Simplicity-First Guidance
VibeGov is not anti-agent.
It is anti-unearned complexity.
The default progression should be:
- workflow before agent
- single owner before multi-agent
- explicit evaluation before extra orchestration
- simple loop before layered harness
- proven need before permanent complexity
Why this matters
AI teams often add complexity too early.
They introduce:
- multi-agent splits before the task is even clear
- evaluator loops before the acceptance contract exists
- memory systems before the basic workflow is stable
- orchestration layers before a single-owner loop has been made reliable
That usually creates more movement, not more control.
The progression VibeGov prefers
1. Start with the smallest coherent workflow
Use a simple sequence first when the work can be handled as a bounded direct flow.
Examples:
- observe -> plan -> implement -> verify -> document
- review -> classify -> create artifacts -> move on
- build -> verify -> package -> release-check
2. Add an agent only when the workflow benefits from it
Use an agent when there is meaningful reasoning, adaptation, or tool-driven work to perform, not just because an agent sounds more advanced.
3. Stay single-owner as long as possible
A single responsible worker is easier to supervise, easier to evaluate, and easier to close cleanly.
Single-owner work should remain the default until there is a demonstrated need for bounded delegation.
4. Add delegation only for a real boundary
Good reasons to delegate:
- a clearly separable work unit
- a distinct competency or tool boundary
- a bounded parallel slice that still has visible supervision
Bad reasons to delegate:
- hiding uncertainty
- avoiding direct ownership
- creating the appearance of sophistication
- compensating for weak issue/spec clarity
5. Add evaluator loops before adding orchestration sprawl
If quality is the problem, the next useful control is often a better verifier or evaluator contract, not a bigger worker graph.
6. Remove stale scaffolding when it stops earning its keep
Model and process improvements should simplify the system where possible.
A control that used to be necessary but no longer adds enough value becomes governance debt.
Fast decision table
| Situation | Default |
|---|---|
| task is clear and bounded | simple workflow |
| one worker can do it | single owner |
| quality is weak but scope is still bounded | strengthen evaluation / verification |
| work crosses a real boundary | delegate carefully |
| complexity was added for past limitations that no longer apply | simplify |
Anti-patterns
Avoid these:
- adding orchestration before proving the simple loop works
- turning delegation into a substitute for supervision
- treating multi-agent as the default sign of maturity
- adding memory, evaluation, and coordination layers all at once
- preserving complexity after the original reason for it has disappeared