Large language models have changed the way we build games inside Unity, but they still suffer from human‑like forgetfulness. As tasks get longer and prompt windows fill up, models pay less attention to the details in the middle of a conversation—a phenomenon researchers call “lost in the middle.” Even when context windows grow to hundreds of thousands of tokens, quality often declines as the prompt fills. Simply giving a model more memory doesn’t stop it from drifting off course.

## Why Make Models Focus?

Most AI copilots, including Coplay, break a request into dozens of small steps: reading files, generating code, and iterating on feedback. Without a structured plan, those steps can blur together. Researchers have shown that models recall information at the beginning or end of a prompt better than details buried in the middle. As a result, long multi‑step tasks often veer off the intended path or repeat work. Bigger context windows help, but they’re not a silver bullet—performance can actually degrade when you stuff too much into a prompt.

The to-do list is our answer. Instead of letting the agent juggle dozens of implicit goals, we give it an explicit to‑do list that travels with it throughout the session. This context‑forward approach anchors every action to the goal, ensuring that even the longest coding tasks stay aligned.

## What is the To-Do List?

With the To-Do list feature turned on Coplay creates a numbered plan as soon as you issue a request. Think of it as a living to‑do list: each item represents a discrete action the agent needs to complete. As Coplay works, it references and updates this list, checking off completed steps and refining the remaining ones. The plan itself becomes part of the model’s prompt, so the agent literally “remembers” what it is doing.

Other tools have adopted simple task lists, but our implementation goes deeper. This isn’t just a UI nicety—it’s integrated with the agent’s planning and execution loops. The same model that writes your code also owns its plan, updating the list every time it makes progress. This constant grounding prevents the model from chasing tangents or forgetting earlier requirements.

## How It Works in Coplay

When you enable the to-do list, Coplay:

1. **Drafts a plan.** On your first prompt, Coplay generates a numbered list of the steps required to fulfill your request. Each step is scoped to a clear, actionable objective—like “update the enemy AI script” or “design UI for the inventory panel.”

2. **Carries the context.** The to‑do list is embedded into the model’s working context. As Coplay reads files, writes code and runs tests, it continually refers back to the plan and updates it.

3. **Checks items off.** When a step is completed, the agent marks it as done and moves on. If a new step arises—say, an unexpected bug appears—Coplay adds it to the list. The list evolves along with your project.

4. **Surfaces progress.** In the Unity editor, you can view the list at any time. Coplay highlights the current step and shows completed tasks, giving you transparency into what it’s doing and why.

Because the plan persists across the entire session, Focus Mode all but eliminates goal drift. The agent no longer relies solely on its fading latent memory—it has an explicit roadmap to guide it.

## Benefits and Looking Ahead

The to-do list delivers measurable improvements for complex, multi‑hour coding sessions. Early users report that long‑running tasks stay coherent through the end, with fewer repeated steps and less contradictory code. The to‑do list also helps you, the developer, understand how the agent is thinking—no more wondering why it’s editing a file or revisiting a function.

We’ve already seen the power of structured planning in other AI tools, such as Cursor’s task lists and Cline's Focus mode. In future releases we’ll extend the to-do list with deeper integration into our planning and execution stages, allowing you to tweak the plan before the agent starts coding and to branch off alternate paths when designs evolve. We expect that this feature will eventually replace the Orchestrator mode.

Larger context windows and more capable models will continue to push the boundaries of AI coding, but they won’t eliminate the need for structure. Coplay's to-do list provides that structure, ensuring that Coplay remains laser‑focused on your goals—so you can stay in flow and bring your game ideas to life.