Run in order. Prompts 1-4 produce the delegation queue — every task lifted off the founder needs a category before it gets a plan; prompts 5-7 build the AI workflow for the first task in the queue and ship it inside this module’s 7-day window, not “soon”; prompts 8-10 write the first-human-hire spec or the explicit hold decision, with economics that don’t pretend the hire is cheaper than it is. The deliberate weighting: AI-first delegation gets prompts 5-7 because for most founders this is the unfamiliar move and the highest-leverage one. Human hiring gets prompts 8-10 because the founder will spend the next twelve months refining it whether or not this module teaches it well. Don’t move to prompt 5 until prompts 1-4 have produced a queue with every task categorised and the first task chosen with a written reason.

1. The delegation queue — every recurring task off the founder’s week, with hours, not vibes. “Pull from M9’s handoff readiness audit (M9 worksheet page 6) every task currently inside the system that is founder-only. Add to it any recurring task that takes founder time more than twice a month but didn’t make the M9 list — these usually hide in ‘small admin’ and quietly consume 6-12 hours/week in aggregate. For each task: the actual hours/week it consumes (measured for one real week, not estimated — estimates run 50-70% low), the frequency (daily / weekly / monthly / ad-hoc), the cognitive load when it has to be done (low / medium / high — interrupting deep work has a different cost than slotting between calls), and the bottleneck it creates downstream when the founder is the only one who can do it (what else doesn’t happen this week if this takes 4 hours). Output as a 5-column table sorted by hours/week descending. Minimum 15 rows — if your queue has 6 rows you are missing the small recurring tasks that aggregate into the founder’s lost evenings. The hours total at the bottom is the M11 prize: what’s recoverable if the queue is worked through honestly.”
2. The task categorisation — AI / human / system / outsource, before any plan. “For each task in the delegation queue from prompt 1, assign exactly one category: (a) AI — the task has a definable input, a definable output, a checkable quality bar, and the judgement required is pattern-recognition or first-draft-then-edit rather than novel judgement. Examples: drafting content from briefs, summarising long documents, first-pass research, customer service tier 1 against an FAQ, data extraction from PDFs, calendar coordination against rules. (b) Human — the task requires real-time judgement, real human voice in real-time conversation, physical presence, or relationship continuity with a specific person on the other side. Examples: closing a $5K deal, leading a partnership negotiation, in-person fulfilment, mentoring another human. (c) System — the bottleneck isn’t a missing person or a missing AI, it’s a missing structure that holds state. The fix is a database, a CRM, a dashboard, a scheduled automation, a written process — not a hire. Examples: ‘remembering where each bespoke deal is in the pipeline’, ‘tracking which customer asked for what last quarter’, ‘knowing which supplier deposit is due when’. (d) Outsource — the task is specialised enough that a freelancer or agency does it better and cheaper than building internal capacity. Examples: accounting beyond bookkeeping, legal contracts, specialised design, regulated compliance. Output: the same 5-column table from prompt 1 with a 6th column for category. If a task could plausibly be two categories, choose the cheapest one — most founders default to ‘human’ when ‘AI’ or ‘system’ is available and end up with a heavier payroll than the operation needs.”
3. The queue priority — what gets lifted first, what waits. “Re-sort the categorised queue using three priority weights, applied in order: (a) hours/week recovered — the biggest founder-time line items go to the top, because every hour back compounds into capacity for the next delegation; (b) downstream unblock — tasks where the founder being the bottleneck is also blocking someone else’s work go up regardless of raw hours (if the team’s content production is gated on the founder approving every draft, the 90-min/week ‘approval’ line is actually a 20-hour/week organisational drag); (c) cost-to-delegate — within the top of the queue, prefer the task with the lowest delegation cost (AI workflows ship in days, SOPs in weeks, human hires in months). Output: the top 5 tasks in lift-off order, each with a one-line justification of why this one before the others, and the realistic week-by-week schedule for lifting them in sequence (week 1: lift task 1 via AI workflow; week 3: lift task 2 via SOP + part-time human; week 6: lift task 3 via system build; etc). The discipline: lift one task fully before starting the next. Founders who try to delegate five things at once end the quarter with five half-delegated tasks the founder is still doing badly and a team that has not been given a clean handoff on any of them.”
