COST DISCIPLINE
Cost discipline, role by role.
Every role gets exactly the AI tier its work requires — heavy where reasoning matters, mechanical where it doesn't, published every cycle.
See what a cycle costs
Heavy
15%
Deepest reasoning, highest capability
The most capable model fires sparsely — once or twice per cycle. One well-reasoned architecture decision shapes weeks of execution. The investment is justified by the leverage.
strategy · design · review-of-design
Standard
55%
Execution and review
Most cycle work happens here. Bounded tasks with defined acceptance criteria — cost is predictable, not open-ended exploration.
engineering · review · platform · independent-review
Mechanical
30%
Structured transforms, no novel reasoning
Deterministic templates on bounded input. Using a heavy model for a git-log rollup is waste — like a principal engineer updating a spreadsheet.
cycle wrap-up · spec sync · cost tally
Want per-role cost attribution and the full breakdown? Token budget rules, governance cross-checks, and the cost calculator.
Cost control detail →
In wakalix, AI cost is declared at the role level, not negotiated at runtime. Every role definition carries its model tier. The tier is fixed before the cycle starts — no per-invocation decisions, no drift. Spend is logged to a metrics file and reported in the wrap-up, so you see the number before you see the invoice. If a role consistently exceeds its tier's cost envelope, the retro flags it as a process issue — not a billing surprise.
See your conventions.
How the team learns your codebase, your standards, and your way of working.
Your conventions