Framework
Goal Vector
A method for decomposing strategic intent into the task specifications that AI agents can execute — preserving alignment from executive objective to agent instruction.
The problem
The gap between executive intent and agent instruction is where most enterprise AI projects fail. A strategy document says "accelerate customer onboarding." Six months later, the deployed agent has optimised completion rate at the cost of accuracy, and the organisation is managing a wave of re-onboarding requests. The problem was not the model — it was that the objective was never translated into a precise, measurable task specification.
The framework
Goal Vector is a four-step translation process: (1) Express the strategic objective as an outcome metric with a direction and a threshold. (2) Identify the constraints that bound acceptable behaviour. (3) Specify the success criteria the agent can measure locally. (4) Define the escalation condition — the signal that the local success criteria is diverging from the strategic objective. The output is a one-page task specification that travels with the agent through its lifecycle.
When to use it
Use Goal Vector at the design stage of any agentic deployment, and again at every major change to the agent's operating environment. It is particularly valuable for multi-agent systems where intermediate agents receive instructions from orchestrators — each hand-off is a potential alignment loss point.
What success looks like
A team that has applied Goal Vector can trace any agent action back to a strategic objective in under five minutes. When an agent produces an unexpected output, the first diagnostic question — "does this satisfy the escalation condition?" — has a clear answer. The framework makes alignment visible and auditable, which is the prerequisite for governance that scales.