Mr Kirkland

Serial entrepreneur, building since 2005.

© 2026 Mr Kirkland

AI and work · 4 min read

I do a lot of my work on a walk now

Rather than open my laptop to an endless list of emails I haven’t replied to, I just have a nice chat.

The short version, on video.

This is part two of Inbox zero was never the goal. In part one I showed you my AI workflow that helps me triage my inbox and tasks. It occurred to me that I could probably do this via a chat, and dispatch the decisions back to a code agent.

The “decision pack”

A Claude Code agent scans my inboxes, my task manager and my calendar. It prepares an ordered list of questions, each with a suggested answer, on how to sort, order and dispatch the agent-friendly tasks. Most mornings that’s 12 to 15 decisions. Claude, being great with tacky jargon, decided to call this my “decision pack”.

D1  A partner wants a quote as a PDF.
    A (recommended): build it from the template and draft the reply
    B: reply that email is enough
    C: hand it to the team

This prep really simplifies my work. I can quickly clear small decisions that might have been wrapped up in gathering all the context. The agent does the opening and reading. I only get the decisions: yes or no, today or Friday, A or B, etc.

The walk

I load the pack into a plain chat on my phone and go for a walk. It reads me each decision and I answer out loud. Mostly I just say “A”, the recommended option.

I do all the work without having to ever look at my inbox/task list. There’s no unread count, no notification, no “while I’m here I’ll just…”. About 20 minutes later I’ve made the decisions, and feel great: an organised task list, and I’ve had a nice walk.

Why two tools

Claude Code connects to everything (email, tasks, calendar, our CRM), but you can’t talk to it in real time (yet). The chat app on my phone has real-time voice, but less integration with tools. So the thinking happens in the chat and I hand off the doing to Claude Code.

The walk routine: Claude Code reads my inbox, tasks and calendar and writes a decision pack; I answer it by voice on a walk; Claude Code then does only what I answered. InboxTasksCalendar 1. Claude Code reads read-only · opens every task and thread writes Decision pack D1 Quote as a PDF? (a) recommended (b) … (c) … D2 … D15 paste into my phone 2. Decide on a walk voice chat on my phone · no tools “D1: quote PDF…” “A.” paste the results back 3. Claude Code dispatches the only step that changes anything Drafts forme to send Tasks sortedand dated Today’s plan,timed
The chat can’t change anything. Only Claude Code acts, and only on what I answered.

The return trip

At the end of the walk the chat gives me a results block. I paste it back into Claude Code, which dispatches the work “so I don’t have to do anything at all” and leaves me a timed plan for the rest of the day.

That feeling of dread before opening the inbox totally disappears, because I just don’t look at my inbox anymore.

How to steal it

  1. Prep: give an agent read-only access to your inbox, tasks and calendar, and ask for a numbered list of decisions, each with options and a recommendation. Nothing else.
  2. Decide: paste it into a voice chat on your phone and answer out loud on a walk. The chat writes a results block.
  3. Dispatch: paste the results back into the agent and let it carry out only what you answered.

Download the prompts (walk-decision-pack.md)
The three prompts, ready to paste. Works with any task app, inbox and calendar your AI agent can read.

More: all writing · LinkedIn