Keep pulling the thread on Erhan Giral & Ryan Manning.
A disruption similar to the one in software engineering will also happen in IT operations, as ops problems are verifiable, though on a slower timescale (hours/days vs minutes).
The AI industry has likely exhausted the potential of learning from existing public text data.
Future AI models will increasingly be trained by being exposed directly to real-world environments to gain a "first-person view," rather than learning from human-written text.
BMC is building training environments called "gyms" where AI agents can learn by autonomously breaking and fixing applications in a controlled data center.
The IT service management market is difficult to serve across small, medium, and large enterprises simultaneously due to both product and go-to-market challenges.
In the Fortune 2000 enterprise market for IT service management, the primary competitors are ServiceNow and BMC Helix.
Around 2010, BMC was slow to adopt cloud technology, which led to the company being taken private by Bain.
KKR later invested in BMC, bringing in leadership from CA Technologies to rebuild the company with an AI-first approach.
BMC Helix uses a Mixture of Experts (MoE) architecture, a pattern where Meta has done leading work, to fine-tune pre-trained models for specific tasks.
In the BMC Helix system, actions that are purely diagnostic are automated, while remediation actions that modify a system still require human-in-the-loop approval.
Recurring problems in IT operations tend to constitute 80-85% of the total problem space.
BMC Helix has a "Deep Root Cause Analysis" agent that uses a hierarchy of sub-agents to investigate major IT incidents in a post-mortem analysis.