Skip to content
all frameworks
09 · Original model

Agent Stack Map

Three independent decisions → eight reference patterns → a portfolio
Agent stack reference patterns and an example large-fintech portfolioThree axes, chosen independentlyHarnessloop, context, sandboxModelquality, place of inferenceToolsactions, identity, execution8 REFERENCE PATTERNScurated · not exhaustiveCore#2 own + API#4 vendor + gatewayScale#6 model routerRestricted#1 air-gapped#3 client + localHigh risk#7 plan ↔ executeSandbox#5 full SaaS#8 personal + MCPEXAMPLE TARGET PORTFOLIOLARGE FINTECHSANDBOX: NO WORK DATA OR AUTHORITYNo universal ranking: status is assigned per scenario and data class.

The argument 'our own client on our own model versus a vendor agent over an API' conflates independent decisions. Harness, model and tools are chosen separately: each axis carries its own data path, cost, failure point and vendor lock-in.

The eight configurations are recurring reference patterns rather than an exhaustive combination of three binary axes. An air-gapped use case points to configuration 1, pilot speed to 5, corporate actions to 4, and high-risk execution to 7.

Core, Scale, Restricted, High risk, and Sandbox form an example target portfolio for a large fintech. Another organization should assign statuses afresh for each scenario and data class. The company typically retains control of gateways, identity, policy, and traces while keeping clients and models replaceable.

How to use this model
01

Decompose your stack along three axes: harness, model, and tools — each is chosen independently.

02

Answer three questions: does data leave the perimeter, are internal actions required, do you have scale and evals.

03

Assign every configuration a portfolio status per scenario and data class, not once for the whole company.

Sources and related materials
Author materials
Share