Skip to content

The loop

/switchback ──▶ printer ──▶ your pen ──▶ phone camera ──▶ /switchback
the agent you print you think in you photograph the agent reads
makes the pages ink, away from the pages the marks back
the screen

Working with agents can turn much of the day into a stream of hand-offs: ask, wait, check, ask again. There is little room left to sit with one hard problem for long enough to think it through. Switchback puts a small amount of deliberate friction back only where the thinking happens: the sitting, the pen, the walk away from the screen, and the photo you have to take to get an answer.

That is the motivation for building it. It is not a measured claim that Switchback makes anyone more productive, and the site does not make that claim either.

The agent could just print a document and ask you to read it, but a document you read is not the same as a page you fill in by hand. Each printed page asks for your thinking; it never prefills the answer, recommends an option, or states a conclusion. One page holds one thinking move, so working through a workbook means making several small decisions in ink instead of one long scroll of text.

The printer prints black on white. Every mark on top of that print is your own ink, in a small colour language described on the ink page — including your Reason ink, the plain black pen for the main content, which is why colour is a convenience here and never the only carrier of meaning.

When the photos come back, the agent does not summarise the whole page. It works in a fixed order: first the photo test — what survived the photos and what did not — then the questions you wrote in your Ask pen, answered in the number order you wrote them, then any notes about the pages themselves, then the content, then the next step. That order exists so that the thing you actually asked for is never buried under everything else the agent noticed.

Marks about the page itself — “these columns are too small,” “this was not a good question” — are kept separate from marks about your problem. The former is feedback on the workbook; the latter is the content the next step is built from. Conflating the two would mean product feedback quietly reshaping your decision, which the agent is built not to do.

Objections you raised and never resolved, decisions you locked in with a circle, and anything you crossed out but kept visible are all carried into carry.md in the workbook’s folder. A Series reads that file back at the start of its next round, so an unresolved concern does not quietly vanish between sittings — it either gets answered, or it comes back the next time you sit down with the pages. A Ritual instead repeats the identical page every time; there, change becomes visible by flipping back through past rounds rather than by reading anything back.