AI In Business: Directing Intelligence · Domain Playbooks
What should AI do in each business function — and what should it not?
14 domains, 85 processes, 491 stages — each stage scored with a four-verdict instrument, with the reasoning and the blockers on the record. Pick a domain to explore its playbook.
- Automate
- An agent does it; humans spot-check.
- Draft
- AI produces; a named human owns the result.
- Assist
- AI at the edges of a human task.
- Avoid
- Not yet — and the blocker is named.
The landscape
Every process from the fourteen playbooks, placed by how much of it AI can own and the Delegation Stack rung it needs: Prompt, Skill, Agent or AI Teams. Agent-rung processes sit in two lanes — a fixed workflow when the path is known in advance, a bounded agent when the next step depends on what it finds. Note the empty bottom row: no core business process is a one-prompt job. Product, the top rung, isn’t plotted: these are processes to delegate, not products someone operates.
The playbooks
Customer Support & Success
6 processes · 32 stages
AI should own the middle of every ticket (classify, retrieve, draft) while humans own the two ends — the policy that constrains it and the exceptions that escape it.
Data & Analytics
6 processes · 33 stages
AI is a superb query-writer and a dangerous analyst — automate the mechanical translation layers aggressively, and keep humans owning every step where a subtle error produces a confident, wrong number that executives will repeat.
Executive & Administration
6 processes · 34 stages
AI can produce almost everything this layer produces, and can own almost none of it — the leverage is in reclaiming synthesis hours to spend on judgment and relationships.
Finance & Accounting
6 processes · 34 stages
AI eats the reconciliation and drafting middle of the pipeline fast, while attestation, judgment calls on estimates, and anything auditors will re-perform stay Draft-at-best.
HR & People
6 processes · 33 stages
automate the paperwork around people decisions; never let the model make, or appear to make, the decision.
Information Systems & IT
6 processes · 36 stages
automate the reading and the drafting aggressively, keep humans on the writes to production and identity — and remember that AI agents are now themselves IT assets that need accounts, permissions, and offboarding like any employee.
Legal & Compliance
6 processes · 35 stages
AI compresses legal reading by an order of magnitude and legal drafting by half, but privilege, sanctions risk, and the duty of candor keep a licensed human's name on everything that leaves the building.
Marketing & Content
6 processes · 36 stages
AI has collapsed the cost of producing marketing assets to near zero, so the scarce skills are now positioning judgment, taste, and knowing which of a thousand cheap variants to ship.
Operations & Supply Chain
6 processes · 35 stages
AI excels at watching, reconciling, and drafting across operations' enormous document and data flows, but every verdict must respect a hard constraint — the cost of an error here is physical, and physical errors don't have an undo button.
Product & Strategy
6 processes · 32 stages
AI has collapsed the cost of the synthesis and drafting layers, which means the product manager's job is shifting from producing documents to owning the calls those documents encode — and the fastest way to fail is to let fluent AI drafts substitute for talking to actual customers.
Sales
6 processes · 36 stages
AI gives every rep a research team and a chief of staff, and the sellers who win are the ones who spend the reclaimed hours in front of customers instead of generating more automated noise.
Software Development
7 processes · 40 stages
code is the most verifiable artifact in business (it compiles, tests pass or fail, CI is binary), so AI can be pushed harder here than in any other domain — *provided* the verification layer (tests + CI) is strong, because that layer is what converts "AI wrote it" from a risk into a non-event. Where verification thins out — ambiguous requirements, architecture bets, production deploys — humans stay in the loop, and that's exactly where you'll feel it in weeks 9-14.
Strategy
6 processes · 38 stages
Deep-research agents have made the analysis layer nearly free, which moves the scarce skill upstream to framing the question and downstream to challenging the synthesis; the bet itself — the choice, the conviction, the accountability — remains stubbornly, correctly human.
Technology & Product Evaluation
6 processes · 37 stages
AI dramatically accelerates the *processing* of evaluation material (summarizing, scoring, cross-referencing, drafting test plans), but the *evidence-generating* steps — talking to users, running your own tests, checking references — must stay independent, because an AI summarizing a vendor's marketing inherits the vendor's framing.