playbooks

Clean Build Run: The Daily Loop

The morning-to-evening Claude Code loop. Clean state, build features, run tests, capture learnings, then reset for tomorrow.

12 min read·

Playbook / 3 Phases / Automation Framework

Clean. Build. Run.

The automation framework distilled from 30 posts and 8 real client engagements. Stop automating chaos. Clean your processes first, build the right systems, then let them run.

CLEAN

  • Audit
  • Delete
  • Document

BUILD

  • Design
  • Test
  • Iterate

RUN

  • Deploy
  • Monitor
  • Evolve

Phase 1: Clean

Phase 1: Clean
STEP 01

Stop Automating Stupid Stuff

Insight

"Automating a broken process makes it faster, more rigid, and 3× harder to undo."

31 active workflows. 14 patching manual steps. 6 duplicate notifications. The rest unknown. They were about to migrate all of it. The most valuable advice: "don't automate this."

If you can't explain a process to a new hire in 3 minutes, you don't have an automation problem. You have a clarity problem.

Do This

List all active automations. For each one, ask: can a new hire understand why this exists? If no, mark it for review.

Phase 1: Clean
STEP 02

Delete 29 Automations

Insight

"More automation is not better automation. The goal is a stack your team understands, trusts, and can maintain without you."

Client added 29 automations after initial setup. Response times got slower. 12 duplicates, 9 "just in case" alerts that never fired, 8 genuinely useful. Deleted 21. Team immediately knew what was running. Engineers started touching the stack again.

29

Automations audited

21

Deleted

8

Kept

Do This

Audit your stack this week. How many automations can you delete without anyone noticing?

Phase 1: Clean
STEP 03

The Numbers Don't Lie

Insight

"The companies with the most automations had the least clarity."

Stats from 8 clients. The pattern was consistent across industries, team sizes, and tools.

47

Avg active automations

31%

Running but not measurable

22%

Duplicates

6+ hrs

Weekly stack management

Only 3 out of 8 clients could say what breaks if they deleted half their automations.

Do This

Run this audit on your own stack. Count total automations, identify duplicates, measure management overhead.

Phase 1: Clean
STEP 04

Before/After: What Clean Looks Like

Insight

"The hours didn't disappear. They moved up the value chain."

15-person growth agency in Australia. 12-week engagement.

Before

  • //20 hrs/week manual research
  • //4-5 day lag on first touch
  • //No messaging consistency
  • //Data spread across 4 tools

After 12 Weeks

  • //Research fully automated
  • //First touch in 2 hours
  • //Messaging standardized
  • //Single pipeline view

The 3 people doing manual work now do strategy.

Do This

Identify your biggest time sink. Calculate hours/week. That's your first Clean target.

Phase 2: Build

Phase 2: Build
STEP 05

Why Your Automation Fails

Insight

"These 3 questions kill about 40% of automation ideas before we write a single node. That's a feature, not a bug."

The 3 reasons automations fail:

  1. // 01Process wasn't defined before automating
  2. // 02Nobody owns it after it ships
  3. // 03Built to solve a symptom not a cause

The 3 questions that prevent this:

  1. // 01Is this process documented?
  2. // 02Who will own this?
  3. // 03Are we solving the real bottleneck or the visible symptom?

Do This

Before your next build, answer these 3 questions. If any answer is "I don't know," stop and clarify.

Phase 2: Build
STEP 06

AI Won't Fix Your Broken Process

Insight

"AI amplifies what's already there. Clean process + AI = force multiplier. Messy process + AI = scaled mess."

Sales team with bad CRM hygiene adds AI to summarize pipeline. AI summarizes accurately. It summarizes inaccurate data. Now bad data looks polished and confident.

Messy Process + AI

Inaccurate data gets a professional summary. Leadership trusts the report. Decisions made on bad data. Problem is now invisible.

vs

Clean Process + AI

Accurate data gets a professional summary. Leadership trusts the report. Decisions are sound. AI accelerates good outcomes.

Do This

Before adding AI to any process, ask: is the data feeding this process clean? If no, clean the data first.

Phase 2: Build
STEP 07

The 3AM Automation

Insight

"90% of flaky automations run on polling when they should be on webhooks."

3:14am, workflow fires. Deal moves in CRM, PDF generated, Slack message sent. Why it works: webhooks, not polling. Polling = checking every 5 minutes. Webhooks = instant notification. The difference between a system you trust and one you check on.

Polling

Checks every 5 min. Misses events between intervals. Burns API calls. Unreliable at scale. You check on it.

vs

Webhooks

Instant notification. Never misses events. Zero wasted calls. Scales cleanly. It runs without you.

Do This

Audit your automations for polling vs webhooks. Every polling-based trigger is a reliability risk. Switch to webhooks where possible.

Phase 2: Build
STEP 08

Tools Don't Fix Chaos

Insight

"Tool-first: I have a problem, what tool solves it? Process-first: I have a problem, what's the process, what tool supports it?"

Company A: Tool-First

Buys tool first. 60% adoption. Buys next tool 6 months later. Repeat.

vs

Company B: Process-First

Documents process. Kills unnecessary steps. Finds simplest tool. 90% adoption. Doesn't touch it for a year.

60%

Tool-first adoption

90%

Process-first adoption

Do This

Next time you consider a new tool, document the process first. Then find the simplest tool that fits.

Phase 3: Run

Phase 3: Run
STEP 09

The System Runs Without You

Insight

"The goal is not more workflows. The goal is a stack your team can maintain without you."

The test: can you go on vacation for 2 weeks and nothing breaks? If the answer is no, you don't have a system, you have a dependency. Every automation should have:

  1. // 01Clear owner: someone is responsible when it breaks
  2. // 02Monitoring: you know when it fails before users do
  3. // 03Error handling: it degrades gracefully, not catastrophically
  4. // 04Documentation: someone else can fix it

Do This

For each automation, answer: who owns this? What happens when it fails? Is it documented? If any answer is blank, fix that before building anything new.

Phase 3: Run
STEP 10

Clean-Build-Run as a Loop

Insight

"CBR is not a one-time project. It's a quarterly cycle. Clean again, build better, run smoother."

Q1: Clean (audit, delete, document). Q2: Build (design, test, iterate). Q3: Run (deploy, monitor, evolve). Q4: Clean again, because the stack grew, new tools arrived, processes changed.

The companies that stay fast are the ones that keep cleaning. The companies that slow down are the ones that only build and never audit.

CLEAN BUILD RUN CLEAN BUILD RUN ...

Do This

Schedule a quarterly Clean day. Put it on the calendar. Audit everything. Delete what doesn't serve you. Then build.

The Framework

Three phases. One loop. Repeat quarterly.

Clean

Audit. Delete. Document.

Avg 40% of automations deleted

Build

Process first. Then tool. Then AI.

3 questions kill 40% of bad ideas

Run

Own it. Monitor it. Evolve it.

Goal: runs without you

Tools for each phase

Clean

  • Notion / Airtable - inventory
  • Loom - document before delete
  • Claude Code - audit automation logic

Build

  • n8n / Make / Zapier - workflows
  • Supabase - data layer
  • Anthropic / Gemini - AI nodes

Run

  • Slack - error alerts + monitoring
  • PostHog - usage analytics
  • sheal retro - session capture

Evening Retro

End every build day with 10 minutes of retro. What shipped? What broke? What would you do differently? Run sheal retro to capture learnings before they disappear. The loop compounds only when you write down what you learned.

Want this running in your stack?

AY Automate builds AI automation systems for production teams.