# Intelligence EXPLOSION: Harness Engineering with Pi Agent, Deepseek, and Gemini

## Executive summary

The video addresses the current 'Intelligence Explosion' in Large Language Models (LLMs), noting rapid model releases and aggressive pricing wars. The core thesis is that modern agentic engineering requires moving beyond selecting a single best-in-class model; instead, engineers must build flexible systems to combine compute from multiple models (e.g., Gemini 3.7 Flash, Deepseek V4 Pro, Claude Fable 5) to achieve superior performance, speed, and cost efficiency. The speaker demonstrates three advanced multi-agent orchestration patterns—Opinion, Debate, and Collaborate—using a custom Fusion Harness V2.

## Key takeaways

- The Principle of Flexible Systems: In an era of rapid technological change, the most flexible system wins. Engineers should focus on combining compute rather than selecting a single model (Combine compute, don't select compute).
- Multi-Agent Orchestration Patterns: Three advanced patterns are demonstrated for leveraging multiple LLMs: **Opinion** (fire one prompt to N models for diverse perspectives); **Debate** (agents argue a claim across multiple rounds, sharing refutations and positions); and **Collaborate** (models propose plans which an 'architect agent' merges into a structured task list with dependencies).
- Cost-Performance Analysis: While state-of-the-art models like Claude Fable 5 are powerful, they can be vastly more expensive than 'A tier' workhorse compute (e.g., Gemini 3.7 Flash or Deepseek V4 Pro), offering better effective agent hour per token cost.

## Technical details

- Fusion Harness V2: The speaker showcases the V2 of his custom Pi coding agent, designed to facilitate multi-agent orchestration. This tool allows engineers to run multiple models side-by-side on a single task (e.g., analyzing DuckDB v2 preview) to compare performance, speed, and cost.
- Multi-Agent Orchestration: Opinion: This pattern gathers N unique perspectives on a single issue (e.g., 'What is the most important feature to focus on from DuckDB v2?'). The output provides multiple, distinct recommendations and accompanying quick experiments.
- Multi-Agent Orchestration: Debate: The system facilitates a structured debate where agents argue a claim over several rounds. This process is valuable for strategic decision-making, allowing models to compare, contrast, and reinforce consensus (e.g., debating if DuckDB should be treated as an embedded analytical engine or a general-purpose server).
- Multi-Agent Orchestration: Collaborate: This advanced workflow has agents propose individual plans, which are then synthesized by a dedicated 'architect agent' into a single, structured task list. This plan includes dependencies, owners, and risk analysis, simulating a full software development lifecycle.

## Practical implications

- Architect your agentic workflows to be model-agnostic and highly flexible (i.e., own the harness) rather than relying on closed or single-model products.
- Utilize multi-agent patterns (Debate, Collaborate) for high-leverage tasks like strategic decision-making and complex planning, as these provide diverse viewpoints that outperform single-model outputs.
- When budgeting compute, prioritize 'A tier' workhorse models (like Gemini 3.7 Flash or Deepseek V4 Pro) over the most expensive state-of-the-art models for optimal cost/performance ratio.

## Topics

Agentic Workflow, LLM Architecture, Build Engineering, Multi-Model Orchestration, AI Pricing Models, Fusion Harness, DuckDB v2.0 Preview, Pi Coding Agent (Fusion Harness)

Source: https://www.youtube.com/watch?v=rqZHR-hRllI