Why AI Hasn't Made Your Work Any Faster

In the past 30 days I've worked directly with over 30 business owners.

Every one of them had AI doing pieces of their work. Email. Research. Content. Admin. Some had it set up across a dozen small tasks.

But when I asked one simple question, they looked at me like they saw a ghost:

What part of your work is everything else waiting on?

Not one of them had a clean answer. That's the mistake I keep seeing people make with AI.

They automate the work that's easiest to automate instead of the work that's actually holding everything else up.

So yes, AI made the email faster, the research faster, and the first draft faster. But the business itself doesn't feel much faster. 

Because TASK SPEED and SYSTEM SPEED are two different things.

Your work is a chain of dependent steps. And that chain can only move as fast as the step everything else is waiting on.

So before automating anything, go in this order:

1. Find the bottleneck.

2. Figure out why it's slow.

3. Look at how the process actually works.

4. THEN automate it.

Before you can fix the bottleneck, you have to learn what one actually looks like. 

You're speeding up the wrong kids

There's a story from Eliyahu Goldratt's The Goal that I think about constantly.

A group of Boy Scouts is hiking to camp.

As they walk, the fast kids pull ahead, and the slower kids fall behind. Everyone keeps stopping so the back of the line can catch up.

Eventually they find the problem.

His name is Herbie. Herbie is the slowest kid in the group, and he's carrying the heaviest pack. The TROOP gets to camp when HERBIE gets to camp.

It doesn't matter how fast the kid at the front walks. Making the fast kids faster just makes the gap bigger. So they move Herbie to the front and redistribute the stuff in his pack among the other kids.

Herbie speeds up, so the WHOLE TROOP speeds up.

Nobody needed to work harder. They needed to fix the one thing setting the pace for everyone else.

The real challenge is spotting Herbie in your own business.

Say you bring in 50 new leads this week. Great. But onboarding is still manual and you can't actually take those clients on. Marketing got faster. The business didn't. 

The leads pile up in front of the exact same constraint. And you can see companies making the same mistake at a much bigger scale.

MIT's Project NANDA looked at hundreds of enterprise AI deployments in 2025. The report was preliminary, so I care less about the viral "95% failed" headline than one detail underneath it: more than half of GenAI budgets were going toward sales and marketing, while some of the clearest returns showed up in back-office automation.

They spent on the visible work. The payoff was hiding in the boring work.

So instead of asking, What can I automate?

Look for where the work is piling up.

Drafts waiting for review. Customers waiting for onboarding. Projects waiting on an approval. Decisions waiting on one person.

That's Herbie. But finding Herbie tells you where to look. It doesn't tell you why Herbie is slow.

"I'm the bottleneck" isn't specific enough

Now, I know what you may be thinking.

"I already know what the bottleneck is. It's me. I just don't have enough time in the day."

You're probably right. But "me" isn't something you can fix. You need to know where your time is actually going. So start with whatever takes the most of it and keep asking why.

Toyota popularized this as the 5 Whys method. It's about as complicated as it sounds.

Here's a real one from a client:

"I spend too much time on email."

Why?

"I sift through noise all day trying to figure out where I actually need to respond."

Why?

"I see a lot of emails that I don't really act on."

Why?

"They all come into the same inbox."

Stop.

We only needed 3 Whys.

You don't stop when you hit 5. You stop when the answer turns into something you can actually go fix. What looked like an email problem was really an inbox filtering problem.That's much more solvable.

But this is also where people jump too far ahead. You found something that sounds like the cause. That doesn't mean you understand the process yet. 

Don't automate the hypothesis. Validate the process first.

The details are where the bottleneck hides

I saw this firsthand at a workshop with Anthropic's engineers.

These are the people building Claude, so I expected their answer for understanding a complicated system and uncovering problems within it to involve some better AI system.

It didn't. Instead, they went back through what actually happened step by step, looked at the underlying details, and checked the logs themselves by hand.

Which sounds almost stupidly simple, but that's the point.

You can't validate a process from a compressed view of it.

A summary can tell you what the process looks like at a high level, but it leaves out the individual steps, handoffs, delays, rework, and exceptions that explain why it's slow.

