Bonus 6 — The AI-Augmented Manager Operating System
Bonus 6: redesign your manager week around AI as an operating layer — what to delegate to AI, what to never delegate, how to govern your team's AI use, and…
On this page▾
- Bonus module 6 of the 12-week program (months 4–6). Theme: AI as an operating layer, not a drafting toy.
- Monthly AI operating review — the ritual you install.
- Same rhythm as weeks 1–12: pre-read, cohort live, ritual, falsifiable homework.
- Reviewer-validated against the gap that earned this module its slot.
By 2026, AI is no longer a productivity hack; it's a layer the manager has to operate on, govern, and reason about. New managers face three new problems the core 12 weeks don't fully address: (1) how to use AI in their own workflow without outsourcing judgement, (2) how to govern their team's AI use (data, quality, attribution), and (3) how to evaluate performance when AI is in the loop. This module is the operating system, not the prompt library.
What the evidence says
- MIT/BCG field experiments (Dell'Acqua et al., 2023–2025): AI lifts performance on tasks inside its frontier and degrades performance on tasks outside it. Managers must know which is which for their team.
- Anthropic, OpenAI, and DeepMind enterprise studies (2024–2026): the manager is now the governance layer — what data goes into prompts, what outputs are reviewed, what attribution rules apply.
- Brynjolfsson et al. (NBER): AI assistance produces the largest gains for novice workers — which changes how managers structure ramp, mentorship, and review.
Pre-read (60 minutes)
- Read: the four classes of manager task — judgement-heavy (never AI), context-heavy (AI assists), pattern-heavy (AI does well), volume-heavy (AI dominates) — 20 min.
- Read: your company's AI acceptable-use policy. If it doesn't exist, that's your first homework — co-author it — 15 min.
- Read: AI in performance review — risks, bias amplification, and the auditability rule — 15 min.
- Reflect (10 min): list everything you did last week. Tag each: judgement / context / pattern / volume. Where is AI under-used? Where is it over-trusted?
Live session (90 minutes)
- 1The manager's AI map (20 min)Each manager classifies their week. Coach challenges: where are you using AI for things that need your judgement (bad)? Where are you doing by hand things AI would do better (also bad)?
- 2Prompt library for managers (20 min)Coach walks through 8 high-leverage prompts: 1:1 prep, feedback draft, perf review draft, calibration prep, comp doc, retro synthesis, change-comms draft, hiring-debrief synthesis. Each prompt has a guardrail: what NOT to put in, what to always review by hand.
- 3Governing your team's AI use (25 min)Coach walks through the team-level governance choices: which tools are sanctioned, what data is allowed in, how outputs are reviewed and attributed, what training is required. Cohort drafts a 1-page team AI charter.
- 4Performance when AI is in the loop (15 min)If AI did 40% of the work, how do you evaluate the IC? Coach walks through the principle: evaluate judgement, taste, and reliability — the parts AI didn't do. Calibration patterns need to evolve.
- 5Wrap (10 min)Each manager commits to: 2 tasks to delegate to AI this month, 1 task to take back from AI, and 1 governance artefact (charter or policy) for the team.
The ritual you install
Once a month, in your 1:1 prep or a 30-minute solo block, review: what did I delegate to AI this month that worked? What did I delegate that didn't? What is my team using AI for that I don't know about? What's our governance posture vs the new risks that have emerged? Update the team AI charter quarterly. This is the operating cadence that keeps AI from running you instead of the other way around.
Modern tools for this skill
| Category | Examples (2026) | Use |
|---|---|---|
| LLM assistants | Claude (Anthropic), ChatGPT (OpenAI), Gemini (Google), Copilot (Microsoft) | Drafting, synthesis, analysis — never delivery or judgement |
| Meeting/transcript AI | Granola, Otter.ai, Fathom, Read.ai | Transcripts, action items — review and own outputs |
| Workflow AI | Notion AI, Linear AI, GitHub Copilot, Cursor for code | In-flow assistance; governance baked into the tool |
| People-AI | Lattice AI, CultureAmp AI summaries, Eightfold | Use for synthesis; never as the deciding voice in calibration or performance decisions |
| Governance | Team AI charter, acceptable-use policy, EU AI Act conformity assessment for high-risk uses | Write it down; review it quarterly; audit annually |
Here's my team's current AI usage [tools used, data going in, outputs review process]. Help me: (1) draft a 1-page team AI charter covering sanctioned tools, data rules, review/attribution rules, and training requirements, (2) identify the top 3 risks in our current posture, (3) suggest 5 high-leverage manager workflows I should automate or augment with AI this quarter.
Homework — falsifiable artefacts
- Manager's AI map completed; 2 tasks newly delegated to AI, 1 task taken back.
- Team AI charter drafted and shared with team for input.
- One performance review draft prepared with AI assistance — and explicitly reviewed against the calibration rubric (what AI got right, what you overrode).
- Governance gap identified and routed to HR/IT/Legal as appropriate.
Success signal
By end of this module, you can explain how AI fits into your manager week without using the word 'AI' as a vague gesture, your team has a written governance posture you can defend to Legal or an auditor, and the parts of management that require human judgement (feedback delivery, hard conversations, calibration decisions) remain unambiguously human in your practice.
Reviewer notes
The AI risk for managers in 2026 isn't that they use it badly — it's that they use it for the wrong things and don't use it for the right things. The wrong things: feedback delivery, performance decisions, anything that requires the manager's own judgement as a signal. The right things: synthesis, drafting, prep, scheduling, pattern-finding. Managers who get this distinction right pull ahead fast.
I treat AI the way I treated email in 1999 — a layer that's going to be everywhere, that I have to govern explicitly, and that I have to make a deliberate choice about every workflow. The managers who refuse to engage with it are getting out-shipped by their peers; the managers who outsource judgement to it are getting found out in calibration. Middle path: AI prepares, you decide and deliver.
Brynjolfsson's work on the productivity J-curve is instructive: the first wave of AI adoption tends to produce modest gains because organisations layer it onto unchanged workflows. The gains come when managers redesign the workflow around what AI does well and what humans do uniquely well. The manager is now the organisational designer of that boundary — that's a fundamentally new part of the job.
Read next
All playbooksA twelve-week, MBA-level program for first-time managers — built around what HR actually escalates, what experienced managers actually need on Monday morning…
Audit of the 12-week program against six reviewer lenses — HR Director, Employment Counsel, 20-year Line Manager, Founder/CEO, Finance Business Partner, OB…
Week 1 of the 12-week new manager program: redesign your week around what only a manager can do, navigate the loss of peer status, and reset the relationships…