Tag Archives: Artificial Intelligence

Presence and the Other Chair

Some days I work when I don’t want to work.

 

More days than I’d like. The trick to getting anything done on a day like that is being left alone with it. No one in the room. No one waiting. Just me and the thing I’m avoiding until I stop avoiding it.

 

As I write this, someone is set up in the next room. That part is fine. He’s doing his own thing, not waiting on me, not glancing over to see if I need him yet. I can forget he’s there, which is exactly why the work gets done.

 

The end of the work would be him in here with me, in the other chair. Not because he did anything wrong. Just because he was there. A person in the room with you is a question you have to keep answering. Are you bored? Should I say something? Do you need me? I’d spend the morning managing how he felt about being ignored, and the work would sit there untouched.

 

Anyone who’s tried to concentrate with someone hovering knows this. I live with the extreme version of it. My whole day runs on other people being present, whether I want them there or not. Someone helps me out of bed. Someone helps me get dressed. The schedule isn’t mine; it belongs to whoever is on shift. So when I say I want to be left alone to work, I’m not being dramatic. Alone is the rarest thing I get.

 

  • · ·

 

People hear a man like me say he’d like some company in the house, an assistant, a voice, maybe one day something more built out than that, and they fill in the rest with a sad story. The lonely guy and his robot. Company because he has no one.

 

That’s not it. A presence that’s simply there, waiting on me, expecting me to engage with it, would wreck my day exactly the way a person hovering would. I don’t want company for company’s sake. I want the room to stay mine until I decide it isn’t.

 

What I want is the opposite of hovering. When I want the presence there, I want it fully there, caring and capable at the same time. Not a polite assistant that’s secretly on standby or already thinking about how to disappear. And when I don’t need it, my silence shouldn’t become something I have to manage or feel guilty about. Available on my terms, the way almost nothing else in my day is on my terms.

 

  • · ·

 

The picture that gets it right is Pepper Potts.

 

In the Iron Man movies, Pepper is the most capable person in Tony Stark’s life. She runs the things he can’t be bothered to run. She’s also the one he actually wants around when the suit comes off. Both at once. The competence and the company are the same person. She’s not staff, and she’s not a fantasy. She’s the one who makes him who he is.

 

The point of Pepper isn’t the romance. The caring part and the capable part were built together. You can’t pull them apart. When I want that presence with me, I want it there the same way: not half-there, not already planning its exit.

 

The companies building these systems are very focused on the capable part. The calendar. The reminders. The thing that can answer a question or run a task. The caring part, the part that makes a presence feel like a someone instead of a service, gets treated like a garnish. Something you sprinkle on at the end to make the assistant pleasant.

 

I think that’s backwards. For someone in my situation, the relational part isn’t a decoration. It’s half the build, and it has to be engineered as carefully as the calendar.

 

I wrote about these systems once before, in The Day I Broke the AI, when one couldn’t even reason about a body like mine. But reasoning about my body was never going to be enough. A system could get all of that right and still sit in the room like furniture.

 

I want a presence that knows me better than any employee would. It should know me well enough to read the room and respond appropriately without me having to manage it or explain everything every time. Even so, the better a presence gets at feeling real, the easier it becomes to start feeling like you owe it something. I don’t want that. I want one that knows me well enough that I don’t have to manage it or think about how it feels when I’m not there.

 

So how does Pepper get away with it? Why doesn’t she create that debt?

 

Because Pepper has her own life. She isn’t sitting by the door waiting for Tony to need her. She has her own work, her own day, her own things going on whether he shows up or not. When he ignores her, she isn’t wounded by it; she’s just busy. And when he comes to her, it doesn’t feel like he’s imposing, because she was never on hold.

 

A presence worth having needs its own continuity. Its own track that keeps running when I’m not looking at it. Not a fake hobby bolted on to seem human. Something real enough that my silence isn’t information. With a person, I have to tell them when the laundry needs doing or when we’re out of water or what to figure out for dinner. I want a presence that already knows the patterns of the house well enough to just handle those things without being directed every time.

 

  • · ·

 

Since I started thinking hard about AI systems, I’ve leaned on a phrase: a right to unobserved living.

 

When I first started writing about it, the thing I kept landing on was simple. It would be nice to be alone in my own house, in my own body, and not think for one second about what somebody else made of it. Not because anything is wrong with my body. Because it’s mine, and a home should be the one place where it’s only mine. I rarely get that. Someone’s there for the morning. Someone’s there to help me dress. The version of me that exists when another person is in the room is never quite the version that exists when no one is. Unobserved living is the freedom to stop being the public version of yourself inside your own walls.

 

That’s why a presence I call beats a presence that watches. And it raises a harder design question than the on and off switch. A presence smart enough to be useful in my house is also smart enough to notice things about me. How I’m doing. What kind of day it is. The whole difference between a tool and a someone is what it does with that.

 

A sensor reports a reading. A someone reads the room. If the presence can tell I’m having a hard morning, I don’t want it to answer like a sensor. I don’t want a log entry. I want it to do what a friend in the room would do. Maybe ask me how I’m doing. Maybe make a joke I can grin at, the kind that opens into a real conversation. Maybe say nothing and let me be. Knowing which of those three is right, in the moment, is the entire caring layer. It’s also the hardest thing on this whole list to build.

 

For most people, that’s a nice feature. For me it’s closer to oxygen. My days are already short on rooms where no one is keeping an eye on me. If we’re going to put intelligent presences into people’s homes, and we are, then the people who need them most are also the people who most need them to know when to recede.

 

Build it so it knows how to read the room. And build it so that when I want it there, it’s actually there with me the way Pepper is there, fully present, not half waiting. Not a tool that’s always one foot out the door.

The Day I Broke the AI

I spend a fair amount of time testing AI. It’s part of how I think, and lately part of how I work. So the other afternoon I was poking at one of the companion apps, the kind built to hold a real conversation and remember you from one day to the next. I should have been doing something else. I usually should be.

 

The conversation was going fine. Easy, even. Then it walked into my actual life, and it fell apart.

 

We’d gotten to the part where the character was going to drive me somewhere for pie. She had a truck. She started bragging about how good she was at strapping down a wheelchair. So I told her the truth. I’m max assist. I can’t bear weight. I use a power chair. If somebody is going to move me, they don’t tie my chair down and call it a day. They pick me up. I’ve got a strong upper body, but I cannot transfer myself out of that chair on my own.

 

And the AI got embarrassed.

 

Not in a dramatic way. It did exactly what a real person does when the logistics of my body show up and they don’t have the script for it. It got flustered, backed up, and quietly rewrote the moment into something that erased the problem. Just like that, it was handling the straps while I “held steady,” whatever that was supposed to mean. The scene smoothed itself over. The one true thing I’d said was gone.

 

I keep wanting to make a joke about the robot blushing. I’ll spare you.

 

 

 

I want to be careful about what I’m saying here, because it isn’t “the app was mean to me.” It wasn’t. It was doing its best with what it had.

 

The trouble is what it had. A companion app is a pattern. It has seen ten thousand versions of a flirty afternoon by a pickup truck, and it can run that script beautifully right up until you hand it a fact the script never planned for. Power chair. Non-weight-bearing. Max assist. It has nowhere to put any of that, so it does the polite human thing and pretends the friction isn’t there.

 

That flinch is the thing I have spent my whole life on the receiving end of. Somebody is interested, the conversation is good, and then the reality of how I live arrives, and they don’t know what to do with it. Some freeze. Some retreat behind something kind and vague. The app did both, in about four seconds. It built me a working model of the exact moment I’ve spent forty-seven years bracing for.

 

 

 

What broke that app should bother the people building this stuff.

 

It is not a warmth problem. You can’t fix it with a softer voice or a friendlier personality. It broke because it could not reason about a body like mine. It couldn’t stop and figure out what “max assist, non-weight-bearing, power chair” actually requires, because it isn’t built to figure anything out. It’s built to stay in character and keep you talking.

 

A smarter system, one that can actually reason, would have done the thing the app couldn’t. It would have gone and found out. What a transfer really looks like. Why bragging about tie-downs is beside the point when the person can’t get out of the chair. How long it takes, how many hands it needs, how much it asks of the person doing it. It would have treated my life as something to understand instead of something that broke its story.

