Autumn 2026How this issue is madeIssue 01
Product manager, data scientist first. Rental applications and payments at AppFolio, data at Oracle and Lively. Now helping small companies put what’s new to work.
The application nobody could finishRebuilt end to end
Native Plant FinderFree tools your agent can run
Applications and payments at AppFolioData at Oracle and Lively
44% to 70%Applications
submitted
Cover: marigolds and hanging amaranth, engraved for this issue.
Product manager, data scientist first. Rental applications and payments at AppFolio, data at Oracle and Lively. Now helping small companies put what’s new to work. I thrive in the unglamorous middle: the customer calls, the half-clean data, the code that has to actually run. I build free tools that AI agents can run.
44 to 70%
Source AppFolio. Started rental applications that were submitted, old form against the rebuilt flow.
3.4days
Source AppFolio. Properties on the rebuilt application, against those still on offline workflows.
99.1%
Source AppFolio. Application-fee checkout, with an assumed-success window and a recovery path for failed payments.
$50M+
Source Lively, Nov 2019 to Dec 2022. From major financial institutions, through APIs and migration flows I built.
6tools
Source llms.txt, which lists every tool and how to run it.
Submissions went from 44% to 70%, and vacancies filled 3.4 days faster. Here is what changed.
The old rental application was stuck. Applicants couldn’t save the form or come back to it. Property managers had no way to see what was working, and every applicant was anonymous.
I rebuilt the experience so it felt like a modern app, added accounts so applicants could finish later and we could measure what worked, and built reporting so property managers could finally see what was happening with their applications.
The result: submission rates went from 44% to 70%1 with a zero change-failure rate,2 and properties on the new flow filled vacancies 3.4 days faster than offline workflows.3
Of every 100 applications started
44 Old form
70 Rebuilt flow
26 more of every 100 started applications were submitted on the rebuilt flow.
College recruiters rarely see athletes who grow up far from the big metro clubs and showcases, no matter how good they are. My brother and I are building Brava for those players: women’s soccer recruiting profiles with coach-verified stats, benchmark rankings and an AI-assisted production pipeline. I design and ship it end to end, from the first research call to the brand on the page.
An early-stage startup with no data science function and no reliable way to explain its data model to financial partners. I built the APIs and migration flows that moved over $50 million in assets from major financial institutions onto our platform. The work grew into product management as the company scaled.
When you change a model, how do you know it made things better? I built a system to run hundreds of evaluations against model updates, so we could point to real statistical significance instead of gut feelings. Concerns about where the data came from eventually led me to leave.
Three steps for building products that fit how people actually think.
“Property managers don’t want an ‘AI Property Manager.’ They want to stop losing Sunday nights to spreadsheets.”
I start with the customer’s real problem, not the technology. Property managers don’t want an “AI Property Manager.” They want to stop losing Sunday nights to spreadsheets and just know their properties are compliant. I keep asking questions until I can name the exact moment someone would pay $500 right now to never do that manual reconciliation again. Every decision after that points back to that one frustration.
Before software can act for someone, it has to understand their world. I map the things customers care about (a Lease, a Security Deposit, a Compliance Rule) and how they connect. When a Lease expires, what should happen to the Deposit? That logic is the heart of the product. It lives in a real model of the world, not in a clever prompt.
I turn that map into a written spec: the job to be done, the pieces involved, how they fit together, and the rules for what “good” means. When the job is compliance, “good” means checking against the real law, with no guessing. I pressure-test the spec before any code exists, then do the same with the plan and the work itself, breaking it into small pieces and checking each one against the spec.
The result is software people can trust with real work, built to fit the way they already think.
Free tools. No accounts. Your AI agent can run them too: each one takes URL parameters and registers WebMCP tools.
Find native plants for your garden based on your location and sun conditions. Support local pollinators while reducing water and maintenance.
Open Native Plant FinderRate eight areas of your life and see which one to invest in next. Then print the matrix as a one-page guide.
Open Life Strategy MatrixWeigh your options with clarity. Score choices against what matters most.
Open Decision AssistCreate a simple, dignified pamphlet for a memorial service.
Open Memorial Pamphlet CreatorBuild a liturgy for a memorial or celebration of life.
Open Service of Life BuilderA one-page form to print and fill out by hand. Record service wishes on paper, no screen required.
Open Memorial Preferences WorksheetA plugin for the Hermes agent that keeps it on today’s best free model, smoke-tested daily with a real tool call, with a local model as the fallback.
AbiertoClaw on GitHubHow to run every tool here: prefilled links, WebMCP, or the data directly.
Read llms.txtWords I’ve put together about life, memory, and paying attention. Some poetry, some prose, some longer stories.
