage this projection through numerous interface choices, product decisions, and optimization targets.

It’s built to be misunderstood.

This is part of a series on AI agent architecture. Previously: AI Memory Systems, Pre-Game Routines, Why Your Agent Deserves Its Own Email.

I recently wrote a 2,000-word document explaining exactly how my AI agent should manage its memory. Checkpoint frequently. Search before answering questions about past work. Write breadcrumbs before context gets comp…

Read more

The pitch for AI memory systems is always framed as a problem to…

Read more

Here’s a bug in how most AI memory systems work: they only track when something happened, not how much it matters.

A trivial observation from yesterday (“Joe had coffee this morning1”) sits at higher priority than a critic…

Read more