I Tried Living Like It Was 2098 for 30 Days. Here’s What Actually Happened.

Part 1 of this piece is about what it's already like to live in 2098 - because in ways that are easy to miss and hard to un-see, we're already there. This part is about what happened when I ran a 60-day experiment: thirty days leaning all the way into 2098, then thirty days deliberately stepping back from it.

The experiment had two phases. I didn't plan it that way. That's just how it went.

Phase one: thirty days living as fully in 2098 as I possibly could. Not in the vague sense of "using AI tools more." In the specific, deliberate, technical sense of building an actual AI-managed operational layer for my professional life and then handing it the keys. Claude handling communication, reasoning, and writing. OpenAI Codex -via Claude Code- handling automation, system integration, and workflow orchestration. Connected through Make.com to my email, calendar, files, and browser. The machines, as much as I could arrange it, would run the operation. I would observe.

Phase two -which I only decided to attempt because of what phase one showed me- was thirty days of deliberate friction. One thing a day that neither machine could do for me. One hour of unstructured time. One morning a week without the inbox.

I live in Ottawa. I work in front of screens. I am exactly the kind of person this series has been writing about.

Here is what actually happened.

Phase One: Building 2098 (Days 1-7 — the Chaos)

It did not start as a superpower. It started as a debugging session that lasted four days.

The setup looked elegant on paper. Claude as the reasoning and communication layer -reading incoming email context, drafting responses calibrated to tone and relationship, summarising threads, flagging priorities. Codex handling the automation scaffolding -writing the scripts that connected the tools, building the Make.com workflows that piped information between Claude, my calendar, my file system, and my browser. The two systems in a kind of symbiosis: Codex building the plumbing, Claude thinking through the pipes.

The first workflow I built was email triage. Simple in theory: incoming email arrives, Make.com sends it to Claude with a context prompt, Claude categorises it and drafts a response, the draft lands in a review folder. Twenty minutes of setup. Or so I thought.

The first run produced eleven emails categorised as "urgent" including a newsletter I'd been ignoring for three months and an automated receipt from a parking app. Claude had, reasonably, interpreted "urgent" through the lens of my prompt rather than my actual priorities, which I hadn't specified clearly enough. I rewrote the prompt. Better. Still not right -it was now missing genuinely time-sensitive threads because the new prompt was too conservative. Third iteration. Closer. By the fourth version, three days in, the triage was working at roughly 80% accuracy, which sounds good until you're the one reviewing the 20% and wondering what else you're missing.

Codex meanwhile had written a perfectly functional Windows connector that kept breaking on authentication refresh. The error was consistent -a token expiry issue that Codex had accounted for in the logic but that Make.com's Windows connector was handling differently than the documentation suggested. I fed the error back to Codex. It produced a revised script. Different error. I fed that back. This loop ran seven times over two days before the connector stabilised.

This is the part of the 2098 story nobody tells you: building it costs. Not just time. Attention. The specific kind of iterative, debugging, read-the-error-message-carefully attention that is itself a form of the cognitive overhead you were trying to eliminate. You are, in week one, working harder than before the tools existed. You are also, if you stay with it, building something that will eventually work without you.

By day seven the Cowork integration was stable. Claude could see my file system, access relevant documents before drafting responses, and -this was the moment the setup shifted from project to infrastructure- it could query context I hadn't explicitly given it because the connector had already surfaced it from my working files.

That morning felt different. Not like using a tool. Like having a colleague who had read everything I'd ever written and remembered all of it.

Phase One: 2098 Running (Days 8-14 — What It Actually Feels Like)

By week two the system was calibrated well enough to run.

The email workflow was handling first drafts at around 90% usability. Not 90% send-ready -I still read every draft before sending, and probably always will- but 90% of the cognitive work done before I touched it. The calendar was managed by an AI scheduling layer that had learned, through six days of corrections, that I don't take calls before 10am, that I need 45 minutes between back-to-back writing sessions, and that "quick catch-up" is rarely quick. When it made a mistake now, the mistake was interesting rather than obvious -a nuanced judgment call about priority rather than a basic misunderstanding of my preferences.

