Leadership with AI
Decision-making, operating models and leverage for people who run teams — with AI treated as a change to how work is organised rather than a tool to buy.
Finish this and you can redesign one of your team’s real workflows around human and machine work, defend the decision with evidence, and put governance around it that your risk function will accept.
-
Decision-making under uncertainty
IntermediateThe core leadership skill, and the one AI stresses hardest: more information, arriving faster, of unknown reliability.
- Farnam Street — Mental Models A working set of frames for decisions. Inversion and second-order thinking are the two that most change how AI proposals get evaluated.
- Thinking, Fast and Slow — Nobel lecture and papers The primary source on the biases every decision framework is trying to correct for. Denser than the popular book and considerably shorter.
- Postmortem Culture — Learning from Failure Blameless review as an operating practice, not a ceremony. Transfers directly to any team, technical or not.
Build this
A decision journal, run for a month
For every significant decision: write down what you expect to happen, how confident you are, and what would prove you wrong. Review after four weeks.
This is the cheapest possible instrument for finding out whether your judgement is actually good, and almost no leader has ever run one.
Done when: A month of logged decisions with a written review of your calibration.
-
Teams and operating models
IntermediateHow work is divided, where handoffs cost you, and what changes when part of a workflow becomes machine work.
- Team Topologies — Key Concepts Four team types and three interaction modes. A compact vocabulary for describing why your current structure is slow.
- Google re:Work Research-backed guides on managers, teams and hiring, with the underlying studies linked. The psychological safety material is the standout.
- DORA — State of DevOps research Multi-year evidence on what actually correlates with delivery performance. Useful when someone claims a tool will fix a structural problem.
-
Redesigning a workflow around human + AI
AdvancedThe practical core of this path: taking one real process apart and deciding, step by step, what a machine should do and what a person must.
- People + AI Guidebook Read the chapters on mental models and errors specifically as a manager: they describe what your team will experience when the system is wrong.
- HAX Toolkit The workbook format works well as a structured session with a team rather than solo reading.
- Stanford AI Index Report The adoption and economy chapters give you defensible numbers for a board conversation, with citations that survive scrutiny.
Build this
One real workflow, redesigned and measured
Map a process your team runs weekly, step by step, with current time and error rate. Identify which steps are judgement, which are retrieval, and which are transformation.
Redesign it, run it for three weeks, and measure the same two numbers again. Include what got worse — quality, morale, review burden — because something usually does.
Done when: A before/after process map with measured time, error rate, and an honest account of the costs.
-
Leading teams that use AI
AdvancedSkills development, review standards and the quiet questions people are afraid to ask you out loud.
- re:Work — Managers The guides on feedback and one-to-ones are the mechanics of the conversations this topic requires.
- Site Reliability Engineering — chapters on process The chapters on toil, on-call and error budgets are a case study in negotiating standards with an engineering organisation.
Build this
A review standard for AI-assisted work
Write your team's actual policy: what may be produced with AI assistance, what must be disclosed, what must be checked by a person before it leaves the team, and who is accountable when it is wrong.
Then take it to the team and change it based on what they say. A policy written alone is a policy nobody follows.
Done when: A one-page team standard, revised after a session with the people it governs.
-
Strategy, risk and governance
AdvancedTurning a set of pilots into a position — and surviving the conversation with legal, security and the board.
- NIST AI Risk Management Framework Voluntary, widely adopted, and the structure most enterprise AI governance is now being written against. Know Govern, Map, Measure and Manage.
- EU AI Act Explorer The risk tiers and what each obliges. Relevant well beyond Europe, because multinational vendors build to the strictest applicable rule.
- MIT Sloan Management Review — Artificial Intelligence Research-grounded management writing rather than vendor commentary. Some articles are metered — the research summaries are usually open.
Build this
A twelve-month adoption plan you would sign
Write the plan: which workflows, in what order, what each is expected to change, what it costs, how you will know it worked, and what would make you stop.
Include the stopping condition explicitly. A plan with no way to fail is a plan that will run for two years regardless of results.
Done when: A costed plan with success measures and a written stopping condition.
Not sure this is your path?
A path is only worth the hours if it is the right one. The counselling session tests the direction you arrived with rather than simply agreeing with it.
Find your directionClose, but not quite yours?
These nine are fixed on purpose — complete, sequenced, and ready today. If your direction sits between two of them, we will build the path around you instead.
Ask for a custom path