What OpenAI's GPT-5.6 and Fidji Simo's Exit Tell Us About the Next Phase of AI

The Day's Thesis

▶

Signal of the Day: OpenAI positioned GPT-5.6 as the preferred model for Microsoft Copilot 365 on the same day its No. 2 executive, Fidji Simo, stepped down permanently due to illness — creating simultaneous product momentum and leadership uncertainty ahead of a potential IPO.

The 30-Second Read:

• GPT-5.6 named preferred model for Microsoft Copilot 365, deepening the OpenAI-Microsoft dependency even as IPO speculation intensifies
• Fidji Simo departs OpenAI's AGI chief role permanently, the third C-suite exit in three months alongside COO Brad Lightcap and CMO Kate Rouch
• AWS GraphRAG deployment cut pharmaceutical R&D cycles by 87%, compressing a six-month data-gathering phase to weeks
• Macquarie warns copper's approach to $14,000/tonne is running ahead of physical fundamentals as stockpiles reach multi-year highs

Today's news turns on a single tension: OpenAI is winning the product war — GPT-5.6 locked into the world's largest enterprise software suite — while quietly hollowing out the executive layer that would manage the transition to public markets. The copper data adds a supply-side caution that matters for every data center construction budget downstream.

---

AI & Research Frontier

AWS GraphRAG cut pharmaceutical R&D cycle time by 87%, the sharpest enterprise AI productivity claim validated by a named deployment in 2026 to date.

GraphRAG — graph-based retrieval-augmented generation, a technique that links structured knowledge databases into a queryable network for AI inference — achieved this by unifying previously siloed proprietary databases, reducing the initial screening phase from more than six months per iteration to weeks.

The historical success rate of that phase was approximately 5%; the deployment does not yet disclose the new hit rate, which is the figure to watch. At that compression ratio, a mid-size pharma running 10 concurrent programs effectively gains the equivalent of four additional R&D years without adding headcount.