From 15 to 150,000 agents: IBM's blueprint for governing AI at scale
Key points
Key takeaways from Source interview with IBM Group Vice President Bruno Aziza (October 2026):
Agent count explosion. Aziza expects the average Fortune 500 company to go from roughly 15 AI agents today to around 150,000 within three years, making continuous governance essential.
The governance gap. Around 80% of governance today is periodic and manual, and it can take six to nine months to detect a model-drift issue or a breach — too slow when agents act around the clock.
Orchestration, not just governance. Aziza prefers the word orchestration over governance, arguing the goal is a continuous system that lets employees innovate while staying in control.
Identity in the agent era. Roughly 70% of breaches are identity breaches rather than password breaches, and with a projected 10:1 machine-to-human ratio, identity and access systems designed for humans alone will not hold.
Sovereignty is multi-dimensional. More than 80% of organisations rank sovereignty among their top three priorities; it spans data, technology, operations and regulation, and IBM's Sovereign Core maps 200-plus compliance frameworks, as seen with Bank of India's half-billion customers across 36 countries.
The AI operating model. IBM frames its AI operating model around four pillars: right-time data (batch and real time), agentic workloads, hybrid cloud, and automation and infrastructure management.
Enterprise-ready, not just AI-ready. Aziza reframes the industry slogan: rather than making data AI-ready, the task today is making AI enterprise-ready, building from AI up with expectations such as real-time data.
Culture as practice. At IBM's Tech Exchange in Atlanta on 26 October, customer stories from American Express, Citi and CVS will anchor a practitioners-led agenda — culture is what you do, not what you put on walls.
Read more: theCUBE
