# 📅 ThursdAI - LIVE from AI Engineer Worlds Fair - OpenAI, DeepMind, EXO, Sakana & more friends

## Executive summary

This live panel discussion from the AI Engineer World's Fair focuses on the critical shift toward local and open-source AI models. Speakers debated the current state of frontier models (like OpenAI's GPT-5.6) versus decentralized, sovereign AI solutions running on consumer hardware. Key technical topics included model routing (Fugu), agentic workflows using tools like Weights & Biases' Coreweave Ara, and the necessity of local inference to ensure data sovereignty and prevent vendor lock-in.

## Key takeaways

- The resurgence of Fable: Fable is back, marking a significant moment for open models. The discussion highlighted that this trend emphasizes the need for decentralized AI solutions over reliance on single cloud providers.
- Local AI and Sovereignty: Running large language models (LLMs) locally is presented as crucial for guaranteeing data sovereignty, preventing vendor lock-in, and ensuring continuous operation regardless of cloud provider restrictions.
- Model Routing and Orchestration: The concept of model routers (like Fugu) was presented as a superior method for achieving high performance, allowing users to dynamically select the best model for specific tasks rather than relying on a single monolithic LLM.
- The Agentic Era and Tooling: Tools like Weights & Biases' Coreweave Ara are emerging to automate the entire AI research loop (auto-research), moving beyond simple chatbots into full agentic co-pilots for ML engineers.

## Technical details

- Model Performance Benchmarking: A comparison was shown between Sonnet 5 and Opus 4.6, noting that while the token consumption of Sonnet 5 might not be significantly higher than 4.6, its cost per task can sometimes be lower with Opus.
- Local Inference and Hardware: The ability to run large models locally (e.g., on Mac Minis or consumer hardware) was emphasized as a countermeasure to geopolitical risks and cloud dependency, promoting 'true sovereignty' in AI.
- Model Pruning and Optimization: The process of pruning models (e.g., GLM 5.2) was discussed, noting that while basic pruning can take days, achieving optimal performance requires dedicated effort over a week or more.
- Agentic Code Review: The Codex agent is evolving into a daily co-worker for non-developers. It will eventually handle code review, with the system designed to flag PRs even if human reviews are bypassed.

## Practical implications

- Build engineers must incorporate local inference capabilities into product roadmaps to mitigate risks associated with centralized cloud APIs and ensure continuous operation.
- The shift toward agentic workflows means that development focus is moving from writing raw code to designing complex, multi-step task chains (agent loops) that utilize multiple specialized models.
- Understanding model routing architectures (like Fugu) is critical for optimizing performance and cost by dynamically selecting the best available LLM for specific tasks.

## Topics

Local AI, Open Source Models, Agentic Workflows, LLM Benchmarking, Model Routing (MoE), Data Sovereignty, wolfbench.ai, Weights & Biases (W&B), Coreweave Ara

Source: https://www.youtube.com/watch?v=eLSf0Zq2kR4