My Claude Daily Workflow
(What I Actually Do)

Honest accounting

This is what happens, not what sounds good

Most Claude workflow posts describe an idealized system. This one describes what I actually do — including what I used to do that didn't work, and the specific things that made the workflow stable enough to sustain over months instead of weeks.

The short version: I run four phases each day, each with a distinct job. Morning open, mid-morning work block, afternoon review, end-of-day close. The phases aren't rigid time blocks — some days the afternoon review doesn't happen. But the structure is real and the handoffs between phases are what make the whole thing work.

Phase 1

Morning open: load context, check status

The first thing I do is load my session file. My session file is a plain text document I keep in my project folder — it contains what I was working on yesterday, what's still open, and what's first today. I paste it into Claude at the start of every session. Takes 30 seconds.

Without this, every morning session starts cold. Claude doesn't remember yesterday. I don't remember exactly where I left off. The first 10 minutes becomes re-orientation. The session file eliminates that.

After loading, I ask Claude for a quick status check: given what's in the file, what's the most important thing to move on today? Claude usually surfaces one or two things I'd half-forgotten. Then I start working.

Here's my session file from yesterday:

[paste SESSION.md contents]

Based on this, what's the most important thing to work on first today?
Flag anything that looks time-sensitive or blocked.
      

Phase 2

Mid-morning work block: Claude as thinking partner, not autocomplete

This is where most of the actual work happens — writing, coding, research — and it's also where most people use Claude wrong. They use it as a fancy autocomplete: type something half-formed, hit enter, get a completion, use it without much thought. The output is fine but the thinking isn't yours.

What works better is using Claude as a thinking partner. I'll write my own draft first — even a bad one — and then ask Claude to react to it. Or I'll describe a problem in my own words and ask Claude what I'm missing. Or I'll have Claude play devil's advocate on a decision I've already mostly made.

You leave the session having thought harder. The thinking is yours. Claude sharpened it.

For coding specifically, I do the same thing. I write the structure, even if it's pseudocode, and Claude fills in or critiques. I don't ask Claude to start from scratch — I find the output less useful and I learn less from it.

Phase 3

Afternoon reviews: ask Claude to find the gaps

After a morning work block, I step away and come back in the afternoon with fresh eyes — and Claude. I paste what was produced and ask Claude to review it with a specific lens: what's missing, what's unclear, what would someone reading this for the first time get wrong.

This is different from asking Claude to "make this better." That produces generic improvement. Asking for specific gap types produces specific findings. "What would a skeptical reader push back on?" returns different feedback than "improve this."

I don't act on every flag. Claude finds more issues than actually matter. But it reliably surfaces one or two things I'd overlooked, and those are worth the session.

Phase 4

End-of-day close: update the session file

The last thing I do each day is ask Claude to generate an end-of-day close. This is a structured summary: what was completed today, what's still open, what's waiting on someone else, and what should be first tomorrow. I paste Claude's output into my SESSION.md file, replacing the previous entry.

Tomorrow's morning open will load this file. The loop closes.

"The session file is the infrastructure. Without it, every morning starts cold. With it, the day picks up where the last one ended."

What didn't work

One long session with everything in it

For my first three months of regular Claude use, I kept one long session open all day. I'd start a conversation in the morning and just keep going — different tasks, different topics, everything in one thread. By afternoon the context was a mess. Claude would reference things from hours ago, lose track of what was active, and give answers that technically responded to my question but missed the point because the relevant background was buried 40 messages up.

Separate sessions with clear handoffs fixed this immediately. Each session is sharp because it starts with exactly the context it needs and nothing else. The session file is the handoff mechanism. It takes two minutes to produce. It's the highest-value two minutes in the workflow.

Next step

Book 1 covers the full system

This post covers the skeleton. Book 1 covers the full system — session file format, the full inbox triage routine, task prioritization, communication workflows, and how to keep Claude useful across projects without losing context between them.

Read Book 1 →