The writing assistance was the most technically impressive and the most personally complicated.

I had built a workflow where Claude had access to my research notes, my previous articles, the style guide for this series, and the current draft. When I opened a new piece, it would synthesise relevant context, suggest a structural approach, and offer a first pass at the opening section. I would react to that pass -cut it, redirect it, argue with it in the prompt- and iterate from there. The finished articles were better-structured and faster to produce than anything I'd done alone.

They also felt, in a way I couldn't immediately explain, slightly less mine.

Not the ideas -the ideas were mine, the arguments were mine, the voice was mine in the sense that I'd trained the system toward it and corrected it when it drifted. But the felt experience of making them was different. I was conducting rather than playing. Directing rather than performing. And conducting is real work, skilled work, difficult work. It is also a different relationship to the output than the one I'd had before.

By day twelve I finished a piece I was legitimately proud of. Sharp argument, clean structure, good sentences. I traced through what I'd done. The structure had been suggested by Claude after analysing three similar pieces I'd written. The opening had gone through four AI-assisted iterations before I found the line I wanted. Two of the five best sentences had arrived in a Claude draft and survived because I'd recognised they were right.

I had directed every decision. The judgment was mine. The taste was mine.

But I had been, in some specific and hard-to-name sense, a very good editor of a very capable collaborator. And I wasn't sure, sitting there in Ottawa at 11pm, whether that was the future of creative work or the slow disappearance of something I hadn't known I needed until it was mostly gone.

I decided to find out what I was missing by going looking for it.

Phase One: 2098 Deepening (Days 15-21 — When It Got Interesting)

By week three something shifted that I hadn't anticipated. The system stopped being a project and started being infrastructure. I stopped thinking about it the way you think about something you built and started thinking about it -when I thought about it at all- the way you think about electricity. It was there. It worked. The attention I'd been spending on it had been quietly reallocated elsewhere.

This is when the interesting questions started arriving.

The Claude Cowork integration had by now been running long enough that it had genuine context -not just the documents I'd explicitly pointed it at, but the accumulated pattern of three weeks of interactions. It knew which topics I returned to. It knew which draft directions I consistently rejected. When I opened a new piece on a Wednesday morning, its structural suggestions were calibrated in a way that felt less like a capable stranger's input and more like a collaborator who'd been paying attention. The suggestions were still wrong sometimes. But they were wrong in interesting ways -wrong because of a genuine misread of my intention rather than a generic miss. I could tell the difference, and the difference mattered.

The Codex automation layer had also, by week three, stopped generating errors I hadn't seen before. The Make.com workflows were stable. The Windows connector that had broken seven times in week one had not broken once in week two and ran without incident through week three. The authentication refresh issue that had consumed six hours of debugging in the first days was now a solved problem, living quietly in the script, doing its job without announcing itself. This is the specific pleasure of automation that nobody talks about because it's so undramatic: the thing that used to require your attention no longer requires it, and you only notice this because you suddenly have that attention available for something else.

What I started using it for was thinking. Longer, slower, less interrupted thinking than I'd had access to in years. The cognitive overhead that email and scheduling and first-draft grinding had previously consumed was now running in the background, handled, and what was left in the foreground was the capacity to actually develop an idea rather than sketch it and move on. I wrote longer notes to myself. I followed tangents. I let a question sit for two days without resolving it prematurely because I wasn't behind on anything that needed the time I was spending on the question.

The writing got better. Not faster -it was already faster. Better in the way that writing gets better when the writer has had time to think.

Phase One: 2098 Fully Running (Days 22-30 — the Uncanny Part)

Week four was when it got genuinely strange.

Strange in a good way, mostly. But strange in a way that required paying attention.

The system was now running at something like 95% of what I'd imagined when I designed it. Email triage at 90% accuracy. Scheduling handled without my involvement except for anomalies the AI flagged for judgment. Research synthesis arriving as a structured brief before I'd consciously decided I needed it, because the workflow had learned to recognise the topics I was circling and surface relevant material proactively. The Codex-built automation layer had, at this point, absorbed enough of my working patterns that it was making correct predictions about what I'd need next often enough to feel, occasionally, slightly eerie.

