For teams that already use workflow-level budgeting and value tags, how does adding a simple rule—‘high-cost workflows must declare an expected outcome type before run’ (e.g., defect reduction, MTTR improvement, migration acceleration)—change developers’ willingness to invoke expensive agents, and does this refinement increase trust and repeatability or just introduce friction that suppresses legitimate high-ROI usage?

Answer

Requiring high-cost workflows to declare an expected outcome type before run usually makes developers slightly more cautious but clearer in when they use expensive agents. Net, it tends to increase trust and repeatability more than it suppresses legitimate high-ROI usage, provided the declaration is quick, non-punitive, and tied to team-level review. If the rule feels like personal pre-approval or paperwork, it mainly creates friction and suppresses high-value use.

Behaviors you typically see:

• Willingness to invoke expensive workflows

  • Drops a bit for vague, low-value tasks (people avoid the hassle of justifying them).
  • Stays steady or rises for well-understood high-ROI cases (incidents, migrations, defect hot spots) where outcome types are obvious.

• Effects on trust and repeatability

  • Trust up when:
    • Outcome types map to existing value tags (e.g., MTTR, defect rate) and reviews stay team-level.
    • The rule clarifies, “These are the accepted reasons to spend a lot,” reducing token anxiety.
  • Repeatability up because:
    • High-cost runs are clustered under a small set of outcome types.
    • Leaders can see, per workflow family, which outcome types recur and tune budgets/variants accordingly.

• Where it backfires

  • If outcome declarations are free-text, audited per-person, or tied to approval, devs:
    • Underuse high-cost workflows even when justified.
    • Revert to cheap but weaker paths to avoid scrutiny.

Net: a light, dropdown-style outcome declaration on high-cost workflows, reviewed at the workflow-portfolio level, mainly sharpens when expensive agents are used and improves governance, with only modest suppression of truly high-ROI usage. Heavy or person-centric implementations flip this: friction and fear dominate, and high-cost usage falls even where it should grow.