My Daily Workflow: Reviewing Diffs Across 6 Projects

March 16, 2026

People ask what the octopus workflow looks like in practice. Not the theory. The actual day.

Here’s last Tuesday. Unedited except for shortening some diffs. This is what 6 projects moving forward looks like when one person runs 8 terminal tabs.

8:15am – The scope block

Kids are at school. Coffee is made. I open iTerm2 and check where each project left off yesterday.

Tab 1: scouter       – needs rate limiting on API
Tab 2: triumfit      – onboarding copy is stale
Tab 3: lucid         – users want PDF export
Tab 4: logline       – search is slow on large libraries
Tab 5: jwp-site      – new tutorial post needed
Tab 6: octopus       – about page first draft
Tab 7: skill-library – marketing skill needs examples section
Tab 8: pwa-marketing – beyond529 landing page A/B variant

I already know what each project needs because I keep a running list. (More on how I scope these tasks.) Nothing clever – a markdown file per project with a “next up” section.

I write scopes for all 8. Some take 30 seconds (the Triumfit copy change is three lines of new text). Some take 2 minutes (the Scouter rate limiting needs middleware placement details). Total scoping time: 17 minutes.

I paste each scope into its tab. All 8 are running by 8:32am.

8:32am – The gap

This is the part that feels wrong at first. You have 8 agents building. You should be doing… something?

No. The agents are working. I’m not needed.

I answer emails. I review yesterday’s analytics for a client project. I text my wife about dinner. Normal morning stuff. The terminals are churning in the background. I can hear my laptop fan.

9:10am – First review pass

I flip through the tabs. Most are done. Here’s what I find:

Tab 1 (Scouter rate limiting): Done. 3 files changed. The middleware is clean, registered in the right order, returns the correct 429 response. I read the diff line by line. Accept.

Tab 2 (Triumfit onboarding): Done. Copy updated, nothing else touched. 30-second review. Accept.

Tab 3 (Lucid PDF export): Done, but the PDF layout is off. The font size is too small and it’s not including the entry date. I type a correction: “Increase body font to 14pt. Add the entry date below the title in gray, formatted as ‘March 24, 2026’.” Run it again.

Tab 4 (Logline search): Still running. The library index rebuild is a bigger task. I skip it and come back later.

Tab 5 (JWP tutorial): Done. Draft is solid. A few voice adjustments needed – it’s using “just” too much and one paragraph is throat-clearing. I make those edits manually in the file. Faster than re-prompting.

Tab 6 (Octopus about page): Done. Good first draft. I’ll revise later but the structure is right. Accept for now.

Tab 7 (Skill library): Done. The examples section has 5 examples. Two are generic. I delete those and ask for two replacement examples using specific campaigns I’ve run. Run it again with the specific context.

Tab 8 (PWA landing page): Done. The A/B variant changes the headline and hero copy. I read both versions side by side. The variant is better. Accept.

Total review time: 28 minutes. Six tasks accepted. Two kicked back with corrections.

9:45am – Second round

The two corrections are done. I review them.

Tab 3 (Lucid PDF – revised): Font size fixed. Date added. Looks right. Accept.

Tab 7 (Skill library – revised): New examples use the specific campaigns. Much better. Accept.

Tab 4 (Logline search): Now done. The search index is rebuilt with a new scoring algorithm. I test it with a few queries against the dev data. Results are noticeably better. The diff is bigger – about 120 lines – so I spend 10 minutes reading it carefully. One function could be cleaner but it works. Accept. I’ll refactor that function in a future scope if it bothers me enough.

10:00am – Scope round two

Three tabs are free. I scope new tasks:

Tab 2: triumfit      – add workout completion animation
Tab 3: lucid         – fix: export button hidden on small screens
Tab 6: octopus       – write first blog post

Paste, run, go make more coffee.

10:30am – The rest of the morning

Same loop. The second-round tasks are smaller so they finish fast. The animation task produces something I don’t love – the easing feels wrong. I give a specific correction (“use ease-out, 300ms, scale from 0.8 to 1.0”) and it nails it on the second pass.

By noon I’ve done 3 scoping rounds, reviewed 14 tasks total, rejected and corrected 4 of them, and shipped commits across all 6 active projects.

What this looks like in git

scouter:       2 commits (rate limiting, test updates)
triumfit:      2 commits (onboarding copy, completion animation)
lucid:         2 commits (PDF export, responsive fix)
logline:       1 commit  (search index rebuild)
jwp-site:      1 commit  (tutorial draft)
octopus:       2 commits (about page, first post)
skill-library: 1 commit  (marketing skill examples)
pwa-marketing: 1 commit  (landing page A/B variant)

12 commits across 7 repos. One morning. Zero lines of code I wrote from scratch. Every line reviewed.

The honest parts

Not every day is this clean. Some failure modes I hit regularly:

Scope that’s too big. I’ll occasionally scope something ambitious and the agent produces 300 lines that are 60% right. Now I’m spending 20 minutes understanding what it built instead of 5 minutes reviewing what I asked for. My fault for scoping too big. The fix is always smaller tasks.

The wrong mental model. Sometimes I scope a task and the agent’s approach is technically correct but architecturally wrong – it solves the problem in a way that doesn’t fit the codebase. This happens when I specify the what but not the how. “Add caching” without specifying the caching strategy. Now I’m looking at a Redis integration when I wanted in-memory caching.

Review fatigue. I wrote a full guide on reviewing AI-generated code — including the 3-pass method that helps with this. By task 12, my reviews get sloppy. I start accepting diffs I should read more carefully. I’ve shipped bugs this way. The fix: stop at 10–12 tasks per day, and review the hardest diffs first when I’m fresh.

The net

One person. 6 projects. 12 commits. One morning.

Not because the AI is magic. Because the workflow is parallel.

The AI does the building. I do the thinking. If you want the setup details and how to get started, or the best practices I’ve learned, those are good next reads. The thinking happens upfront (scoping) and afterward (reviewing). The building happens in the background, across all 8 arms simultaneously.

That’s Tuesday. Most Tuesdays look like this.