We had planned to take our five-year-old to the free Darth Vader skate down at the waterfront but the weather was not cooperating.
It's the kind of pop-up event we find buried in our group chats or newsletters: a free outdoor sakting event in Brooklyn Bridge Park, costumes encouraged, and, since our son had just discovered Star Wars, the kind of thing that makes you feel like you're winning at parenting if you actually make it there. We had the bag packed. Our son had his lightsaber... But.... By 11am, the rain settled in for good.
So, I opened an email my mvp product / hack / AI bot sent that morning.
I've been hacking the email for about a month and calling it "Family Friendly", a weekend planner to help two tired parents get out of the apartment on the weekends and explore their city. Family Friendly consists of five listings, picked from a corpus of nine hundred events I've been scraping from the city's cultural edges. It's still rough, but in a city so dense, with parents so exhausted, it becomes a real pain point for us and we find ourselves repeating stale patterns.
So my wife scrolled through it, paused on one of the five ideas, looked up, and said "let's do the museum."
Twenty minutes later we were paying what we wished at the Brooklyn Museum. We spent the day with other families who'd clearly made the same calculation we had. We were off our phones, with our kid, in a building full of art. We were making arts and crafts in a free workshop - art that's still displayed our fridge, actually.
There's so much AI hype and hate right now that I've largely tuned out both, instead opting to hack on a handful of AI-enabled products for fun the past few months to figure out for myself what kind of product AI makes possible.
Two personas, one need
The standard model for personal AI right now has two modes. There's the assistant, which waits for you to ask it something and then dutifully answers. And there's the agent, which goes off and does things for you while you watch.
Both of these add to your queue.
The assistant adds to your queue by giving you options to evaluate. Twenty different ways to phrase your email. Pick one. Cognitive load goes up along with decisions fatigue.
The agent adds to your queue by doing things in your name that you now have to monitor, edit, or undo. Did it book the right flight? Did it create the right email draft? Are you sure that's what you wanted? Cognitive load: still up.
The category I feel like we're currently under-indexed on is help.
Help is what a great chief of staff does, what a friend who knows you well does. It's what a coach does. It takes weight off, instead of adding more. It makes you better at the thing you're trying to do, not a passive recipient of work done in your name.
Help has three properties.
Unburden. Focus. Empower.
Help operates on three axes: load, attention, agency.
It unburdens. The assistant says: here's the report you asked for. Help says: I took that off your plate. Help is what hands you one good plan and gets out of the way.
It focuses. Most digital products want to expand your options because options feel like value. Help points at the one thing or 5 incredible choices with you as final decider.
It empowers. An agent takes the action from you; help makes you better at taking the action yourself. One replaces you. The other amplifies you.
Help makes you more, not less. More agency, returned attention, less load. AI that operates this way feels different from AI that performs for you. It feels like a peer who knows what you're trying to do, like power steering or driver assist.
Coaching - a working example
I'm currently a startup coach and I was wondering how AI might affect the practice last year. I thought: 'hrmm, like all good technology, it'll make it cheaper (free), always available when you need it (on the internet), democratized (for all) and with access to the best thinking in leadership development'.
So last year I built a product idea called Ed Bot with my friend and executive coach, Ed Batista: it's a custom GPT on OpenAI's platform, designed to be a thinking companion for founders and operators. It doesn't tell coaches what to do. It doesn't take action on their behalf. It asks questions Socratically. It surfaces frameworks based on 1,000,000 words Ed's written on his blog over the past twenty years on leadership development. It's awesome and I find myself using it to surface relevant articles Ed's written for my clients.
In the past year, it's had 100,000+ conversations across 8 versions and earned 4.9 stars from folks who, professionally, are a very tough audience.
I believe this is the help posture. It unburdens: the founder doesn't have to hold every framework in working memory. It focuses: the right question at the right moment. It empowers: the founder is still the founder, they're just better-informed.
Here's a few takeaways:
Help-shaped AI can work at scale. This isn't a thought experiment. It's been operating on the largest LLM platform, across more than a hundred thousand real conversations.
It works across stacks. Ed Bot runs on OpenAI now. The chief-of-staff system I'll describe in a minute runs on Anthropic. The version of either you'd rebuild on next year's models could run wherever. What carries across stacks is the knowledge base, not the runtime. Help becomes the product value, not the platform.
