LADLE for deep work.

An always-available AI assistant is a productivity multiplier and a focus killer. Both are true. Which one wins depends entirely on the pattern you develop for when to reach for it.

The default failure mode.

You sit down to write for two hours. Halfway through paragraph one, you get stuck on a word. Instead of pushing through, you tab over to LADLE for a synonym. The synonym takes 30 seconds; the re-entry into your writing flow takes 10 minutes.

Repeat this six times and you've spent the two hours in a state that felt productive but produced 20 minutes of actual writing. This is the pattern most people fall into with any real-time assistant.

The pattern that works.

Batch your assistant use. During deep work sessions, note the things you would ask, but don't ask them. Keep a running list — "look up which font that guide used"; "check if 'perpendicular' is the right word here"; "get the exact 10-K quote". Then, at a natural break in the work (end of a section, end of a Pomodoro, coffee refill), spend 5-10 minutes clearing the list.

The result: you get the assistant's value, and you keep the flow state. Individual interruptions get replaced by planned batches.

Which tasks are batch-safe and which aren't.

**Batch-safe** (do these at breaks): - Fact-checks and citation lookups. - Word-choice questions. - Prompts to critique the section you just finished. - Long research questions that would otherwise fragment your attention.

**Not batch-safe** (do these mid-flow): - Debugging a runtime error that's blocking your next step (you're not in flow anyway — the error broke it). - Writing where the assistant is the writing partner from the start (a co-drafting session, not deep work). - Meeting prep where you need the assistant to shape your incoming session.

The rule of thumb: if you're in flow, the assistant waits. If flow is already broken, the assistant can go now.

Project setup for deep work.

Have a dedicated Project for each deep-work stream. Attach your voice guide, your reference materials, and the immediate context of what you're working on. This means when you do reach for the assistant during a batch break, you don't spend three prompts giving it context — it's already there.

What deep work with LADLE looks like at its best.

Two-hour session, phone on airplane mode, LADLE tab closed. Ninety minutes of writing / coding / thinking. Twenty minutes at the end where you open LADLE and clear the batch list you kept during the session — every question gets a fast, contextually-aware answer because your Project is set up. Ten minutes of applying those answers.

Total output: two hours of forward motion, with the assistant's value fully captured. This is meaningfully better than either "no AI, all flow" or "AI always on, no flow".

What to disable.

If you have any real-time AI assistants that autocomplete as you write (some IDE plugins, some writing tools), turn them off during deep-work sessions. Autocomplete is the enemy of the flow you're trying to preserve.