# Build Hour: Valuemaxxing with GPT-5.6

## Executive summary

The session provides a deep dive into 'Value Maxxing' with GPT-5.6, shifting focus from merely maximizing token usage to optimizing AI agents for concrete business outcomes and cost efficiency. Key strategies include selecting the optimal model (Sol, Terra, Luna) based on workload needs, implementing advanced API features like programmatic tool calling and prompt caching, and structuring agent workflows to minimize redundant context processing.

## Key takeaways

- Shift from Token Maxxing to Value Maxxing: Progress should be measured by the value generated (e.g., time saved, quality improved), not just the number of tokens consumed. This requires defining clear outcomes and measurable 'good' for AI agents (evals).
- Model Selection Strategy: The GPT-3.5 family includes Sol (flagship/complex tasks), Terra (balanced intelligence/cost/latency), and Luna (high-volume, cost/latency sensitive) to optimize for specific workloads.
- Optimizing Agent Workflows with Codex: For day-to-day coding, starting with `GPT-5.6 Soul` on medium reasoning is often sufficient. Developers can also trade tokens for time using Fast mode or utilize Chronicle to build task memory.
- Advanced API Techniques (Harness Optimization): Implement programmatic tool calling (using a JavaScript sandbox) and prompt caching to significantly reduce input token costs and processing time. Persistent reasoning and compaction also boost performance and cache efficiency.

## Technical details

- GPT-5.6 Model Reasoning Levels: The model offers new reasoning modes: Max (fully takes the leash off) and Ultra (a mode on top of Max, designed to natively use subagents). Using these modes is highly token intensive and recommended only for complex, heavy-hitter use cases.
- Programmatic Tool Calling: This feature allows models to write code into a JavaScript sandbox rather than reasoning through logic. This saves tokens by moving computation out of the model's chain of thought and can save entire model turns.
- Prompt Caching (KV Cache): Caching reusable parts of prompts (like system instructions or date information) drastically reduces input costs. The cache is most effective when context remains append-only; modifying tools or system prompts breaks the cache.
- Subagent Orchestration: When using subagents, they are given only the specific context needed for their task by the main agent, which is more efficient than providing the full chat context. The effectiveness of multi-layer delegation (e.g., Soul -> Terra -> Luna) can be lossy.

## Practical implications

- Audit your agent's traces to identify redundant outputs or unnecessary steps that can be automated with code.
- When building production agents, prioritize caching system prompts and tool definitions by setting breakpoints on stable context elements.
- For complex tasks, consider using programmatic tool calling to offload computation from the LLM's reasoning process, saving both tokens and time.
- Benchmark cost-performance not just on token count, but on 'cost per overall task to completion.'

## Topics

AI Agents, LLM Optimization, API Development, Cost Management, GPT-5.6, GPT-5.6 blog, Programmatic tool-calling guide, Compare models, Migrating a production AI agent to GPT-5.6 (Ploy), Build Hour repo

Source: https://www.youtube.com/watch?v=jyuyY86GJnA