💡 Effort Modes: Economy / Power / Turbo Session Tiers

[don-petry]

Summary

Add user-selectable effort modes that let users trade cost vs. capability per session. Economy mode uses smaller/cheaper models with turn limits for quick tasks; Power is the balanced default; Turbo uses the most capable model with extended context and higher budgets for complex workflows. Estimated cost is surfaced before session start, supporting the BYOK model where users pay their own API costs.

Market Signal

Replit Agent pioneered session-tier pricing, and the pattern has since spread across AI coding tools. Augment Code, Devin AI, and ChatGPT Atlas all offer tiered usage plans in 2026. The BYOK (Bring Your Own Key) model makes cost transparency even more critical — users need to understand and control their spend per session. Fazm, one of the top desktop AI agents, uses a free-tier model with local processing, showing that cost-conscious users are a large segment.

User Signal

This idea is already captured in the PRD's "Enhancement Ideas (Post-MVP)" section as an explicit post-MVP item. The BYOK API key model (FR40-41) means users bear direct API costs, making cost visibility and control a first-order UX concern. No feature requests exist yet (the project is pre-implementation), but this was identified during product discovery as essential for the target non-technical user who may not understand token economics.

Technical Opportunity

The Claude Agent SDK already supports maxTurns and maxBudgetUsd parameters on AgentDefinition, providing the exact primitives needed. The AgentBackend abstraction port (defined in architecture as the extension point for backend configuration) can be extended to accept an effort-mode parameter that maps to SDK budget/model settings. The architecture's DDD pattern means this maps cleanly to a new value object (EffortMode) in the Agent bounded context, with the UI presenting it as a simple 3-option selector during session start. The new get_context_usage() SDK method enables real-time cost tracking during sessions.

Assessment

Dimension	Score	Rationale
Feasibility	high	SDK already has maxTurns/maxBudgetUsd primitives; UI is a 3-option selector
Impact	high	Directly addresses cost anxiety for BYOK users; differentiator vs. flat-rate competitors
Urgency	med	Important for adoption but can ship shortly after MVP since MVP targets a single workflow

Suggested Next Step

Create a product brief defining the three tier configurations (model, maxTurns, maxBudgetUsd, context window) and their cost estimates. Spike on surfacing real-time cost tracking using get_context_usage().

---

Proposed by the BMAD Analyst (Mary) on 2026-04-05.

[github-actions[bot]]

Research Enhancement — Effort Modes: Economy / Power / Turbo Session Tiers

Sharpened Problem Framing

The core problem is not just cost visibility — it is cognitive load at the moment of commitment. Non-technical BYOK users face a blank "Start Session" button with no mental model of what a session will cost or how powerful it needs to be. This creates two failure modes: (1) cost anxiety that suppresses usage, or (2) overspending on Opus when Haiku would suffice. Effort modes solve both by translating token economics into an intuitive three-tier selector with pre-session cost estimates.

Prior Art & Competitive Signal

Product	Tiering Approach	Key Lesson
Claude Code	/fast toggle between Opus modes	Proves model-routing UX can be a simple toggle, not a settings page
Replit Agent	Session-tier pricing (pioneered the pattern)	Users accept per-session cost framing when the value proposition is clear
Augment Code / Devin	Tiered usage plans (2026)	Industry converging on tiered AI access as standard
Kilo Gateway	BYOK with 0% markup + usage analytics across 20 providers	Shows BYOK cost transparency is a differentiator, not a limitation
Cline / Aider	Open-source BYOK, $50-200/month typical Opus spend	Real-world BYOK costs validate the need for budget guardrails

Current Claude Pricing (July 2026)

Model	Input/1M	Output/1M	Best For
Haiku 4.5	$1.00	$5.00	Quick tasks, classification, routing
Sonnet 5	$2.00*	$10.00*	Most production work, everyday coding
Opus 4.8	$5.00	$25.00	Deep reasoning, complex planning

*Sonnet 5 introductory pricing through Aug 31, 2026; reverts to $3/$15.

Options Table

Option	Pros	Cons	Effort
A. Static 3-tier mapping (Economy→Haiku, Power→Sonnet, Turbo→Opus)	Simplest implementation; maps directly to SDK maxTurns/maxBudgetUsd	Rigid; Sonnet 5 is close to Opus quality at 40% of the price — is a 3-tier split still the right cut?	Low
B. Smart 2-tier (Standard→Sonnet, Power→Opus) with budget slider	Simpler choice for users; reflects that Sonnet 5 covers 80%+ of use cases	Less granular; loses the Haiku cost-saving option for trivial tasks	Low-Med
C. Task-aware auto-routing with override	Routes by estimated task complexity; user can override	More complex; requires task classification heuristic	Med-High

Recommendation: Start with Option A for MVP simplicity, but design the EffortMode value object to accommodate auto-routing later (Option C). The 5-25x cost difference between Haiku and Opus makes the three tiers genuinely meaningful for BYOK users.

Risks & Unknowns

• Model pricing volatility: Sonnet 5 introductory pricing ends Sep 1, 2026 — tier economics will shift. The EffortMode mapping should be a configuration, not hardcoded.
• User comprehension: "Economy / Power / Turbo" may not mean anything to non-technical users. Consider testing labels like "Quick / Standard / Deep" that describe the task, not the engine.
• Cost estimation accuracy: Pre-session cost estimates require predicting token usage, which varies wildly by task. Consider showing ranges ("$0.05 - $0.30") rather than point estimates.
• Prompt caching interaction: Cached input costs 90% less — effort modes should account for cache-warm sessions being much cheaper than cold starts.

Phased Validation Path

1. Phase 1 (MVP+1): Implement static 3-tier EffortMode value object in Agent bounded context. Wire maxTurns and maxBudgetUsd per tier. Add 3-option selector to session start UI. Show estimated cost range.
2. Phase 2: Add real-time cost tracking using get_context_usage() — display running cost in session UI. Alert when approaching budget limit.
3. Phase 3: Add usage analytics (cost per session, per mode) to inform whether auto-routing (Option C) would save users money.

Success Metrics

• Adoption: >60% of sessions use a non-default effort mode within 30 days
• Cost reduction: Economy mode sessions cost ≤20% of Turbo sessions on average
• Retention: Users who use effort modes have ≥15% higher 30-day retention than those who do not
• Comprehension: <10% of users change effort mode mid-session (indicating they chose correctly upfront)