As AI moves from pilot projects into daily production, IT teams are discovering that deployment is only the start. Managing live AI systems now means continuous monitoring, cost control, governance, and reliability work - much of it falling on teams that were never staffed for it.
Grant Thornton's 2026 survey of 950 U.S. senior executives found that 78% lack confidence they could pass an independent AI governance audit within 90 days. Only 14% said AI was fully integrated into operations. KPMG's Q2 2026 U.S. AI Pulse Survey found that just 26% of organizations have real-time visibility into their AI costs.
The pattern is clear: companies are building the machinery to manage AI, but most are still struggling to make the pieces work together.
The unplanned workload lands on IT
Once an AI system is live, teams must monitor model performance and drift, track token and infrastructure costs, enforce usage policies, answer security and compliance questions, and decide when models need updates. These tasks are becoming part of IT's day-to-day operations.
Bhupendra Chopra, co-founder and CRO at Kanerika in Austin, Texas, says much of the unexpected work is concentrated in cost and governance. "Governance and cost are currently the areas where I see the maximum amount of unplanned effort being absorbed," he said. "Security and monitoring had mature tooling and playbooks to draw from. AI governance and cost management are being invented in real time. The questions of who approved a specific agent action, why it made a particular decision, and what that decision cost are now going to IT desks with no established process to answer them."
Many companies still expect existing IT teams to absorb this work alongside everything else they manage. "It is not sustainable, and the strain is starting to show," Chopra said. The teams doing better, he noted, assigned clear responsibility for AI governance, monitoring, and costs early - even if it was only part of someone's job at first.
Treat AI like any other software system
Kurt Guntheroth, an independent software engineer and author, argues that AI systems aren't a special category of software. An AI system is simply a software system that happens to use AI, he said, and enterprise IT can regain control by applying regular software discipline.
That means sticking to core engineering basics: clear requirements, experienced developers, and proper discipline. "But people aren't very experienced with AI, and often don't think carefully about the requirements of the system in their urge to quickly type up some prompts and get AI to do all the thinking for them," Guntheroth said. "That's not a plan that leads to success."
AI does bring unique operational headaches around model drift, costs, and governance. But the fundamentals of building reliable software still hold. Once live, it's simply another system IT needs to manage.
What good AI operations look like
Organizations handling this well aren't creating large AI ops teams. They're making ownership clear from day one. That means continuously monitoring for performance drift, tracking token and cloud costs to understand where spend originates, and putting strict security and access controls around enterprise data.
AI also affects several teams at once - IT, security, legal, compliance, finance, and business. Grant Thornton's finding that only 11% of corporate boards have completed basic AI oversight actions suggests this coordination is happening informally, meaning IT absorbs the burden by default.
For IT managers looking to build these operational skills, structured training can help close the gap. An AI Learning Path for IT Managers covers the monitoring, governance, and cost management responsibilities now landing on IT teams.
Top-performing businesses treat AI as a continuous operational commitment, not a launch milestone. That starts with answering one question: who is ultimately responsible for AI operations, and what does that responsibility cover? If the answer is unclear, the work gets informally distributed to whoever is closest to the system.
Closing that gap requires realistic resourcing. Continuous monitoring, compliance, and cost tracking cannot be sustained without proper tooling, dedicated headcount, and clear authority.
Why this matters for managers
For most IT teams, managing AI is already part of the daily job. The only question is whether companies staff for it intentionally or leave IT to absorb the extra workload by default. Managers who assign explicit ownership for AI governance, monitoring, and cost - and budget for the tooling and headcount to support it - will have more stable deployments than those who let the work land wherever it falls.
Your membership also unlocks:
