The Bottleneck Is You: How to Get Out of Your Own Way with AI

From 20-minute Zapier wins to an AI agent running 36 jobs while I sleep.

You know the symptoms. A flood of AI-written email. Slop a colleague fired off without reading it back. Frankenstein social posts stitched from three prompts.

I see a deeper issue. My business stopped the moment I closed my laptop.

Emails sat unread. A report I’d promised “by end of day” was still a blank Google Doc, because I got pulled into three meetings and never made it back.

Nothing was broken. Everything was just… waiting. Waiting for me.

I am the bottleneck.

Not AI. Not my team. Not our budget. Me.

Every task that needs my attention creates friction and delay. Every decision I have to make is a decision that isn’t getting made while I sleep or sit in a meeting that should have been an email.

And I’d bet the same is true for you.

Here’s my goal for 2026. It’s not to work faster. It’s to remove myself from the loop. Build systems that run without me. Check in only when something breaks. Let the machines handle the repetitive stuff while I focus on the decisions.

(Watch the companion video: You Are the AI Bottleneck (Here’s How to Fix It).)

The bottleneck audit

Before you automate anything, you need to see where you’re stuck.

Here’s the exercise. For one week, every time you touch a task, tag it with one of three labels:

Label	What it means	Examples
DECIDE	Only I can make this call	Brand positioning, partnerships, strategy, pricing
REVIEW	I look at it before it moves forward	Email drafts, content approvals, reports, campaign setups
TOUCH	I do it because nobody and nothing else does	Forwarding emails, updating spreadsheets, scheduling, filing

You don’t need a fancy tool. A sticky note on your monitor works. A tally in a notebook. Whatever you’ll actually do.

What you’ll find

Most of your day is TOUCH and REVIEW. Very little is DECIDE.

When I ran this audit on myself, I was spending about 80% of my time on work that didn’t need my brain. Work a system could do just as well, or better, because systems don’t forget steps or get distracted.

What to do with the results

TOUCH tasks: Automate them. Today. These are pure waste. Every minute you spend forwarding an email or updating a spreadsheet is a minute you’re not doing the work only you can do.

REVIEW tasks: Build quality gates so you only see exceptions. If 95% of what you review is fine, you don’t need to review 100% of it. You need a system that flags the 5% that isn’t.

DECIDE tasks: Protect them. Guard them with your life. This is where your value lives. The rest is overhead.

Your goal for 2026: flip the ratio. Get to roughly 10% TOUCH, 25% REVIEW, 65% DECIDE.

The speed spectrum

Most tasks don’t need an AI agent. That’s the mistake I see again and again. Folks hear “automation” and immediately think they need to build Skynet or buy Agentforce.

Match the complexity of your solution to the complexity of the problem:

Level	What it is	Setup time	Example
1. Zapier / Make	If-this-then-that triggers	20 minutes	New lead to CRM plus welcome email
2. Scheduled scripts	Code running on a timer	1 to 2 hours	Nightly task archiving, weekly reports
3. AI plus workflow	An LLM making judgment calls	Half a day	Email triage with AI categorization
4. Skill pipelines	Multi-step AI with model routing	1 to 2 days	Content pipeline: research, draft, review
5. Autonomous agent	Always-on AI that runs itself	Weeks to months	Full operations agent

Start at Level 1. Only move up when you hit a real limitation, not when you get excited about the tech.

The marketers who actually speed everything up? They have 20 Level 1 automations, a handful of Level 2 scripts, and maybe one or two things at Level 3 or above.

The ones who burn out? They try to jump straight to Level 5 on day one.

A CIO recently bragged to me about all the agents they had running, so I asked him to show me the outputs. It was just massive slop.

You get paid because you’re reliable. Agents today are unreliable. If you have a lot of agents, you have a lot of problems to fix. They will get there. But right now they’re hallucinating a meaningful chunk of the time, the models shift every couple of months, and your team doesn’t know how to fix them when they break.

If you want the full map of where each level fits across your marketing function, that’s the AI marketing automation pillar this piece sits under.

Zapier wins you can build today

These take zero coding. You could set them up during a lunch break. Seriously, 20 minutes and a free Zapier account.

Automation 1: The lead response machine

Trigger: New form submission (Typeform, your website, a landing page, whatever)

What happens automatically:

Lead gets added to your CRM
Personalized welcome email fires immediately
You get a Slack or email notification
A follow-up task gets created for 48 hours later

Time saved: 5 to 10 minutes per lead. Doesn’t sound like much until you realize that’s hours per week if you’re running any kind of lead gen.