Who waited on who. What got sent back. What happened out of order. Which "one-off" exception happens three times a week. Where somebody had to stop and ask for approval.

Those are the details that show you how the process actually works. And the only way to truly understand it? You guessed it, rolling up your sleeves.

You have to get uncomfortably close to the process.

I call this Sauron's Eye, after the literal Eye of Sauron in Lord of the Rings that fixes its gaze on one thing and ignores everything else (iykyk).

Pick one system. Then put all of your attention on what actually happens inside it.

Not the whole business. Not ten workflows at once. One.

And when I do that, I look at the process three different ways.

> Their version. What does the customer, manager, or teammate actually experience? Buy your own product. Trace your manager’s request to its response. Complete your teammate’s next step using only their handoff.

> Your version. What do you literally do? Write every task down in the order it happens. Not the version you remember. The version you actually perform.

> The written version. Pull up the SOP or process doc. Go line by line and mark everything that isn't true anymore.

The gaps between those three versions are what Sauron's Eye is looking for.

One of my favorite ways to check this is to have Claude draw a Mermaid diagram, basically a flowchart, of how the process works today.

Then challenge every box.

• Does that step actually happen?

• Is that really the next step?

• Who owns it?

• Where does it go when something goes wrong?

• Where does the work wait?

I do this with clients who run their own businesses and they're regularly amazed when they realize they don't actually know how one of their own systems works.

Which makes sense. Processes drift.

Someone added a workaround. A handoff changes. One person starts reviewing something twice. The documentation stays exactly the same. 

Six months later you're trying to improve the process you think you have instead of the one that's ACTUALLY running.

Automate before you know how the process actually works, and you just make the wrong process faster.

AI comes last

Boris Cherny, who runs Claude Code at Anthropic, has a simple piece of advice:

"Find the bottleneck, solve the bottleneck."

The important part is the order. Find it first. Solve it second. Then build the automation.

Most people do almost the exact opposite. They find the easiest task. Or the most repetitive task. Or the thing somebody made look cool in a demo. Then they automate it and wonder why their day feels basically the same.

The best automation isn't necessarily the one that saves the most minutes. 

It's the one that removes the thing everything else is waiting on. And if it works, something interesting happens. Your bottleneck moves.

Say writing was holding everything up. You fix writing. Now review might be the slow step. Fix review and approval might become the slow step. That's not a failure. That's what improvement looks like. 

The goal isn't to remove the bottleneck forever. You can't. The goal is to keep finding it and moving it.

After I automate something, I want two things answered:

Did work stop piling up where I made the change? 

And 

Where is it piling up now?

Then we go again and again.

I was the owner of a 15 person 7 figure agency, while having a full time job.

I was the COO of a tech startup that raised over $30M.

I'm now 9 months into my most recent business, have 13 people and just getting started.

This is the exact process I run to solve problems, and it scales, I know that for certain. But there are two major things to consider after you decide to automate anything…

Two buckets before you build

There's one final filter I use here.

Every automation should fit into one of two buckets.

Constraint-driven: it fixes something limiting the system. A bottleneck, recurring blocker, or step everything else is waiting on.

Enhancement: it makes something you're already doing cheaper, faster, or better, but nothing was waiting on it.

Both are worth building.

But if you've got AI everywhere and you're wondering why your actual output hasn't changed, start with the first bucket.

And if you can't explain what constraint the automation removes or what measurable thing it improves, don't build it.

It's that simple.

So this week, do two things.

1. Find your Herbie. Take whatever is eating your time and keep asking why until you hit something you can actually change.

2. Go look at it. Pick one process. Diagram how it works today. Challenge every box. Find out why that step is actually slow. Then automate it.

AI has compressed the time, but it can't remove the work.

Cheers, see y'all in the next newsletter.

Whenever you're ready, here's how I can help:

For Business Owners & Executives:
1. Apply for my Executive AI Coaching Program: Linked here
2. Build a Team of AI Agents that save you 10+ hours a week: Linked here

For Everyone:
3. Use BuildPartner.ai to build faster with Claude Code (try free): Linked here