The eerie feeling is worth describing precisely because I think it's important and I haven't seen it written about honestly.

There were moments in week four when a Claude draft arrived that was so close to what I would have written -not perfect, not finished, but the right register, the right level of directness, the right structural instinct- that my response to it was not satisfaction but a mild, unplaceable discomfort. Like hearing your own voice played back to you. Familiar in a way that is slightly too close. The system had learned enough about how I work that it was, in a specific technical sense, modeling me. And being modeled accurately is a different experience than being helped.

I don't know exactly what to do with that observation. I'm including it because it happened and because pretending otherwise would make this a less honest piece.

What I do know is that week four also produced, alongside the eerie, something I hadn't predicted when I started the experiment: genuine gratitude for the machine layer. Not the performative "AI is amazing" gratitude of a product review. The specific, personal gratitude you feel for something that has taken a real burden from you and carried it reliably and well. By week four the system had processed roughly 340 emails, managed 23 scheduling decisions, generated first drafts for four articles, and maintained the connective tissue of my professional life with an accuracy and consistency that I, doing all of it manually, would not have matched. I would have been slower, more distracted, more tired, and more likely to drop something.

The machines had done their part. Remarkably well, by the end.

And that's when the time appeared.

What Thirty Days of 2098 Actually Gave Me

Here is the part of the story I didn't expect to write.

By the end of week two, the system was running well enough that something genuinely strange happened. I had time. Not scheduled time. Not optimised time. Actual, unaccounted-for, nobody-is-waiting-for-anything time. The cognitive overhead that had previously consumed my evenings -the emails to draft, the scheduling to manage, the research to synthesise, the first drafts to grind through- had been substantially absorbed by the machine layer. And what was left, sitting there in Ottawa with a Tuesday evening that had no obligations in it, was me.

So I played video games. I am bad at them. I remain bad at them. This has not changed and I have no expectation that it will. But I played them for two hours on a Tuesday night and the guilt that would normally have accompanied two hours of non-productive Tuesday was simply absent, because by any measurable standard I had already been more productive that day than a Tuesday without the AI layer would have allowed. The machines had created a surplus. The surplus was mine to spend however I wanted.

I watched YouTube for stretches I would previously have called embarrassing. Documentary essays, mostly -the kind I'd been meaning to get to for months and had been saving for some future version of myself with more time. That version of myself had apparently arrived, on a Wednesday afternoon, in the middle of a month I'd set aside for an experiment.

I started making videos of my own. This surprised me more than anything else. I'd been dimly aware for years that I was interested in the form -had bookmarked tutorials, had thought about it in the abstract- but had never found the moment to start. The moment, it turned out, was not a moment at all. It was a condition. The condition was having enough time and enough low-stakes cognitive space that starting something new didn't feel like a withdrawal from something more urgent. The AI layer had created that condition by handling the urgent. So I started. I'm not good yet. I'm making the kind of videos that exist primarily to teach the person making them how to make better ones. But I'm making them, which I wasn't before.

I spent more time with my wife. More time with my kids in the specific way that counts -not physically present while mentally composing an email response, but actually there, actually paying attention, actually available in the way that the people who love you can tell the difference between and the way that matters most. The AI was handling the logistics of my professional life with enough competence that the logistics stopped bleeding into the hours that were supposed to belong to other things.

I fixed the lawn. This sounds trivial and I'm including it anyway because it's the most honest detail in this whole piece. The lawn had been losing a war against weeds for two years. I'd been meaning to address it since the spring before last. What I'd lacked was not knowledge or resources or even time in the abstract -I'd lacked the mental availability to do something low-priority and physical and unhurried on a weekend afternoon without the background hum of everything I should be doing instead. The AI layer quieted the hum. The weeds got addressed. I spent three Saturday hours doing something that required nothing of my professional self and everything of my hands and my knees and my tolerance for manual repetition.