Why this one matters most: Speed-to-lead is one of the strongest predictors of conversion. The widely cited figure is that leads contacted within 5 minutes are 21x more likely to convert than those contacted after 30 minutes. You can’t respond in 5 minutes if you’re in a meeting. Zapier can.

Automation 2: The content distribution chain

Trigger: New blog post or newsletter published

What happens automatically:

Social posts get created for LinkedIn and X
Posts get scheduled through Buffer or Hootsuite
Content gets added to your newsletter queue
Your content tracker spreadsheet updates

Time saved: 30 to 45 minutes per piece of content.

Why it matters: Most content dies because distribution is manual. You write the post, feel accomplished, then forget to share it everywhere. Automate distribution, and every piece of content gets its full reach. Every time. Without you remembering.

Automation 3: The meeting prep brief

Trigger: Calendar event starting in 30 minutes

What happens automatically:

Pulls attendee info from your CRM
Checks recent email threads with those people
Creates a one-page brief in Google Docs
AI writes the summary
Drops it in your inbox

Time saved: 15 minutes per meeting. More importantly: you walk in prepared without spending a single minute preparing.

Notice what all three have in common. You are not in the loop. They fire automatically, execute reliably, and only notify you when something needs attention.

That’s the difference between “working faster” and “removing yourself.” Working faster means doing the same tasks with less friction. Removing yourself means the tasks happen whether you’re there or not.

Levels 2 and 3: scripts and AI-assisted workflows

Zapier handles simple if/then logic. But some tasks need judgment, and that’s the job for scheduled scripts and a little AI.

My task archiving system (Level 2)

If you read the Claude Code deep dive a few months back, you know this one. I had a task management system in markdown that ate hours of manual housekeeping every week. Copy completed tasks to an archive. Reorganize. Update dates. Every. Single. Week.

So I had Claude Code build me a Python script. It runs every night at 9 PM, automatically:

Scans my task file for completed items
Moves them to weekly archive files
On Sundays, does a deep clean. Anything older than 60 days gets pushed to quarterly archives
Handles all the date formatting, file creation, and folder organization

Cost: $0 per month. It’s a cron job running on my machine.

Time saved: About 4 hours per week. That’s roughly 208 hours per year, over five 40-hour work weeks I no longer spend on housekeeping.

No AI involved. Just a script on a timer doing something I used to do by hand. This is the sweet spot for most people.

My email triage system (Level 3)

This is where AI starts earning its keep.

I get 270+ emails per day. Before I automated triage, my morning routine looked like this:

Open each inbox
Scan every subject line
Mentally categorize each one (client? lead? newsletter? spam?)
Draft responses for the routine ones
File or archive the rest

Time: 90 minutes of pure inbox processing. Every morning. Before I’d done a single productive thing.

Now here’s what happens:

AI categorizes every incoming email by type: client, lead, newsletter, internal, and so on
Drafts responses for the routine stuff (meeting confirmations, simple questions, thank-you notes)
Flags anything that actually needs my decision
Archives the rest automatically

What I see each morning: 10 to 15 flagged emails that genuinely need my brain. Not 100+.

Time saved: 60+ minutes daily. That’s over 5 hours per week I got back just by automating the sorting.

The key insight here: I didn’t automate my entire inbox. I automated the triage. The sorting, prioritizing, and first-draft responses. I still make every important call. I just don’t waste 90 minutes getting to the point where I can make them.

How to build your own

Step 1: Pick your most time-consuming recurring task. The one that makes you groan.

Step 2: Describe it to Claude Code or ChatGPT: “Build me a script that does [this]. Here’s how I currently do it manually: [paste your process].”

Step 3: Test with a dry-run mode first. Always. Let it show you what it would do before it does it.

Step 4: Schedule it. Cron job on Mac or Linux, a slash command, Task Scheduler on Windows, or a simple launchd config.

Step 5: Check the output every day for a week before you trust it. Trust is earned.

Levels 4 and 5: agent pipelines and autonomous operations

Levels 1 through 3 handle individual tasks. Levels 4 and 5 handle entire workflows: multiple steps, multiple AI models, running without you.

This is the serious end of the spectrum, and the part I have the most fun with.

My content pipeline (Level 4)

When I create a deep dive like the one you’re reading right now, the pipeline looks like this:

Research phase. Gemini, Perplexity, and Grok gather sources, stats, and competitive examples. They’re great at pulling in large volumes of information and synthesizing it.
Outline phase. Claude structures the argument, maps the sections, identifies gaps.
Draft phase. Claude writes the hook, matching my voice and style from previous pieces.
Review phase. A separate AI pass checks for factual claims that need citations, and for structural flow.
I write it. Make edits. Rework. Add the personal stories and insights that only I can provide.

