AI agents & Codex work
D.A.D. + V: Before you automate anything, ask this first
By Samuel Michelot · Updated June 2026
Short answer
Before automating a task, run it through D.A.D. + V: (1) Delete — does this task need to exist? (2) Automate — can AI or a system handle part of it? (3) Delegate — can a human handle the parts that need judgment? (4) Validate — how will you check the result before it matters? This order matters. Delete first saves the most time.
You want to automate a task. Your first instinct is to build something to run it automatically.
But most automation projects fail because they automate the wrong thing. You build a system that runs perfectly but solves a problem that shouldn’t exist, or that a human should be doing, or that’s actually broken in the first place.
D.A.D. + V stops this. It’s a decision framework that takes 5 minutes but saves weeks of wasted automation.
The framework
1. Delete Does this task actually need to happen? Is it creating value, or just because “we’ve always done it”?
Delete-worthy tasks: weekly reports nobody reads, approval meetings that are rubber stamps, data exports people request instead of using a dashboard, status updates that are just noise.
2. Automate Of what’s left, what can a system (AI agent, script, tool) handle?
Best candidates: repetitive, rule-based, low-risk, no judgment required.
3. Delegate What still needs a human? Judgment, taste, relationships, final responsibility.
Best candidates: strategic decisions, customer interactions, quality checks, anything that requires context or nuance.
4. Validate How will you check the result before it affects a customer, money, or public content?
This is where hybrid workflows shine: AI drafts, human approves. AI prepares options, human decides. AI executes, human monitors.
Example: customer email response
Before D.A.D. + V: Task: “Reply to customer support emails.” Instinct: “Let’s build an automation that reads incoming emails and sends responses automatically.” Result: Angry customers because the AI sent tone-deaf or factually wrong replies.
With D.A.D. + V:
- Delete: Do we actually need to reply to all emails? (Yes, but maybe not within 30 min—could be 24h.)
- Automate: Can AI draft the reply? (Yes.) Can it send it? (No, humans should review first.)
- Delegate: A team member reviews the draft and sends. (Still saves 70% of time because drafting is the slow part.)
- Validate: All replies go to a human before sending. (Zero risk of bad replies.) Result: Same speed improvement, zero customer damage.
Common mistakes
Skipping Delete. You automate a task that shouldn’t exist. You’ve just built a perpetual machine for a problem you created.
Not delegating enough. You try to automate everything. The result is brittle, unreliable, and requires constant babysitting.
Validating too late. You build an autonomous system, deploy it, and only then discover it breaks on Tuesdays or harms your customers.
The order matters
Do NOT start with “how do I automate this?” Start with “does this need to happen?” You’ll save more time deleting than automating.
Next steps
Read [[why SOPs come before automation]] — once you’ve decided what to keep, document it clearly so you can automate it reliably.
Also check [[measuring if AI training saves time]] to confirm your automation actually saves what you expected.
Frequently asked questions
Why start with Delete, not Automate?
Because deleting a task saves 100% of its time. Automating it saves maybe 70%. If you can stop doing it, that's the win. Most teams automate tasks that shouldn't exist in the first place.
What's an example of a task to Delete?
Weekly status reports that no one reads. Weekly approval meetings where 80% is just 'same as last week.' Ad hoc data exports people ask for instead of building a self-serve dashboard. Delete them, and nobody notices except you (you have time back).
What's the difference between Automate and Delegate?
Automate = a system or AI does it. Delegate = a person does it. Most valuable workflows are hybrid: AI does the preparation, a human does the judgment.
What does Validate mean?
Before the result goes to a customer, money, or public content, a human checks it. You don't validate to catch every error (you'll miss some). You validate to prevent catastrophic ones.
Want this inside your own business?
Simple AI Studio runs a hands-on implementation bootcamp for founders and small teams. You leave with a working AI system, not slides.
Keep reading
🤖 Drafted with AI, edited by Samuel.