Core Values

0. Users Care Only About Their Task

Users want only one thing: to get their task fulfilled. Donew, on the user plane, must only focus on this concern. Implementation details should not be a user concern unless explicitly requested.

1. Code-First Expression

If it can’t be expressed in code, it should not exist. Donew inherently generates code to interact with its realm. A realm is a set of stateful capabilities you define, and it can even include another Donew instance. It must not rely on model capabilities such as tool calling. Each Donew task, under the hood, resembles a futuristic code that drives the task.

2. Composability

Users must not take extra steps to configure how agents interact with each other. Passing a Donew instance to another must work by design.

3. Predictability

Regardless of task complexity and setup, when users define their expected input/output:

• Donew must work with that input and return the outcome
• Must try its best to either self-correct or fail gracefully
• Must leverage its code and In-Context Learning architecture
• Half-baked or unreliable results are not acceptable

4. In-Context Learning

Donew learns from every interaction:

• Each task builds context and captures key insights
• Uses a dynamic timeline instead of static databases
• Adapts to your domain language and business context
• Learns continuously from real-world usage

5. Universal Runtime with Isolation

Each Donew instance must:

• Work independently
• Be packaged as a portable unit with all necessary components
• Be deployable in any compliant runtime environment

6. Continuous Optimization

Donew must:

• Track its success vigilantly
• Alert users when task optimization is possible
• Eliminate unnecessary AI-generated components where possible
• Optimize initial prompts
• Treat self-improvement as a core assignment

7. Bring Your Own X

Donew must come with batteries-included and sensible defaults included. But also, users can bring their own stuff if they follow rules from above.