Each step routes to the right AI model for the job. Research goes to models with strong web access and large context windows. Writing goes to models with strong creative output and voice matching. Review goes to models that are good at structured critique. (That routing reflex is its own skill, and I unpack it across the AI marketing hub.)

Time before this pipeline: 8 to 10 hours per deep dive, spread across multiple days.

Time now: under 2 hours.

My overnight operations agent (Level 5)

This is the far end of the spectrum, and an advanced tactic. 95% of you should not do this. I’m sharing it because it’s useful to understand the agentic issues, not because I think you should build it tomorrow. You should not.

I have a dedicated Mac Mini running as an always-on AI operations agent. 24x7x365, while I sleep:

Morning brief at 5:30 AM. Summarizes overnight activity, flags issues, previews my calendar. I wake up to a briefing, not a pile of unknowns.
Task alerts 3x daily. Reminds me what’s due, what’s blocked, what changed since last check-in.
Nightly research. Deep dives into topics I queued during the day. I tag something “research this” and it’s done by morning.
Code review at 2 AM. Reviews any code changes from the day, files detailed findings for my morning review.
Security scans at 1 AM. Checks for vulnerabilities, monitors token expiration, rotates credentials.
Email processing. Listens for inbound email around the clock, routes and categorizes.
Meeting prep. Builds briefing documents before every meeting, with attendee context pulled from my notes and past conversations.

In total: 36 scheduled jobs across three specialized AI agents.

The result: I wake up to a brief that tells me what happened, what needs my attention, and what’s already been handled. My day starts with decisions, not catch-up.

Should you build this? Not yet. Gemini, Claude, and ChatGPT are all racing toward versions of this you’ll get out of the box. Wait.

This setup took months to build and refine. It breaks all the time. It needs daily maintenance. Some mornings I wake up to a brief that says “5 jobs failed overnight” and I spend 30 minutes debugging instead of strategizing.

But here’s what matters: I didn’t start here.

I started with a Zapier automation. Then I added a cron job that archived my tasks. Then AI email triage. Then a content pipeline. Each level taught me what the next level needed to be.

The progression is the strategy. You can’t pick up a guitar and expect to become John Mayer. Each automation you build teaches you where the next bottleneck is. You can’t see the Level 5 problems until you’ve solved the Level 1 through 4 problems. Don’t skip levels.

The human-only filter

Here’s the part most “automate everything” articles leave out: some things should never leave your hands.

Your art and craft.

Removing yourself from the loop doesn’t mean removing yourself from what you love. It means being intentional about where you spend your attention.

My rule:

If removing myself from a task would damage trust, dilute what makes my business distinctive, or lose the human element my audience actually cares about, I stay in the loop.

Everything else is fair game for automation.

Your move: the 3-week speed sprint

I’m not giving you a 30-day roadmap this time. If you read the Claude Code deep dive, you already have one. This is faster.

Week 1: audit plus quick wins

Run the bottleneck audit. Tag everything DECIDE, REVIEW, or TOUCH for five days.
Set up 2 or 3 Zapier automations for your biggest TOUCH tasks. Lead response, content distribution, meeting prep. Pick the ones you hate the most.
Target: reclaim 3 to 5 hours per week.

Week 2: your first script

Look at your REVIEW tasks. Find the one that eats the most time.
Build one scheduled script or AI-assisted workflow to handle the triage.
Test it for a full week before you trust it. Watch the output every day.
Target: remove yourself from one daily recurring process.

Week 3: pipeline thinking

Pick one end-to-end workflow. Content creation, client reporting, email outreach.
Map every step. Identify which need AI judgment and which are simple automation.
Build the first version. Ugly and simple is fine. Working beats pretty.
Target: one workflow that runs start to finish with minimal input from you.

The weekly gut check. At the end of each week, ask yourself one question:

“What still stops when I stop?”

Whatever the answer is, that’s your next automation target.

Build the machine. Then get out of its way.

Find your bottleneck, then kill it

Every Friday I write up the actual systems I’m running, what broke that week, and the one automation worth your weekend. If you’ve ever closed your laptop and felt your whole business go quiet, this is the newsletter that helps you fix it, one level at a time.

Subscribe free and get the next deep dive in your inbox.

What’s the one thing that stops when you stop? That’s where you start.