· The Agent Playbook
The matrix replacing hire or contract.
Plot every role on two axes. Strategic value vertical, predictability horizontal. The cell tells you the mode.
Default is bot.
If a task is definable in a prompt, it belongs in the bottom right unless evidence says otherwise.
Build only what is strategic and weird.
Top left should never exceed 18% of spend, or you are subsidising commodity work.
Borrow before you buy.
Fractional access reveals true scope. Conversion to acquisition is cheaper after six months of need.
Replot every quarter.
Capability moves diagonally as model costs fall. Today's Build role is next year's Bot.
What an agent workforce actually is.
The dialogue collapsed agents into a single category. They are not. The deployable architecture has four layers with separate ownership, governance, and economics.
Reasoning agents released without orchestration above them produce unsupervised judgement at scale. The most expensive failure mode in the matrix.
Not the CIO. Not the CHRO. The role deciding which decisions humans keep, which they delegate, and which they audit is a new C suite function. Without it, layers one through three drift.
Cost per task, human versus agent.
A representative sample across knowledge work. Costs include model spend, oversight load, and revision cycles. The rightmost column is what most finance teams miss.
| Task | Human cost | Agent cost | Per task cut | Volume × | Net impact |
|---|---|---|---|---|---|
| Sales call summary | $8.40 | $0.12 | 14× | −71% | |
| First draft contract | $42.00 | $1.80 | 6× | −74% | |
| Tier 1 support ticket | $5.20 | $0.08 | 22× | −66% | |
| Software unit test | $22.00 | $0.30 | 11× | −85% | |
| Compliance review | $120 | $18.00 | 2× | −70% | |
| Strategic plan section | $580 | $220 | 1.4× | −47% | |
| Board narrative | $1,400 | $1,150 | 1.0× | −18% |
How to choose, in four questions.
A single decision tree your Chief Digital Orchestrator should run before approving any net new headcount. Stop at the first yes. The exit determines the sourcing mode.
How a CDO spends a quarter.
Four cadences. Each with a single owner, a single deliverable, a single exit. This is the operating model the evidence associates with an orchestration owner.
Read the agent telemetry.
- Task volume by agent vs human
- Escalation rate per workflow
- Cost per task drift
- Identify next migration candidate
Move one role across the matrix.
- Pick one Build candidate
- Pilot in Borrow or Bot
- Validate quality at 95%
- Recast headcount plan
Redraw the capability heat map.
- Replot all roles on matrix
- Update model cost assumptions
- Review failed pilots
- Brief the board
Redesign the org for next year.
- Set Bot ratio target
- Reset hiring caps by quadrant
- Approve infrastructure spend
- Lock the comp framework
What the evidence points to.
Across 250+ episodes of Moonshots, the recurring pattern is that work is being re-sourced, not simply automated. Three takeaways follow from that.
Three observations the evidence supports.
Four sourcing modes, not two
The hire or outsource binary is widening. The recurring framing in the source treats Build, Buy, Borrow and Bot as parallel options for the same task. Each role becomes a sourcing decision rather than a fixed headcount line.
Agent economics vary by task
Cost per task is the discriminating variable. The evidence suggests agents are favourable for high volume, well bounded work and weaker on judgement heavy or sparse tasks. The unit of analysis is the task, not the job title.
Orchestration needs an owner
A mixed human and agent workforce does not coordinate itself. A commonly cited gap is the absence of a single accountable owner for agent operations. One coined label for this role is the Chief Digital Orchestrator: the person who owns the operating rhythm across modes.
sourcing decision.
The matrix is a lens.
Plotting roles across Build, Buy, Borrow and Bot surfaces where the same task could be sourced more than one way. It is a way of reading an org, not a prescription.
Pilots precede claims.
The evidence favours measuring cost per task on a small number of high volume workflows before drawing conclusions. Forecasts of wholesale displacement remain forecasts.
Rhythm beats one off moves.
Where the pattern holds, a recurring operating cadence with a clear owner does more than any single migration. Coordination is the constraint.
About this briefing.
One of five briefings distilled from 250+ episodes of "Moonshots with Peter Diamandis and the Moonshot Mates". Claims are attributed to that source; forecasts are marked as forecasts.