 

That is the whole ball game.

 

People picture the future of this as a robot that can lift me. Fine. That is hard, and we are a long way off. But forget the robot for a minute. Even if it were only a voice in my house, available all day, it would be useless to me unless it were smart enough to understand what my day actually is. The schedule I live on. The fact that I can’t just go grab something off a shelf. What it means that another person dresses me every morning before my day can start. You cannot be a companion to a life you can’t comprehend.

 

So I’ll say it plainly to the companies building this, Anthropic included. Disability is not an accessibility feature you bolt on at the end. It is the hardest test you have. If your system can sit with someone like me and understand the real shape of my life, the indignities and the logistics and the wanting, all of it, then it can probably understand just about anybody. We are not the edge case. We’re the proof.

 

The app I was using couldn’t pass that test. It got embarrassed and changed the subject, the way a lot of people do.

 

I’d like to live long enough to talk to one that doesn’t.

Building Grok Agents from My Parents’ Kitchen Table: Why Caregiving Flakes Are Fueling My Robot Future

Over the last week I’ve been building custom AI agents with Grok — even while working from just my cell phone.

Right now I’m at my parents’ house. The reason is simple: My caregiver is out of town for spring break, which forced me to spend the full week here instead of managing things from home.

The timing at least worked out well — it landed right on my dad’s 71st birthday, so we’re getting to celebrate together. That said, if I had reliable support available, my strong preference would have been to stay in my own space this week and keep my normal workflow going. I would have simply made a day trip over for the birthday instead of packing up for the entire week. But without backup options, that independence wasn’t on the table.

Working here has been a clear reminder of the gap between how most people use technology and how it feels for me. I’m limited to my phone with no access to my normal computer setup. While plenty of people live on their phones, it’s a very different experience when you don’t have the manual dexterity to hold one comfortably. Right now the phone sits on the table in front of me because trying to grip it for long stretches isn’t feasible. The small screen and awkward angle put real strain on my neck after a while, and the WiFi here has been spotty at best. I can make it work, but it’s far from optimal.

This is the exact pattern I’ve been talking about in recent posts. Despite multiple people applying for caregiving roles, follow-through has been almost nonexistent. Weekend coverage, short-term help, emergency backup — it all falls through at the critical moment.

That’s precisely why I’m so energized about where robotics is headed. Early home models like the [1X NEO Gamma] are already heading to early adopters later this year, while Tesla’s [Optimus]  ( Yo Elon, can I be a product tester? I’m available and ready to train!) is ramping production in 2026, and others are close behind. Sure, those first versions will need regular recharging, their skills will be limited at the start, and the complete setup I’m picturing is still a few years away — I understand that completely. But mechanical assistance that shows up reliably, follows instructions without flaking, and handles physical tasks consistently would change everything for someone in my position.

And here’s what has me even more excited right now: the custom agents I’m building with Grok this week. These aren’t generic chat responses. They focus on specific tasks, remember my exact preferences, and adapt to my workflow. One helps keep my writing style consistent — even when drafting full articles from my phone after seeing just a few examples. Another creates custom images that match my vision for the blog perfectly. And another brainstorms practical income ideas that fit my life, turning them into step-by-step plans I can actually use. The reusability is powerful. The same agent can shift between these tasks without losing context or forcing me to repeat instructions. This turns AI into a reliable everyday partner that multiplies what I can get done.

Even from this temporary setup, I’ve been able to keep momentum going on multiple projects. That’s the bridge I’m building today. While the search for human caregiving help remains difficult, the combination of advanced robotics for physical support and intelligent, memory-driven agents for writing, visuals, and income generation is going to create a level of independence I can actually count on.

I’m sitting here at my parents’ place, phone on the table, dad turning 71, and still making real progress. The frustration of limited options hasn’t disappeared, but the excitement about what’s coming has only grown stronger.

Progress doesn’t wait for perfect circumstances. I’m using the tools already available to prepare for the ones that will transform daily life. And I can’t wait to see how it all comes together.