4. The irreplaceable list — what the founder is going to be doing MORE of after M11, not less. “Pull from M9 prompt 10 the tasks labelled ‘never’ on the founder-only column — the work that requires the founder’s judgement, the founder’s voice, or the founder’s network specifically and is structurally not delegable. Cross-check against three filters: (a) is this honestly never delegable, or is it ‘I haven’t tried because it feels too personal’ (most founder-only-voice work can in fact be partially delegated to AI trained on the founder’s voice rubric from M6 — the founder edits rather than drafts); (b) is the irreplaceable list small enough (10-15 hours/week of irreplaceable work is healthy; 35+ hours/week is the founder hiding behind ‘only I can do this’ to avoid building the SOPs); (c) does the irreplaceable list match what the business actually needs from the founder over the next 12 months (strategic decisions, key relationships, brand voice, capital, hiring of senior people) rather than what the founder happens to enjoy doing. Output: the refined irreplaceable list (target: 10-15 hours/week), the justification per line, and the explicit time those hours should be protected to over the next quarter. The point of M11 is not less work for the founder. It is more of the right work. If the founder lifts 20 hours/week off the operation and immediately backfills with more operations, the delegation has failed even if it looked successful.”
5. The AI workflow spec — pick the first task from the queue and ship the workflow this week. “Take the top AI-category task from the priority queue (prompt 3). Design the live AI workflow that lifts it off the founder’s calendar this week — not ‘eventually’, this week. Spec structure: (a) the input — what triggers the workflow (a form submission, an email arriving in a labelled folder, a scheduled time, a manual paste — be specific about the trigger surface); (b) the brief — the prompt or prompt chain the AI runs against, including the voice rubric from M6, the SOP rules, the examples of good output, the explicit instructions on what NOT to do; (c) the output — the format the AI produces (a draft email, a structured table, a one-page brief, a Slack message), the destination it goes to (founder’s inbox for review, team’s Notion for editing, customer’s inbox direct after approval), and the quality bar that defines acceptable output; (d) the human checkpoint — where the founder or team reviews before the output goes live (for the first 30 days, every output gets reviewed; after 30 days, the founder spot-checks 1 in 5 if the quality has held); (e) the failure mode — what does the workflow do when the input is malformed, the AI output is wrong, or the system is down (default to escalation to a human, never silent failure). Output: a one-page workflow spec, plus the actual prompt(s) ready to paste into Claude/GPT, plus the week-1 daily checklist for the founder reviewing every output. Ship the workflow within 7 days of finishing this prompt. An AI workflow that isn’t running by day 7 is one that won’t run.”
6. The SOP that the AI runs against — the description that turns tacit knowledge explicit. “Write the SOP (standard operating procedure) for the task being lifted in prompt 5. This is the document the AI workflow runs against AND the document a future human hire trains against — the same SOP serves both, which is the structural reason AI-first delegation makes the next human hire 10x more effective. Structure: (1) Purpose — what this task achieves and why it exists (one paragraph, no jargon); (2) Inputs — what triggers this task and what data is needed before it starts (named, specific); (3) Steps — the actual decision tree, written so a smart person who has never seen the business before can execute it correctly on their first attempt (include the if/then branches, not just the happy path); (4) Quality bar — what ‘done well’ looks like, with at least one example of a good output and one example of an unacceptable output and why; (5) Escalation — the explicit list of situations where the task should stop and route back to the founder (a customer asks something unusual, the data doesn’t match the template, the request is outside scope); (6) Review cadence — when this SOP is re-read and updated (every 90 days minimum, or whenever the founder catches the workflow doing something wrong). Output: the full SOP as a single document, target length 600-1,200 words. The discipline that makes this prompt hard: writing the SOP forces you to discover the parts of the task you have been doing on autopilot and can’t actually describe. The ‘I’ll know it when I see it’ steps in your head are the ones that have to become explicit rules here. That discovery work is the whole point.”
7. The 30-day audit — did the AI workflow hold, what broke, what gets fixed before scale. “Build the 30-day audit framework for the workflow shipped in prompts 5-6. The audit runs at day 30, not ‘whenever I get to it’ — written into the calendar before launch. Audit structure: (a) throughput — how many outputs did the workflow produce in 30 days vs the projection (under-delivery usually means the trigger is wrong, not the AI); (b) quality rate — of the outputs reviewed, what % were shippable as-is, what % needed light edit, what % needed rewrite, what % were unusable (target: 60%+ shippable as-is by day 30; if below 40%, the SOP is the problem, not the AI); (c) failure modes — every escalation and every wrong output catalogued by root cause (input quality, SOP gap, AI limitation, edge case the SOP didn’t anticipate); (d) founder hours actually recovered — measured, not estimated, against the projection from prompt 1 (gap between projected hours recovered and actual hours recovered usually means the founder is still doing the work in parallel ‘just in case’ — a delegation only works when the founder stops shadowing it); (e) the next iteration — the 3 changes to the SOP or workflow that will be made in the next 30 days, ranked by which fixes the most-common failure mode. Output: the audit template the founder fills in at day 30 plus the calendar entry. Without the scheduled audit, the workflow drifts — wrong outputs become normalised because the founder stopped reading them after week 2, and the system silently produces low-quality work the founder no longer sees.”
