Funding

Who pays for the platform decides how people behave.

Almost no operating model literature treats the funding model as a design variable. It is filed under accounting. The exception is the MIT CISR operating model work, whose decision band names funding as a first-class choice: central, chargeback, or hybrid. Primary source This page takes that choice and applies it to the agent era, where the platform is a model nobody in the building controls and the failures land in cells the budget never reached.

SourceRoss, Weill, and Robertson, 2006 Primary source

Cells06 Resources and 12 Incentives

Applied inProtocol step 07

I / The choice

Three funding models, and the cells each one starves

Who pays for a shared platform changes behavior more reliably than who reports to whom. Each model buys one kind of good behavior and quietly taxes another. The starved cells are where to look first when a deployment on that model goes wrong.

Central

The center pays. Units consume the platform at no visible cost.

What it does to local behavior Adoption is fast because nobody is charged for it, and the platform team hears about demand as complaints rather than as orders. Units treat capacity as free, so the queue grows until the center rations by politics.

Characteristic failure Nobody local owns the cost of an agent that runs badly, so nobody local funds the evidence that would show it. The center pays for the platform and the unit keeps the savings, and the Proof band stays blank because it belongs to neither.

Starves

12 Incentives 10 Evidence

Chargeback

Each unit pays for what it uses, at a price the center sets.

What it does to local behavior Units economize and the platform team gets a demand signal it can plan against. The price also makes the platform something a unit can decline, and the cheapest way to avoid the charge is to build a copy or buy around it.

Named failure Units avoid the shared platform to avoid the charge, and integration dies quietly. Each unit's agents work, none of them share evidence or a halt path, and the failure that crosses two units has no owner.

Starves

09 Coordination 08 Learning

Hybrid

The center funds the base and units pay for what they add on top.

What it does to local behavior The most common answer and the hardest to run, because the line between base and add-on has to be redrawn every planning cycle. Where the line holds, units get the shared halt path and evidence trail free and pay only for their own variety.

Characteristic failure The line drifts. Whatever is expensive to prove becomes an add-on, whatever ships a demo becomes base, and the funding model quietly reproduces the demo-attached anti-pattern it was chosen to prevent.

Starves

02 Authority 06 Resources

II / The rule

Fund the blank cells, not the demo

Fund the blank cells, not the demo. Most agent budgets go to Composition while the failures happen in Authority and Evidence. That is the agent-era instance of the same logic: the funding model decides which cells get money, and the cells that get money are the ones with something to show. The allocation rule below runs the other way. It ranks blanks by shortest path to an incident, and funds the top of the list first.

Tick the ones that are true of your deployment. The first ticked box is where the next dollar goes. Advisory This ordering is the site's own position, not a finding from the literature.

01Authority blank. An agent is running and no named person can halt it faster than it can do harm.Klarna is the documented case: the assistant ran for fifteen months before the reversal.

02Evidence blank. The work is regulated and the system cannot produce its own record without being asked.The human oversight duty requires understanding, intervention, and halt, and each one has to be shown.

03Mastery blank. Someone is on the roster to overrule the machine and nobody has checked that they can.The credential for supervising a hybrid team does not yet exist, so the check has to be local.

04Composition blank. The human to agent split was set by the pilot, not by the work.Most agent budgets already go here, but to the tooling, not to the decision about what the tooling touches.

05The rest blank. Purpose, Structure, Strategy, Resources, Flow, Learning, Coordination, Incentives, in whatever order your blanks fall.These fail slower. They still fail.

The demo-attached cell

Cells with no visible demo attached lose to cells that have one, regardless of risk. That is the anti-pattern this rule exists to break, and it is written up in full as the demo-attached cell. A Composition cell has a screen recording. An Evidence cell has an audit trail nobody has asked for yet. In every planning cycle where the two compete, the recording wins, until the incident.

The hybrid model is the one most exposed to it, because its base-versus-add-on line is redrawn every cycle, and whatever ships a demo drifts into base. The chargeback model is exposed differently: no unit pays for evidence that only matters when two units fail together.

Where this sits in the method

The funding choice is scored under Resources and Incentives on the canvas, and applied as step 07 of the protocol. The rule above is the one-page version; the protocol step says who is in the room and how long it takes.

Protocol step 07 06 Resources 12 Incentives