The direct answer
Hand the AI the work that is pattern and prep. Keep the work that is judgment. The line runs between four kinds of work: decisions that need your experience, work that needs preparation, work that needs a first draft, and work that repeats. The last two are the AI's. The first one is yours, and the system should be clearing the way to it.
The fact: McKinsey estimates generative AI could add $2.6 trillion to $4.4 trillion in value annually, with the biggest impact in knowledge work like customer operations, marketing, and software engineering.
SOURCE: McKinsey, June 2023 — Read the analysis
Notice where McKinsey drew the line too. Knowledge work, the work that starts with a context problem, is where the value sits. Not the judgment itself.
The fact: Microsoft found 78% of AI users bring their own AI to work, rather than getting it from their company.
SOURCE: Microsoft & LinkedIn, 2024 Work Trend Index — Read the report
That number is the warning label on delegation done badly. When people bring their own tools, the quality control is gone. A system, with one loop and one voice, is how you get consistency.
Work that needs your judgment
Pricing. Positioning. Who to call back first. What to walk away from. That is the $10K work. The AI can prepare everything around the decision. It should not make it.
Work that needs preparation
Research, meeting prep, briefing documents, competitor scans. This work is expensive because it is invisible: it doesn't produce anything the client sees. Hand it over. You get the same brief in minutes that used to eat an afternoon.
Work that repeats
Follow-up drafts, content calendars, weekly summaries, the same email you have written forty times. State it once. The system remembers the pattern and the voice.
Where to start
Pick one task from each bucket this week. Run the AI's two. Keep the judgment one, but notice how much time the other two just gave back.