8. The first human hire — specialist, integrator, or hold for another quarter. “After the top AI-delegations have lifted what they can lift, name the first human hire — or name the explicit reason there isn’t one yet. Diagnostic: (a) which task at the top of the queue (prompt 3) is genuinely human-category and is consuming more than 8 founder-hours/week; (b) is that task a specialist workstream (paid ads, content production, fulfilment, customer success, sales) where the right hire is someone who is already excellent at exactly that one thing, or is it actually ‘running the SOPs the founder has built across multiple workstreams’ where the right hire is an integrator who knows how to operate other people’s systems; (c) what is the realistic fully-loaded annual cost of that hire at the level of judgement actually needed (not the level the founder hopes will work — entry-level hires require 6-9 months of founder mentoring before they produce; senior hires require 3-4x the salary but produce from week 2); (d) does the business’s current cash position (from M9 / M10) support that hire for 12 months with a 3-month productivity ramp baked in; (e) if not, what AI delegations or system builds can substitute for another 90 days while cash builds. Output: the first hire specification — role title, the one task it lifts off the founder in the first 30 days, the 3 tasks it lifts off the founder in the first 6 months, the fully-loaded cost, the start date — OR a written hold decision with the explicit cash threshold that triggers the hire later. Do not output ‘a generalist who can help me out’. That hire fails 80% of the time, regardless of the founder.”
9. The hire’s first-30-days plan — the SOPs they run on day 1, the work they own by day 30. “For the hire specified in prompt 8 (or the hire that will follow when the cash threshold trips), write the first-30-days plan. Most founder-led businesses lose their first hire within 6 months not because the hire was wrong but because the first 30 days were unstructured. Plan structure: (a) day 1 SOPs — the 3-5 SOPs from the M11 worksheet (built in prompt 6 and across the queue) that the hire is reading, executing under supervision, and giving feedback on by end of week 1; (b) week 2 — the first task the hire owns end-to-end with founder review on output only, not process; (c) week 3-4 — the second and third tasks owned, the founder review tightened to spot-check; (d) day 30 milestone — the explicit list of tasks the hire owns without daily founder involvement, plus the weekly check-in cadence (60 minutes, fixed time, agenda template) that replaces founder oversight; (e) day 30 trigger — the explicit decision point of whether this hire works (criteria: do they execute the SOPs to quality, do they raise the right escalations, are they making the founder’s calendar lighter or heavier net of the time spent managing them — if heavier, the hire is wrong for the role even if they are smart). Output: the 30-day plan as a single page, ready to share with the hire on day 1. Hires given this plan on day 1 outperform hires given ‘figure it out, ask if you need anything’ by every measure that matters.”
10. The audit cadence — the quarterly review that catches SOP rot before the next quarter. “Build the M11 quarterly audit — the review that catches the slow rot in delegated work before it becomes a quality failure visible to customers. Cadence: every 90 days, 2 hours blocked, written into the founder’s calendar for the next 4 quarters now. Audit structure: (1) SOP currency — which SOPs have been edited in the last 90 days, which haven’t been touched in 6 months and are likely silently wrong because the business has changed around them; (2) AI workflow performance — for each AI workflow, the day-30 quality metric from prompt 7 re-measured (quality drops in AI workflows usually mean a context that has shifted — the brand voice updated, the product mix changed, the customer profile evolved — without the prompt being updated to match); (3) human hire output — what did the hire own this quarter, what got promoted into bigger ownership, what should be re-allocated because it’s not working at the level needed; (4) new tasks creeping back onto the founder — the founder lists every task they did this quarter that should be on someone else’s plate by now and assigns the lift-off (re-running prompts 5-9 on each); (5) the irreplaceable list re-test — has the irreplaceable list grown or shrunk since last quarter; growth means the founder has taken on new founder-only work that may need scrutiny, shrinkage means more has been successfully delegated and the freed hours need to be reallocated deliberately not absorbed into more operations. Output: the audit template plus 4 calendar entries (next 4 quarters, 2 hours each, agenda pre-filled). The audit is the difference between a delegation system that compounds and one that decays. Most delegation projects fail at month 5, not month 1 — and the fix is the scheduled review the founder didn’t put in the calendar at launch.”