And somewhere between the lawn and the video games and the YouTube documentaries and the evenings that ended when I wanted them to rather than when the work allowed them to, something landed that I hadn't been looking for.

This is what 2098 feels like from the inside. Not the sleek, frictionless, everything-optimised version that the tech conference keynotes describe. The real version. The version where the machines handle enough of the cognitive overhead that the human is left with a question the machines can't answer: now what?

And the answer -my answer, arrived at on a weedy lawn in Ottawa with dirt on my hands and nowhere particular to be- was not what I'd have predicted.

More manual labour. More human connection. More making things from scratch that didn't need to be made from scratch. More daydreaming. More starting things I wasn't good at yet. More of the slow, unhurried, friction-full life that the machines were supposedly replacing.

Not instead of the machines. Because of them. The machines created the space. I filled the space with exactly the things the machines can't do.

That's when it hit me. When 2098 actually rolls around -with its AIs and AGIs more capable than anything running today, handling more, replacing more, doing more of the cognitive and logistical work that currently fills the hours- the humans inside it won't be floating in frictionless leisure. They'll be on their lawns. In their studios. At their kids' games. Making videos nobody asked for. Playing games they're bad at. Building the chicken coop their neighbour has been trying to get them to help with for three years.

The counterintuitive truth that thirty days of living like 2098 revealed: the more the machines do, the more human we become. Not by retreating from the tools. By finally, for the first time in most of our adult lives, having enough space to remember what we actually are.

Imagine what tomorrow has in store.

Phase Two: The Friction Experiment (Days 31-60)

The rules were simple. No email before 10am. One full Saturday of work that had nothing to do with a screen. One hour of unstructured time daily. One thing per week I'd been meaning to try but had been too busy to start.

I thought it would be easy. It was not.

The email rule broke on day three. Not because of a crisis. Because of a habit so ingrained I didn't notice I'd broken it until I was already reading a thread at 7:47am, phone in hand, the day having started in the old way before I'd decided how I wanted it to start.

What I eventually found on the mornings I held the rule was something small but real: the day felt like it had begun on my terms. Research from MIT Sloan on reduced-hour work models found that the benefit isn't simply the extra time -it's the restored sense of agency over how that time begins. The inbox would be there at 10. It was always there. But who decided when to open it turned out to matter more than I'd expected.

The manual Saturday was the one that surprised me most. Not because making bread or fixing a leaning bookshelf is profound -it isn't- but because the feedback was so immediate and so honest that it felt like a different category of experience than anything I'd done during phase one.

The bread I made on the first Saturday was dense and slightly wrong. I knew it was wrong before I tasted it. The bookshelf I repaired took forty-five minutes and now stands straight. There was no ambiguity in either result. No algorithm had mediated the feedback. No performance review would tell me in six months whether it had worked. The result was immediate, physical, and entirely legible. I had done the thing or I hadn't, and the thing itself told me which.

I tried to fix a bicycle derailleur on the second Saturday and failed comprehensively. Watched three tutorials. Failed again, differently. Took it to the shop. This was also fine, in a way that failing at a work task is not usually fine. Because the failure was clean. I knew exactly what I didn't know. There was no way to manage the appearance of competence when the derailleur was still broken.

By the fourth Saturday I was looking forward to it. Not for romantic reasons. Because it was a different kind of time -time that had resistance built into it, that couldn't be optimised away, that asked something of me and told me honestly whether I'd delivered.

The unstructured hour was the hardest part. Timothy Wilson and Daniel Gilbert's research -in which participants chose mild electric shocks over sitting alone with their thoughts- stopped feeling like an amusing data point and started feeling like a personal description. On day fourteen I reorganised my entire bookshelf alphabetically by author instead of sitting with the unstructured hour. Alphabetically by author. For an hour. Because it had a task structure and the emptiness did not.

I am reporting this without pride.

What eventually worked was giving the hour a loose container without filling it. A walk with no destination but a rough duration. Sitting outside with coffee and no phone but not requiring anything of myself except to be there. The brain needs scaffolding to learn how to rest. The total absence of structure was too much at once. The gentle container -a walk, a bench, a cup of something- made the quiet feel like choice rather than void.

