Google’s Delayed Gemini Pro: Coding Challenges and Internal Friction
Google is months behind schedule on delivering the next version of its flagship AI model, Gemini Pro, because the technology has fallen short of internal coding goals. (Bloomberg)
Background
Ten current and former Google employees share their insights on the mounting frustration within the company. The upgrade was expected at the May developer conference but hasn’t materialized due to:
- Competition: Internal factions compete, with Google Cloud, DeepMind, and Android teams all developing AI coding tools.
- Performance Gap: Outperformers like Anthropic and OpenAI have released models surpassing Google’s current offerings in code writing.
- Departures: A wave of senior researchers is leaving for Anthropic and other labs due to frustration with Google’s competitive position.
Challenges and Efforts
- Structural Issues: Structural problems within the company include fragmented efforts and a push from co-founder Sergey Brin for faster AI coding.
- Consolidation: Koray Kavukcuoglu leads efforts to unify internal AI coding tools, while Sebastian Borgeaud tackles the problem specifically at DeepMind.
- AI Integration: Google claims 75% of code is now AI-generated and has consolidated tooling under Antigravity.
Impact
- Capacity Constraints: Engineers often face capacity constraints due to internal competition for computing power, impacting both internal and external customers.