Operations | Monitoring | ITSM | DevOps | Cloud

Enterprise AI agent programs overspend because leaders optimize token costs, which represent just 20-25% of variable run costs. Human oversight accounts for 70-75%. The real fix is reducing exception rates and embedding governance directly into agent execution from the start.

Codex is OpenAI's terminal-based autonomous coding agent, bundled into ChatGPT plans and billed through token-based credits. Claude Code is Anthropic's terminal-based coding agent, sold through Claude subscriptions or the API. Codex is cheaper to enter; Claude Code is stronger on complex work. Both run $100 to $250 per developer monthly at active use.

Tuning guardrails can take time. To accelerate this, we combined the power of Claude Code, Coralogix Telemetry and the CX CLI, to build a loop that iterates on guardrails. To test our loop, we built a fictional chat interface for an iGaming operator, that took bets, gave odds on markets and more, but with a catch: this agent was not permitted to make predictions about where the market was going. With our goal in place, we got to work designing our loop.

Across the global landscape, telecommunications operators have already deployed machine learning for predictive maintenance, customer care chatbots, and anomaly detection. However, as the industry transitions toward Autonomous Networks Level 4 (AN L4), where networks can make intent-driven, predictive decisions and perform closed-loop management with minimal human intervention, a fundamental architectural challenge has emerged.

Canonical is now part of the Open Secure AI Alliance, announced by NVIDIA with partners across cloud computing, cybersecurity, enterprise software, open source foundations, and AI research. The alliance aims to develop and share open technologies, techniques, and tools to safeguard software and agents in the age of AI.

AI is already good at writing SQL, which is why people keep asking whether AI agents will replace database IDEs. But letting an agent work on a production database is still a much bigger leap. In tests reported by BIRD, a text-to-SQL benchmark built from 95 real databases across 37 professional domains, human experts reached 92.96% execution accuracy. The best single-model entry reached 80.04% in June 2026. The model failed roughly one task in five; the humans, about one in fourteen.