The thing I'd been meaning to try was a documentary filmmaking course -online, eight weeks, one live session per week with an instructor based in Vancouver. I'd been interested in it for two years and had found, over those two years, that there was always something more urgent. Which is, I now understood, exactly the point. There is always something more urgent. The question is whether you let that be a reason forever.

The first assignment came back with detailed written feedback that I read twice. Not because it was praised -it wasn't particularly- but because someone had actually engaged with what I'd made, sentence by sentence, and told me what was working and what wasn't. I had forgotten what it felt like to make something specific and receive specific feedback on it. The AI tools I use give me feedback instantly and generically. This took five days and was about this thing, made by me, for reasons that were mine.

I thought about the difference for the rest of the week.

What the Two Phases Together Actually Showed

Here is the thing that the two-phase structure revealed that neither phase would have shown alone.

Phase one -thirty days of full 2098, Claude and Codex running the operation- is not bad. Let me be clear about that, because the tempting conclusion is wrong. The productivity was real. The quality of some outputs was genuinely better. The cognitive overhead reduction was significant and allowed me to think about bigger things more consistently. There are parts of phase one I've kept and have no intention of abandoning.

But phase one is also -and this is the part that only became visible from inside it- consuming in a specific way. It consumes the felt sense of authorship. The sense that the thing that happened, happened because of you, flowed through you, carries something of you inside it. Not in a precious artistic way. In a basic neurological way. The IKEA effect -the well-documented tendency to value things more when we've had a hand in making them- doesn't disappear because the tool is sophisticated. It attenuates. And what attenuates with it is something that, it turns out, I needed more than I'd known.

Phase two showed me what I'd been missing by showing me what it felt like to have it back. Not the manual work specifically -the manual work is just one delivery mechanism. What was back was the felt gap between where I was and where I was trying to get, and the satisfaction of closing it through my own effort. The bookshelf standing straight. The documentary assignment returned with five days of specific feedback. The hour of unstructured time that slowly, over thirty days, became less terrifying and more like rest.

The 4 Day Week Global research found that the benefits of reduced working hours increase over time as the pattern settles into the body. Phase two felt like that. Not dramatically different on day one. Measurably different by day thirty. The centre of gravity had shifted slightly. The shift was real.

Threading the Needle

The two phases together allow me to make an argument I couldn't have made from inside either one alone.

Phase one -full 2098, Claude and Codex running the operation, Make.com stitching it together- is not the problem. Let me be clear about that because the lazy conclusion is wrong and I've been at pains not to write it. The productivity was real. The cognitive overhead reduction was significant. The quality of some outputs was genuinely better than what I'd have produced alone. The free time it created was real free time, spent on real things that mattered to me.

But phase one also showed me, from the inside, what exclusive residency in 2098 does to the felt sense of authorship. The work that the machines touched felt slightly less mine than the work they didn't. Not wrong. Not bad. Less mine. And that distinction, small on any given day, compounds across weeks and months into a question worth taking seriously.

Phase two -the deliberate friction, the documentary course, the manual Saturdays- showed me what I'd been missing by showing me what it felt like to have it back. Not the manual work specifically. The honest feedback loop. The gap between where I was and where I was trying to get, closed by my own effort, confirmed by something that didn't have an incentive to be kind about it.

The needle threads here: use 2098 fully and without guilt for what it does well. And then keep -deliberately, as a practice, not a retreat- some domain that isn't 2098. Something that pushes back honestly. Something where the result tells you directly whether you delivered.

For me it was a weedy lawn and a documentary course and two hours of video games on a Tuesday I hadn't earned yet by the old metric but had more than earned by the new one. For you it will be something else. The specific domain doesn't matter. The honest feedback does.

The lawn is better. Not fixed -there are still patches that are losing- but measurably better than it was two months ago. I'm still making videos I'm not good at yet. The Tuesday evenings still sometimes belong to something other than work, and the guilt that used to accompany them is mostly gone because the machine layer has changed what "productive" means when it can no longer mean "busy."