The kind of audience matters more than the number. Founders are a cool canary for the question, "does AI augment the human, or replace them?" They're acutely conscious of the boundary in their own work and in tools they evaluate. Each week Ed receives an unsolicited review from someone who has found the virtual version of him and been impacted. Some of his clients even bring their conversations with the bot to their sessions. (Some of mine too).
Calm tech and cultural products
There's a strand of technology thinking that goes back to Mark Weiser at Xerox PARC in the early '90s. He called it calm technology. The idea was that the best tools shouldn't capture your attention, they should sit at the edges of it, and recede. Calm technology was a critique of the personal computer's tendency to demand everything you have.
That critique applies even more to phones, and even more sharply still to today's AI products, which are largely racing to be the thing you spend your day inside of. At the consumer level, the business model is still your attention.
I don't think we want more of that. I don't think we need more of that. I think we need less of that, to be clear.
The Family Friendly email took thirty seconds to read and got us out of the house. It pointed at the Brooklyn Museum, which had a free craft workshop, and which my son will remember as the place his cool collage came from. The product worked precisely because we didn't spend the afternoon with it or our devices. We actually spent the afternoon with each other, at a cultural institution, making and being inspired.
What if personal AI makes us into more interesting people, not just more efficient ones?
Think about some non-AI products that do this naturally. Letterboxd makes its users into film critics. They watch differently, write differently, talk differently, because the product is shaped to reward that. Strava makes its users into athletes. Goodreads makes its users into readers. The category isn't "saves you time." The category is "makes you the kind of person you strive to be."
Help-shaped AI could do the same. Family Friendly, if it works, would make its users into more engaged New Yorkers, better connected to the city (and to each other). The category is "makes you more of a citizen," not "saves you fifteen minutes."
That's what calm tech and cultural products have in common: they're not asking for your attention, they've giving it back and they're shaping who you become.
Two new bets
Ed Bot was one kind of proof, here's what I'm hacking on in the same vein now.
The first is Kubrick, my open-source chief-of-staff stack. It unburdens by capturing context as I talk, so I don't have to keep it in my head. It focuses by surfacing what I should do next, instead of presenting twelve options. It empowers by leaving me as the decider, but a better-informed one. The arc is public: four stages, build-in-public, an essay for each.
The second is Family Friendly, the email I described above. The consumer-facing version. One user, a corpus of nine-hundred-and-something events, a weekly Saturday email that picks five things, recommends one, and is short enough to read while you're making coffee. And for the nerd in me an MCP server so anyone can start hacking on this.
Ed Bot for founders. Kubrick for any professional. FF for parents in their city. All three are experiments asking whether personal AI can sit at the edges of your attention and make you better, instead of in the middle of it making demands.
When google launched it was just a text box (backed by thousands of servers and proto AI) to a consumer. Family Friendly is, in form, simply a newsletter. Newsletters are an old, familiar format. What makes this one different is that it doesn't exist without a corpus, without synthesis from the corpus, without a model that knows it's raining outside and that we have a five-year-old and that the carousel is closed today. Newsletters used to be authored. This one is curated, in something close to real time, by an instance of help. I also just think this is the right channel to start in, it's familiar. MCP server, messaging, alerts and real time personalized conversations could easily come next.
While I'm playing around with novelty, I'm simply trying to take an old format and find its new shape.
A love letter to the city
I started Family Friendly where I live because New York City is the inspiration for so much of my life. Foursquare did something like this twenty years ago, it turned check-ins into a relationship with place. ITP, the program at NYU where I went to grad school, treated the city as the canvas for everything we made. There's a long lineage in this town of building technology that puts you back into physical contact with where you live.
I want Family Friendly to do that, eventually for more cities than New York, but first for my neighborhood and then, school district by school district, for all five boroughs. It's meant as a love letter as much as a product.
That's the kind of AI I'm interested in building. Help-shaped. Calm. Specific. Connective. Memory making. The kind that gives you your weekend and your family. The kind that takes you to the Brooklyn Museum on a rainy Saturday and lets you forget, for a few hours, that any of this software exists at all and leaves a project stuck to your fridge to remind you.