JPMorgan Chase’s $20 billion technology budget is a statement of architectural intent. The bank is restructuring entire business units, like its commercial and investment bank, to accelerate AI integration. Its approach is systematic: an internal generative AI platform serves over 200,000 employees, while engineering teams are measured by their adoption of tools like GitHub Copilot. The strategy is to build and embed, aiming to match every dollar spent with operational savings.

Goldman Sachs maintains a disciplined $6 billion tech spend, constrained by return requirements. Its OneGS platform, now in its third iteration, focuses AI on specific workflows like trade accounting, avoiding a broad frenzy. Citi’s model is bottom-up, with tools reaching 70% adoption across 84 countries. Its generative AI systems reportedly save 100,000 developer hours weekly on code reviews, a tangible engineering win.

System design choices vary. Wells Fargo employs a ‘hub and spoke’ model for its AI agents, targeting high-impact cases. Bank of America has embedded AI from consumer banking to institutional divisions, claiming over 90% employee use of its long-running Erica assistant.

The engineering trade-offs are becoming clear. Productivity metrics, like Wells Fargo’s cited 35% engineer boost, are promising. However, these systems introduce new attack surfaces. Regulators are already questioning banks about AI models like Anthropic’s Mythos, which can autonomously generate software exploits, prompting defensive hardening. Justifying billions in spend requires proving these gains outweigh the escalating security complexity and the workforce reductions already in motion.

Source: Webpronews