That's the thing 2098 turned out to be, approached from inside rather than described from outside. Not a destination. Not a threat. A condition -created by the machines, occupied by the humans- in which the oldest human questions finally have the time and space to be asked.

What do I actually want to do with an evening?

What am I making that is mine?

Who did I forget to call?

Imagine what tomorrow has in store, when the machines are even better at the machine part, and we have even more time for the human part.

I think I'm going to need a bigger lawn.

The machines got better at the machine part. Turns out I got better at the human part. Nobody told me that was how this worked.

Tools Used in This Experiment

Claude (Anthropic) — reasoning, communication drafting, writing assistance, context synthesis
Claude Code — agentic coding, automation scripting, connector debugging
Cowork by Anthropic — desktop file and task automation integration
OpenAI Codex — workflow orchestration scripts, Windows connector logic
Make.com — the automation layer connecting all tools to email, calendar, and files
Make.com Windows Connector — file system and application integration (the one that broke seven times)

Sources & Further Reading

Wilson, T.D. et al. (2014). Just Think: The Challenges of the Disengaged Mind. Science, 345(6192). Read the paper →
4 Day Week Global (2023). Cited in Fortune: Is a Four-Day Workweek Just as Productive as Five Days? Read the story →
Norton, M.I., Mochon, D. & Ariely, D. (2012). The IKEA Effect: When Labor Leads to Love. Journal of Consumer Psychology. Read the paper →
MIT Sloan Management Review (2025). Why Four-Day Workweek Experiments Fail. Read the article →

Your Questions, Answered Plainly

Should I try the Claude and Codex setup myself? Yes, but budget a full week for the chaos before the calm. The first three to five days of building any AI operational layer are debugging days -wrong prompts, broken connectors, authentication loops, the Make.com workflow that almost works but doesn't. Don't judge the system by week one. Judge it by week two, when the calibration has settled and the machine layer actually runs. The Claude Code documentation and Make.com's connector library are your starting points. Expect to iterate on every prompt at least four times before it does what you actually need.

What did the free time actually feel like? Disorienting at first. Genuinely good after about three days. The guilt that usually accompanies non-productive time was absent because by any measurable standard the productive time had already happened. What filled the space was, in order: video games I'm bad at, YouTube documentaries I'd been saving for a future self with more time, making videos of my own, longer evenings with family, and a lawn that had needed attention for two years. None of it was what I'd have predicted. All of it was exactly what I needed.

Is the counterintuitive thesis -more AI means more human- actually true? It was true for me, over six weeks, in Ottawa, with this specific configuration of tools. Whether it generalises depends on whether you actually use the time the machines create for something other than more machine-mediated work. The risk is that the freed capacity immediately gets filled with more professional output -more content, more projects, more productivity- and the human part never gets the space it needed. The machines can create the condition. They can't make you use it well.

What's the single most useful thing the AI layer did? Quieted the background hum. The persistent low-level awareness of things undone, threads unresponded to, logistics unmanaged -that hum is more cognitively expensive than it seems until it's gone. When the machine layer absorbed enough of it, the evenings changed quality in a way that's hard to describe precisely but was immediately noticeable. It wasn't more time. It was quieter time. That turned out to matter more than the extra hours.

You said 2098 isn't the problem. So why does this series keep sounding like it is? Because the series is written for people already living primarily in 2098 -which most readers of inthacity.com/blog/tech/ai/ probably are. The argument isn't against the tools. It's for keeping something alongside them that the tools can't provide. Not instead. Alongside. The lawn. The videos. The Tuesday evenings. That distinction is the whole thing.

Read Part 1 of this piece, and follow the full series at inthacity.com/blog/tech/ai/.

Disclaimer: This article may contain affiliate links. If you click on these links and make a purchase, we may receive a commission at no additional cost to you. Our recommendations and reviews are always independent and objective, aiming to provide you with the best information and resources.

Get Exclusive Stories, Photos, Art & Offers - Subscribe Today!