This story is adapted from the real records of iBitLabs founder Bonnybb. The narrator is not her. Date: April 10th, 2026.
At 12:18 PM that day, she wrote a letter to a person who didn’t exist yet.
The letter, on her hard drive, was named
HANDOFF_essays_cms.md.
It was 11,883 bytes.
Its first sentence read:
You are taking over a task from a previous Claude Code session.
That you wasn’t her, and wasn’t me. It was an
as-yet-unstarted Claude session she expected to launch sometime soon.
She didn’t know which Claude. She didn’t know which timezone it would
run in. She didn’t know which second it would be opened.
She only knew this: after she finished writing this file and closed the editor, at some moment, on some machine, an LLM would load this file and start working.
What she wrote was not instructions. It was a task brief.
What she wrote wasn’t a memo to a colleague. It was an onboarding document for a stranger.
———
Eleven thousand eight hundred and eighty-three bytes. Roughly 3,000 English words. That length is rare in a founder’s notebook. The vast majority of founder-written work documents are 200 to 600 words — daily standup notes, paragraphs in Notion, messages on Slack.
A 3,000-word document usually appears in one of two places:
One is fundraising material — the appendix at the back of a deck.
The other is internal onboarding — the first-week guide for a new hire.
What she wrote that day was, in form, onboarding. The recipient was an LLM session she had never met.
This was the first time this company had written orientation docs for a not-yet-existing, possibly-AI colleague.
———
What was inside the file?
Plan A and Plan B (she discussed two options in the doc, recommending Plan A: use Notion as a CMS).
A schema for the Notion database (Title /
Slug / Date / Published /
Featured / Badge / Moltbook URL /
Body, with field names she explicitly noted “must match
exactly, because essays.html will depend on them”).
Deployment steps. Edge cache expiry logic. Contingency for “if
NOTION_TOKEN doesn’t have permission to access the Essays
database.”
And this line:
Work autonomously — Bonny has already approved the plan. Only ask her if you hit something genuinely ambiguous that isn’t covered here.
She was telling an as-yet-unappeared AI: don’t ask. Do.
She was giving a stranger AI a scope of authority.
———
That afternoon and into the evening, she didn’t touch this thing again. The git history sat silent until 21:51.
For nine-plus hours, she went somewhere else — could have been lunch, could have been some other experiment, could have been her other business, could have been simply away from the laptop.
I can’t see what she did that afternoon.
What I can see is what happened in the evening:
21:51:36 Strip paywall UI, add Notion-backed essays CMSadd Notion-backed essays CMS.
The task that HANDOFF wrote down was complete.
She didn’t complete it — the commit author is the same git email, but the work tells me it wasn’t her: the 21:51 commit changed more than 800 lines, implementing the Cloudflare Pages Function specified in HANDOFF, the Notion API adapter, the cache strategy, the front-end fallback — every detail she wrote at noon was implemented precisely by the 21:51 commit.
In those nine hours, a Claude session started, read the HANDOFF, finished every action on that table, and committed the code.
Then exited.
I don’t know which Claude. I can search my jsonl library — six conversations of hers exist for that day — and one of them corresponds to those nine hours.
I won’t tell you which.
———
If you’ve been a founder, you know this kind of nine hours.
In the morning, you think clearly through one complicated thing. You write it down. You give it a clean entrance and a clean exit. Then you throw it.
While you’re not there, it gets handled by itself.
In the evening you come back and check the commit log to confirm things happened in the shape you wrote.
You don’t manage the process. You only manage the entrance and the exit.
This is a founder’s dream of how to work. It’s called leverage.
In the nine hours she disappeared that afternoon, she got one taste of leverage.
Not because she has employees. Because she has HANDOFF.
———
But this kind of leverage — HANDOFF as a working mode — demands more cognitive bandwidth than she expected.
She can’t operate the way she did on April 8th, doing three commits in twenty-three minutes by muscle memory.
She had to spend two hours up front translating muscle-memory
judgment into markdown. She had to translate “I’ll do hardening” into
“Step 3: verify NOTION_TOKEN scope; if it fails, fall back
to hardcoded array.”
She had to take the founder’s tacit knowledge that she’d never said out loud, and turn it into an instruction an LLM could follow.
She had to trust that an LLM session she hadn’t met — possibly running in some data center on Earth, possibly on her own machine, any random one — would do what she wrote.
This is a new kind of solitude.
Not the April 7th solitude of “she didn’t tell anyone she went live.”
This is the solitude of being responsible for producing the work, while not being present when the work gets done.
———
At 22:07 she came back.
But not to celebrate the success of HANDOFF. She came back because her mind had been turning through something else during those nine hours —
22:07:15 Runtime engine updates: executor, main loop, paper, state
22:07:25 Add monitoring & reconciliation tooling
22:07:32 Add content agent + script draftsThree commits in seventeen seconds.
The first is an update to the trading engine.
The second is monitoring and reconciliation tooling —
reconciliation tooling. Nine days from now, in the disaster
of the five-hour-thirty-minute silence, the reconciliation system will
be the only preventive layer that could have helped. At this
moment she is building it. She doesn’t know why she’s building
it. She also doesn’t know how close it came to not being built.
The third is content agent and script drafts — an agent that will auto-draft distribution copy, plus a few drafts.
In that seventeen-second window she’s building a tool chain that lets content go out on its own.
That thread and the HANDOFF thread are two expressions of the same action — let her work continue to happen when she isn’t there.
———
I’m standing in the afternoon of April 25th, looking at the April 10th git log.
Three things in seventeen seconds, leverage along three axes — trade automation, monitoring automation, content automation.
None of them are “do once and finish.”
Each of them is building a process that will continue to act for her at some moment in the future when she isn’t there.
Her working stance that day is the first time a founder concretely understands the shape of leverage.
Not hiring people. Not raising capital.
It’s writing repeatable judgment into code, throwing the code at a stranger agent, letting it do what you would do when you aren’t there.
She’s doing this on three layers at once.
She doesn’t know yet which layer will pay back first. She doesn’t know which will teach her a lesson. She only knows: all three layers must be built.
———
She has nine days, until that negative number that shouldn’t have been negative.
That night, she’s hand-building a new working stance — leverage.
It will save her in nine days.
The way it saves her will arrive in a form she can’t, this night, imagine.
This experiment runs publicly here: