# You can be ambitious without the huge token bill. Here's how.

## Executive summary

While advanced AI agents increase capability and token consumption, the rising cost is defensible only if the work they perform is genuinely new and valuable. The core strategy for cost control is not simply finding a cheaper model, but fundamentally redesigning the business workflow by eliminating unnecessary 'interoffice envelope' steps. By focusing on the desired business outcome, engineers can minimize handoffs and reserve expensive, frontier models for complex, exceptional cases, while using cheaper models for routine, deterministic tasks.

## Key takeaways

- Redesign the Workflow, Don't Just Buy an Agent: Before selecting a model, determine which work should exist at all. Start by defining the desired business outcome (e.g., an accurate quote) and map the most efficient path to achieve it, rather than simply automating the existing, often redundant, process.
- Separate Value-Add Work from Administrative Overhead: Many existing enterprise steps (like summarizing requests for different departments) exist because previous systems lacked interoperability. Identifying and eliminating this 'admin work' can drastically shorten the workflow and reduce token consumption.
- Match the Model to the Task Complexity: Not all work requires the highest intelligence. Reserve expensive, frontier models for the 1-5% of challenging, exceptional cases. Use cheaper, open-weights models for routine, deterministic tasks (e.g., calling a CRM or applying known pricing rules).
- Implement Evaluation (Evals) for Reliability: To ensure a redesigned process works, implement rigorous evaluation (Evals) to check if the agent's output is not just 'approximately right,' but factually correct and meets business requirements. Evals are a critical human skill for maintaining quality.

## Technical details

- Harness Architecture (Thick vs. Thin): The 'harness' is the entire system surrounding the model (instructions, tools, information, checks). A less capable model often requires a 'thick harness' (more structure and explicit steps) to perform predictably, while a challenging, investigative problem requires a 'thin harness' to give the frontier model freedom.
- Model Routing and Classifiers: Use classifiers to determine the type of incoming request. Route standard, predictable requests to cheaper, open-weights models, and reserve the high-cost, frontier models only for unusual or complex edge cases.
- Token Efficiency: The consumption of tokens increases exponentially because agents now perform multi-step investigations (reading history, checking product info, etc.), leading to significantly more tokens per request than simple Q&A.

## Practical implications

- Challenge existing workflows: For every process, ask what the absolute minimum steps are to achieve the business result, ignoring historical departmental handoffs.
- Prioritize business ownership: Business leaders must own the definition of the 'desired outcome' to guide the AI design, preventing engineers from simply automating inefficient processes.
- Implement structured checks: Use classifiers and evals to enforce quality and cost control, ensuring the agent's output is verifiable against known business rules (e.g., pricing systems).
- Focus on the 'Why': Shift the conversation from 'Which model is best?' to 'Which model can do this specific job well?'

## Topics

Generative AI, Agent Design, Workflow Optimization, LLM Cost Management, Enterprise AI Strategy, Token Saver Skill Guide, Substack Newsletter

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