Family Fun Quest + Financial Quest + Thriving Wage in SF

Hey hey,

The last three weeks have been an emotional roller coaster. I’ve been modeling what it actually costs to raise a family in San Francisco, trying to be realistic and pragmatic at the same time, and it’s daunting.

I’m not talking the lifestyle of the rich and famous. Wholesome basics like fiscal responsibility (short, medium, and long term savings), education + enrichment opportunities for kiddos (daycare, good school, music teacher, summer camps), reactive + proactive healthcare, and a small sum for one international and one national vacation a year. In one of the richest places on earth, to do all of those things one has to be … well rich. It makes me wonder about which of those categories the vast majority of Bay Area dwellers are foregoing to live here. It’s inspiring, in a sort of spontaneous combustion under my derriere way, and I’m mostly just sitting with seemingly impossible math.

In between the token burn runs: an EPIC seal squad river adventure with friends, a stretch of family time, and more camping on the horizon. More river than screen is a decent antidote.

In motion

Thriving Quest #003: Family Fun

I never sent this one, so here it is. There are far more great things to do around the Bay Area than our family will ever find or remember at the right time. Discovering them means searching scattered calendars, filtering out obvious mismatches, and coordinating tickets and childcare before the opportunity disappears.

So I built Family Fun, an agent-assisted possibilities calendar that watches trusted event sources, applies our family’s preferences and constraints, and puts the best options somewhere my wife and I will actually see them, without blocking time on the real calendar. The target: one fun family event a week and one culturally stimulating date night a month. The goal is less hunting and coordinating, more adventures together. I want agents to create shared anticipation, not another family inbox.

The honest status is that the technical proof is ahead of the family proof. We have a run of trips coming up, so the weekly surfacing has mattered less lately. Build notes here.

Proactive Family Finances

Ahh, my least favorite roller coaster. So important, and so hard to stomach.

Three words now organize the whole system: a forecast is what happens if our behavior doesn’t change, a budget is where we deliberately commit to change it, and actuals get compared to both. Two different signals fall out: drift (did life go as expected?) and discipline (did we hold the line?).

The first production forecast covered September through December, and it matched the rollup I’d built by hand to the cent. Then my wife and I sat down and committed our first real budget against it. The rule: any line that differs from the forecast must carry one sentence saying why. I’m hopeful those sentences will feed my learning engine for iteratively improving my methodology, and that’s proving surprisingly difficult to architect gracefully. No great priors I’m aware of.

Frankly, the first showing to my lovely lady was a lukewarm success (as intended). I needed to push something “good enough” and not allow the perfect to be the enemy of the good. The forecasts were close in some months and off in others, and the savings picture is still missing from the screens. October is the first month where drift and discipline can actually be measured.

And the exercise has already proven useful to inform our spending decisions for Sept. Eat down the pantry a bit. Restaurants once a week each. Keep the rideshares down. And also to inform our shorter term planning for bigger life choices. Just wish it wasn’t so daunting here …

Full notes in the quest log.

The Thriving Wage

A living wage (opens in a new tab) is a calculator developed by economists at MIT defining what it costs to survive in a place. I am extending that methodology to a Thriving Wage calculator for what it costs a family to participate in the economy, save reasonably, and provide meaningful experiences for kids and adults. Of course, my first pass was for San Francisco: live, participate, stabilize, and thrive, layer by layer. Testing this on our family is a big part of the roller coaster above.

I look forward to refining it and publishing it with more data sets soon, so it can be a lot more useful to you.

Agents in the cloud

I deployed my first agent to Vercel’s Eve framework this past week and have really enjoyed it. It’s a real step up in admin overhead to build and maintain, but it unlocked a lot of room for refinement.

At work, I found that an open-source model can run one of my agentic loops for 1/17th the cost with only a marginal quality tradeoff. To get there I had to build a golden eval set and run it repeatedly with prompt tweaks until the cheap model reached parity with the expensive one. It took real technical know-how and a pile of tokens. But 1/17th is 1/17th. I’m going to pour more of my recurring loops and cron jobs into this framework to bring costs down and free up token budget for my own frontier work.

On my mind

It’s really easy to forget how deep in the AI echo chamber I am. Andrej Karpathy’s recipe for becoming an expert (opens in a new tab) has three steps: take on concrete projects and learn on demand, teach or summarize everything you learn in your own words, and only compare yourself to younger you. The middle step is exactly how I feel about working with AI. It’s so fun to be at the epicenter, and it is absolutely an echo chamber. Organizing my thoughts on Thriving with AI well enough to share them with you is a big part of my joy in writing today.

Taking in

33 Questions Executives Ask About AI (opens in a new tab) by Natalia Quintero and Mike Taylor at Every. I highlighted more of this than anything I’ve read in months. The line I keep repeating: it’s easier to raise the ceiling than the floor. One fluent person building ten reusable skills beats ten people each building one, and skeptics follow their peers out of practicality. That’s invitational excellence. My one disagreement: I would not pick a single platform and lock in right now. Build model-agnostic from the start while the models are changing this fast.

Has Anthropic solved the peptide-binder design problem? (opens in a new tab) by Claus Wilke. The release of Fable 5.1 came with a lot of fanfare about scientific progress, and my brother was kind enough to send me this teardown from a domain expert. I love Wilke’s test for any claimed breakthrough: can you explain the core new idea in two or three sentences? If not, be skeptical, because people are great at confusing themselves and AI is exceptionally good at finding loopholes that satisfy the letter of a problem without doing what you meant. Spoiler: the answer is that Fable did a good job on a known problem with known solution paths. This may have positive impacts on the field, maybe, but it’s far from a breakthrough. And the bottleneck is the slow, complex “wetware” problems of a physical lab testing the ideas that come out of computational modeling.

Part 1: My Life Is a Lie (opens in a new tab) by Michael W. Green. A while back, my brother sent me this article when I was grumbling about living costs and it’s shaped a portion of my Thriving Wage concept. Green takes a number he had accepted his whole life, the poverty line, recovers where it came from (fascinating!! and so broken today …), and rebuilds it from what a household actually has to buy today. I appreciate his callout of a “participation ticket”: the costs that are technically optional but functionally required to work, parent, communicate, and belong. That’s why the calculator has a Participate layer sitting between the basics and the future.

Until next time,

~h