My daughter has my dark circles. The same ones I studied in mirrors as a boy and wished away. On her face I feel only tenderness, and something in me settles: what I carried alone, she won’t.
I used to treat being a good dad as a knowledge problem. I’ve modeled my kids’ genetics down to the allele, the coin flips and the certainties, because the math is the one part I can’t change. So I’m changing everything else: the garden, the food, the sunscreen, the early bedtimes. But kids can’t read protocols. They read faces.
The two most durable things I’ll ever make are asleep down the hall.
Read the essayHow this issue is made. Every value comes from theme.json, compiled to tokens.css. Switch the edition in the masthead or the menu to see both.
Orange, acid green and a dash of black.Orange, acid green and a dash of black. The bands are sized to each colour’s share of a page: Gondola is the ground, orange and green are fields.
| Role | Night | Contrast |
|---|---|---|
ink / ground | #EFE8DC | 10.7 · 11.4 |
ink-quiet | #B8AFA3 | 6.0 · 5.6 |
spot small text | #FF8460 | 5.4 · 5.1 |
spot-large 24px+ | #FE5F35 | 4.3 · 5.1 |
leaf | #CAD576 | 8.3 · 5.1 |
on-spot-fill | Char on orange | 5.6 |
on-leaf-fill | Char on green | 10.8 |
on-leaf-fill-quiet | Moss on green | 6.7 |
outline controls | #A0978C | 4.5 · 4.2 |
spot-wash chart | #503730 | 8.9 · 10.1 |
Table 1 Every pair is checked against WCAG AA by the test suite. Night first, then Day.
Fraunces, Newsreader and Instrument Sans.Fraunces for display, Newsreader for reading, Instrument Sans for anything you press or scan. Weight and width carry the hierarchy, not size alone.
Ben Siverly
Nameplate · Fraunces 800 · Optical size 144The application nobody could finish
Display · Fraunces 640 caps · Soft 100Map the world first
Headline · Fraunces 360 · Soft 503.4d
Figure · Fraunces 300 · Soft 0Nobody claps. That’s how I know it’s working.
Pull quote · Fraunces italic · Soft 100Native Plant Finder
Title · Fraunces 480 · Soft 30Applicants could not save their place. Here is what changed.
Deck · Newsreader italicI care about shipping things that actually work, staying close to my family, and building a few things that might outlast me.
Body · Newsreader 17px · 2 columnsProperties on the new flow, compared with offline workflows.
Sidenote · Newsreader 14px01 · Work · AppFolio
Kicker · Instrument Sans · 85% width05 · Ben Siverly · Issue 01
Folio · Instrument Sans · 80% widthFilm grain, grainy fades, one ornament a page.One tile of fine film grain, light and dark, strongest on the orange fields. Fades are grainy, dithered ramps where one ground meets another, used twice. One ornament per page.
As shot
As printed
A square of orange closes a piece.
Material 3 anatomy and states, in pills.Familiar anatomy and states. Pills, condensed tracked caps, no shadows. One orange button per page; on a colour field, buttons turn Char.
| State | Primary | Secondary |
|---|---|---|
| Enabled | Primary | Secondary |
| Hover 8% | Primary | Secondary |
| Focus 10% | Primary | Secondary |
| Pressed 12% | Primary | Secondary |
| Disabled |
| State | Layer | Also |
|---|---|---|
| Hover | 8% of the label colour | Text buttons underline |
| Focus | 10% | 2px ring, 2px offset; Char on fields |
| Pressed | 12% | Pill eases to 12px corners |
| Disabled | 12% container, 38% label | No layer |
Tool bar, steps, slider and the paper proof.The Workshop tools sit on the plain ground: no fields, fades or seals while you work. A tool bar, progress steps that are real buttons, a slider with a bar handle, and what prints, set on paper. One orange button a screen.
Life Strategy Matrix
Your priority this month
Health
Invest The chart wash: Toxic Orange at 12% on paper, 14% on Gondola.
The feature, the poem, the long read.The Writing pages are printed on the insert stock: paper in Night, charcoal in Day. An essay opens like a feature, a poem keeps every line its own, and the long read numbers its chapters and gives the machine its own voice. The words are set, never restyled.
Prose
Nobody Claps
On raising children without passing on fear, caring for a father in decline, and the unglamorous work nobody claps for.
My daughter has my dark circles. The same ones I studied in mirrors as a boy and wished away. On her face I feel only tenderness, and something in me settles: what I carried alone, she won't.
I am a person who wants to put my hands around things
and call it care.
2025
5
> how do you cut a coho so it doesnt rip
Find the spine with your thumb first, and let the knife follow your thumb. Head to tail in one stroke if you can, the edge riding just over the bone. But I can only tell you where the bones are. I can't put it in your hands. For that you need the women on the point, and a few more ruined fish of your own.