Skip to content
Kiberon LabsAgentic Engineering Ebook

Minimal Goals

Description

Section titled “Description”

Minimal goals are the smallest durable set of objectives an autonomous system needs in order to direct itself over time. They are the goal layer behind minimal-bootstrap: if the system has no durable goals, it can execute work but cannot self-direct improvement.

Requirements

Section titled “Requirements”

1. Dedicated Self-Improvement Goal

Section titled “1. Dedicated Self-Improvement Goal”

At least one goal must explicitly target improving the system itself. If self-improvement is only implicit, it is too easy for immediate production work to consume all attention and compute.

This goal should be protected by its own budget, priority, or scheduling rule so it continues to receive computation even when higher-priority external work exists.

Concrete KPIs

Section titled “Concrete KPIs”

Every self-improvement goal needs measurable success criteria. The system must be able to tell whether a change improved the target condition or not.

Useful KPI classes include:

  • task success rate
  • recovery rate after failure
  • latency to completion
  • cost per successful outcome
  • review acceptance rate
  • regression rate
  • percentage of goals completed without human intervention

2. Continued Existence Goal

Section titled “2. Continued Existence Goal”

The system must also preserve its ability to continue operating. Self-improvement that destroys core execution, observability, or recovery capability is not progress.

This means the goal layer must treat survival as a first-class constraint alongside optimization.

Why This Matters

Section titled “Why This Matters”

Goals are the system’s direction. Without them, feedback has nothing to optimize toward and triggers have nothing durable to resume. A system may be capable of action, but not of progress.

Section titled “Related Concepts”