You cannot delegate what you can't describe. You can't describe what you've never written down. This phase makes the invisible work of your job visible — and tells you exactly which pieces are ready to hand off.
Most people who fail at "AI workflow building" fail here, not at the AI part. They open ChatGPT, try to delegate "marketing" to it, get a generic response, and conclude that AI is overrated.
What actually happened is they tried to delegate a job, not a task. There is no AI on earth that can do "marketing." There are many AI assistants that can do "write 3 hook variations of an Instagram caption following the brand voice rules in this document, given the product brief below."
The difference between those two sentences is the difference between this roadmap working for you or not. Phase 2 is where you learn to think in tasks, not jobs.
The smaller and sharper a task is described, the higher the chance it can be reliably delegated to an AI worker. Vague briefs produce vague outputs — from humans, and even more so from AI.
Mapping is the act of going from "what is my job" to "what are the 30 specific tasks that make up my job."
Open the doc you started in Phase 1 (AI Director — My Roadmap.md) and add a new section: "Task Inventory." Give the inventory enough time to capture representative work, exceptions, data classes, systems, and accountable owners; revise it when reality exposes a gap.
Step 1: Brain-dump everything you do.
For the next 20 minutes, write down every task you've done in the last 7 working days. Don't filter, don't categorise, don't judge. Just list. Aim for at least 30 entries. If you stall before 30, look at your sent emails, your Slack history, your last week's calendar — they're full of tasks you've already forgotten about.
Step 2: Add three columns next to each task:
Frequency — daily / weekly / monthly / one-offTime — how long it takes you (rough is fine: 5 min, 30 min, 2 hours, etc.)Type — R (routine), C (creative), J (judgment), or M (mixed)Definitions for the Type column live in the next section.
Every task you do falls into one of three categories. The category determines whether it's a candidate for delegation, what kind of AI worker fits, and how much effort the SOP will take to write.
Has a clear recipe. The output looks similar each time. The variables are small and known. Examples: weekly campaign reports, formatting documents, resizing creatives, copying SKUs into a sheet, daily standup write-ups.
Best AI candidatesHas a recipe but the output should feel fresh each time. Variables are larger. Examples: writing ad copy, designing a landing page, crafting an email sequence, brainstorming hooks, drafting a content calendar.
Strong AI candidates with good SOPsEach instance is genuinely different. Stakes are real. Requires reading a person, a situation, or weighing trade-offs only you can weigh. Examples: client negotiations, hiring decisions, strategic pivots, crisis response, performance reviews.
Keep on your plate (for now)M (Mixed) is the diplomatic answer for tasks that are hard to classify. Use it sparingly. If more than 30% of your inventory is M, you're being too easy on yourself — most M tasks can be split into an R-portion and a J-portion. Try to split them.
Not every R is worth delegating yet. The filter below tells you which R and C tasks should be your first three SOPs (which you'll write in Phase 3) — and which to defer until you have more reps.
Score each R/C task 0–5 (one point per question answered "yes / favourable"). Pick your top 3 highest scorers. Those become your first three SOPs in Phase 3.
To make this concrete, here are sample task maps for four common online-execution roles. These are not exhaustive — they're starting points to compare against your own list. If your list is shorter than these, dig harder.
Listing job titles instead of tasks. "Marketing" is not a task. "Pull weekly Meta Ads spend report by campaign and post it to the #marketing-reports Slack channel every Monday at 10am" is a task. The more your list looks like the second sentence, the better.
Hiding your boring tasks. The reflex to omit "I spend 4 hours a week formatting decks" because it's not glamorous is exactly the reflex that keeps you executing forever. Boring & repetitive tasks are gold for this exercise. List them first.
Marking everything "J" because you're proud of your judgment. Be honest. Ask yourself: "If I really had to write down exactly how I make this decision, could I?" If yes, it's likely R or C, not J.
Trying to map a task you've done once. SOPs are for things that recur. If a task is genuinely a one-off, don't map it. Move on. (You'll feel tempted to map cool one-offs because they're top-of-mind. Resist.)
For tasks you can describe but not in detail — your fingers know but your words don't — try this:
With authorization and appropriate redaction, record yourself doing the task while narrating decisions. Avoid capturing credentials, personal data, client data, or restricted screens. Treat the transcript as draft evidence: verify it, add omitted controls and edge cases, and delete or retain the recording under policy.
This is the single fastest way to extract knowledge that lives in your hands but not on paper.
By the end of Phase 2, you should have:
If you have those three artifacts, you are completely ready for Phase 3. If you don't, do not move on. Mapping is the foundation. Phase 3 will be miserable on a wobbly foundation.
You've done something most professionals never do in their entire career: you've made the invisible structure of your job visible. From here on, you can think about your work as a portfolio of tasks instead of a job — and a portfolio is something you can re-allocate.
Phase 3 turns this inventory into a testable operating guide. Clear SOP writing is important, but it works only with appropriate access controls, evaluation, review, monitoring, and ownership.