NEW: Granular Control with Effort Selection and Model Picker

We recently introduced predev Sprints, giving you the ability to sequence work and steer your coding agent session. Today, we are taking that control a step further. You can now define exactly how deep your agent goes on every sprint and which models execute the work.

Effort Selection and the Model Picker are now live for all customers and projects, giving high-agency engineering teams the tools they need to balance speed, cost, and architectural rigor.

Effort Level: How Deep Each Sprint Goes

Not every task requires a full research and verification pipeline. With Effort Selection, you can dictate the agent's workflow depth per sprint.

You can set the effort level using the /effort slash command:

• Auto (Default): Let pre.dev automatically route the sprint based on the complexity of the task. If you don't want to think about it, this is the recommended setting.
• Vibe (Low): A fast, direct loop with minimal ceremony. Perfect for UI tweaks, copy changes, or small features where speed matters more than process.
• Todo (Medium): A direct loop utilizing a visible task list. The agent plans its steps and works through them. Ideal for multi-step features that don't require an upfront research phase.
• Build (High): The complete predev pipeline. The agent executes full research, writes the code, and runs rigorous verification against acceptance criteria. Use this for complex features, integrations, and critical architecture.

Note: Higher effort levels use more compute credits, as the research and verification phases require additional model work.

Model Picker: Pin Models to Execution Phases

Every sprint on predev moves through distinct phases: Chat, Research, Coding, and Acceptance. Now, you can pin a specific LLM to each individual phase using the /model command.

Want to use a fast, cost-effective model for chat, but bring in heavy hitters for writing the actual implementation? You can do that.

• Chat: Interactive turns and the ad-hoc loop.
• Research: Exploring the codebase and requirements before coding.
• Coding: Writing the implementation.
• Acceptance: Verifying the result against criteria.

Typing /model in your chat walks you through the selection process: pick a phase, then assign your preferred model (e.g., GLM 5.2, Kimi K2.7, Claude Sonnet 4.6, or GPT 5.5). Set any phase back to "Default" to return to GLM 5.2.

Pro Mode: Maximum Quality on Demand

If you want the absolute highest quality across the board without configuring individual phases, use Pro Mode.

By toggling /pro, you instantly pin every phase of the agent's work to our designated Pro model, Claude Opus 4.8.

A common, token-efficient workflow: Iterate on your task using standard models and lower effort while you shape the feature. Once you are confident in the direction, flip on Pro Mode and set effort to "Build" for the final, production-grade pass. You can toggle these settings between sprints without losing any project state.

What’s Next

These new controls integrate perfectly with our autonomous Autopilot mode and individual Sprints. In our next post, we will cover how to scale this control by running multiple sprints across parallel agent sessions.

Head over to your pre.dev project to test out Effort Selection and the Model Picker today.

Let Us Know Your Thoughts and Feature Requests

We want to hear from you! We are incredibly excited to see what you build and learn how you are using In-Line Media Generation to accelerate your projects. Join the conversation and share your feedback over on our X and LinkedIn pages.

About predev

pre.dev is built to provide professional engineering teams with hyper-intelligent and cost-efficient coding agents. These agents deliver scalable, self-verifying, and model-agnostic software development.

Professional software development requires agentic coding that deeply understands system architecture, mitigates technical debt, and respects your compute budget.