Why I’m the Bottleneck in My Own AI Workflows

Last updated: 2026-09-17

Here’s an admission. In my own AI workflows right now, I’m the bottleneck. The work gets produced. I’m not reviewing the outputs fast enough.

Once AI can handle most of the execution, the constraint moves onto the human who has to define the goal and the human who has to approve what comes back. Right now that’s me. I know, that’s an annoying place to be stuck.

Does this sound familiar? You get the output. Then review becomes the pile.

The 10-10-80 rule

The first 10% is human. Defining vision, strategy, and constraints.

The middle 80% is AI. Execution, breaking down tasks.

The last 10% is human again. Review, approval, shipping.

The bottleneck is no longer execution. It’s your ability to define the goal and review the output. The productivity focus moves from doing to directing.

What does that first 10% actually include? Vision. Strategy. Constraints. Skip those, and the middle 80% has nothing solid to execute against. The AI can break work down. You still have to define the goal.

Audit your current projects. Identify where you’re stuck in the middle 80% (execution) versus the first 10% (definition). If you’re actually stuck in review, that’s the last 10%. That counts as a bottleneck too.

Find the one constraint

Want the fastest way to make progress? Eli Goldratt already named it in The Goal. Identify the single biggest constraint blocking you. Resolve that first. Breaking everything else down can wait.

In an AI workflow, the constraint is often the human review process. I’m living that. I’m the bottleneck because I’m not reviewing outputs fast enough.

Here’s the thing with Jevons Paradox. As AI makes certain tasks cheaper and faster, like legal drafting or coding, demand for those tasks increases. You’ll do more of them because the barrier to entry is lower. Cheaper contracts mean more contracts, which means more lawyer review time. Humans end up with more work to review and manage. Don’t assume AI will eliminate your workload. Assume it will expand the scope of what you’re responsible for overseeing.

Identify your current bottleneck. Is it lack of clarity? Lack of time? Or lack of review capacity?

A vision, then two execution styles

Now let’s get back to turning a vision into work you can actually do. You still need a vision either way. The difference is how you execute.

Brooks uses a Work Breakdown Structure as the systematic way to avoid tackling-and-blocking chaos. Borrow it from project management and apply it to personal goals. Start with the Goal. Break it into Phases (major deliverables), then Sub-tasks, then Individual Work Packages (atomic actions). That hierarchy prevents overwhelm by showing the path from vision to daily action.

Take one vague goal, like launch a course, and apply the WBS framework: Goal → Phase 1 (Content) → Sub-task (Outline) → Work Package (Write Module 1). Keep going until the phases have real work packages.

If you struggle with detailed planning, use tackling and blocking. Define a rough vision, then do the next immediate thing to unblock progress. Repeat. It’s less systematic than WBS. It’s effective for maintaining momentum when perfectionism causes paralysis.

Tackling and blocking and WBS both start with a vision. They differ in execution style. One gives you a hierarchy from goal to daily action. The other is the next unblock, over and over.

If you’re stuck in analysis paralysis, switch to tackling and blocking for 24 hours. Rough vision. Next unblock. Repeat.

The obviousness trap

When AI breaks a complex goal into steps, the output feels obvious and simple. Generating that structure from scratch on a blank page is difficult. That’s why people resist planning. The steps look too simple, so you dismiss the structure and stay on the blank page.

The value lives in the clarity of the structure. Use AI to bypass the friction of starting from zero.

Next time you feel stuck on a big project, don’t try to write the plan yourself. Prompt an AI to break this down into a Work Breakdown Structure, then refine it.

Leading versus lagging

Are you tracking the thing you can actually control?

Lagging indicators tell you what happened. Total sales. Marathon finish time. Leading indicators predict the outcome. Number of prospect meetings. Weekly training sessions.

Focus your daily energy on the leading indicator. It’s the one thing you can control to influence the lagging result.

Pick one quarterly goal and identify its single most important leading indicator. A membership growth target is a lagging result. The leading indicator is conversations you can actually hold, not outreach volume. Outreach volume can look like work. Put the daily energy on the conversations.

Notes without a third wheel

The last 10% needs a record of what was said.

Unlike bots that join as participants, Granola records computer audio locally. It matches transcripts to calendar events to identify speakers without appearing in the meeting. Ideal for 1:1s where you want a record but not a third-wheel bot. That’s my personal recommendation for a tool that respects those social dynamics.

Try Granola for your next 1:1 coaching or client call. You get a clean transcript without the awkwardness of a bot joining.

Audit one live project today. Mark whether you’re stuck in the first 10% (definition) or the middle 80% (execution). Then name the single constraint: clarity, time, or review capacity. Fix that one before you ask AI for more.