Substitute, Augment or Elevate: What to Hand to AI

Hook

You have built a workflow. You know AI can help with almost everything on your desk. That is exactly the problem. "Almost everything" is not a plan, it is a paralysis and most executives freeze at the same question: where do I actually start, and how much of this task do I hand over?

There is a one-pass answer. Every task you touch in a week falls into exactly one of three tiers: Substitute it entirely, Augment your own judgment with it or Elevate past it, using AI to clear the time so you spend more of your week on the part only you can do. Sort your task list into those three buckets once, and the paralysis question stops being a question.

Context

The instinct most executives have is to pick one relationship with AI and apply it everywhere: full delegation ("AI just does it now") or full caution ("I review everything myself"). Both instincts are wrong for the same reason. They treat AI's capability as the only variable, when the variable that actually decides what to do with a task is what kind of value that specific task creates.

McKinsey is explicit that this number is "technical potential in tasks, not the inevitable loss of jobs." That distinction is the whole lesson. Just because a task can be automated does not mean it should be substituted. Some of that 57% is exactly the mechanical work you want off your plate entirely. Some of it is work where the automatable part is only the first draft, and your judgment on the output is still the point. And a meaningful slice of your calendar was never a candidate for automation in the first place, because the task is your judgment, not a byproduct of it.

That is the level this framework operates at. Not "should I use AI in my role." Task by task, three tiers, one decision each.

Steps

Step 1: Ask two questions about the task in front of you

Before sorting anything, ask these two questions about a single task:

1. Is it rules-based and repeatable, with an answer that is right or wrong rather than a matter of judgment? A meeting summary, a first-pass data cleanup, a standard policy document, a routine status update: yes. A succession decision, a hard performance conversation, a call on whether to walk away from a deal: no.
2. Does getting it right depend on context that lives in your head and nowhere else? Institutional history, a relationship, a political read of the room, an unwritten judgment about what your organization can actually absorb right now: yes, that context is yours. A defined format applied to defined inputs: no, that context is portable.

Your answers to these two questions place the task in exactly one of the three tiers below. There is no fourth category and no task that is exempt from the sort.

Step 2: Substitute, when the task is rules-based and the context is portable

If the task is repeatable and does not depend on context only you hold, let AI do the whole thing, start to finish, with a light spot-check rather than a full review.

Concrete Substitute-tier work: first-pass meeting notes and action items, routine data formatting and cleanup, drafting a standard policy document from an approved template, generating a job description from a role's requirements, summarizing a long document into a structured brief. (The document processing pipeline from the previous lesson in this track is built for exactly this tier.)

Step 3: Augment, when the mechanical work is heavy but the call is yours

If the task has real mechanical weight, research, drafting, first-pass analysis, but the decision at the end depends on judgment only you should exercise, let AI do the heavy lifting and keep the decision.

Concrete Augment-tier work: a workforce-planning memo where AI drafts the options and you choose the recommendation, a board narrative built from an AI-drafted first pass that you shape and own, a hiring shortlist where AI screens for the criteria you set and you make every advance-or-reject call. The rule from this track's lesson on your first AI workflow still applies here: define the output format before you write the prompt, and a human makes the decision the output feeds.

Step 4: Elevate, when the task is the judgment itself

Some tasks are not candidates for AI assistance at all, because the task is not the mechanical work around the judgment call. The task is the judgment call. A hard conversation with a direct report. Reading a room during a sensitive reorg. Holding a board's confidence through a leadership transition. Telling the CEO what they do not want to hear.

You do not hand these to AI, not even partially. Instead, you Elevate: use AI to Substitute and Augment everything else on your calendar, and redirect the hours that frees up toward this tier. Elevate is not a task-level instruction to AI. It is a calendar-level decision you make once you see how much of your week the first two tiers actually free up.

That gap between believing AI matters and actually deploying it is, in practice, a Substitute-and-Augment gap. The tasks that could free up hours for the Elevate tier are sitting undone, not because the judgment work is being neglected, but because nobody has done the sort that shows where the freed hours would come from.

Step 5: Run the sort on your actual week, not a hypothetical one

Take your real task list for the next seven days, not a generic list, the actual items on your calendar and to-do list. Tag every one: Substitute, Augment or Elevate. Most executives who do this exercise honestly find the same pattern: 40 to 50% of their week is Substitute-tier work they are still doing by hand, 30 to 40% is Augment-tier work where they are either over-reviewing or under-reviewing the AI-assisted draft, and 10 to 20% is genuine Elevate-tier work that consistently gets crowded out by the other two.

Move the Substitute tier to AI this week. Fix the review discipline on the Augment tier using the trap check in Step 3. Protect the Elevate tier on your calendar like it is a client meeting, because it is the one tier no one else in the building can do for you.

Recap

• Every task falls into exactly one of three tiers: Substitute (rules-based, portable context, AI does it end to end with a spot-check), Augment (heavy mechanical work, judgment stays yours) or Elevate (the task is the judgment itself, protect the time rather than delegate the task).
• 57% of US work hours are technically automatable today, but McKinsey is explicit that technical potential is not a mandate to substitute everything. Some of that 57% is Augment-tier, not Substitute-tier.
• The most common misclassification is letting an Augment-tier task quietly slide into Substitute, approving AI drafts without genuinely weighing them.
• Elevate is not something you ask AI to do. It is the calendar decision you make once Substitute and Augment free up real hours.
• Sort your actual task list for the coming week into the three tiers today. The paralysis of "where do I even start with AI" resolves the moment the sort is done.

You have completed the AI Workflows track. You now have the four core capabilities that separate AI users from AI-powered professionals: the workflow mindset, the chaining patterns, the document processing pipeline and, now, a one-pass way to decide what any task actually gets from AI.

The next track applies these foundations to the work everyone is talking about and few can define: what an AI agent actually is, how it connects to your software and when it is the right tool.

Back to AI Workflows track →