The Promise Meets the Paper Trail
When organisations first deployed AI in their customer operations, the pitch was simple: reduce cost, accelerate resolution, free your people for higher-value work. The reality, according to new Freshworks research highlighted by CX Today, is proving considerably messier. AI complexity is quietly becoming a CX problem in its own right — one that is inflating IT workloads, consuming governance budgets, and pulling strategic attention away from the customer outcomes the technology was supposed to deliver.
The research surfaces a pattern that will feel familiar to anyone who has lived through a rushed AI rollout. Multiple point solutions accumulate. Each one requires monitoring, maintenance, and fine-tuning. Governance gaps emerge between what the models were trained to do and what customers are actually asking them to do. And then comes what 2026 research is increasingly calling "botsitting" — the very human, very expensive practice of stationing staff to watch AI systems in case something goes wrong. The automation dream, in other words, has generated a new category of manual labour.
What Botsitting Actually Costs You
Botsitting is rarely a line item in an AI business case. It shows up instead as creeping headcount in IT, unexplained QA overhead, or customer satisfaction scores that plateau despite fresh technology investment. When an AI agent misroutes a complaint, misreads sentiment, or simply loops without resolution, someone has to intervene. If that someone has not been formally designated, the work lands on whoever is closest — often a frontline agent already handling their own queue, or a team lead who should be coaching, not babysitting software.
The governance gap compounds this. Many contact centres deployed conversational AI under pressure to show results quickly. Training data was good enough to pass procurement but not good enough to handle edge cases at scale. Policies around escalation, data handling, and quality assurance were written after go-live rather than before it. The result is a patchwork of informal fixes that looks functional until a high-value customer hits a gap and walks.
Complexity as a Strategic Risk
CX leaders need to reframe AI complexity from an IT inconvenience into a strategic risk. When your AI stack demands constant human supervision just to stay operational, it is not automating your customer service — it is restructuring it around a more fragile set of dependencies. Your best agents are context-switching between their own conversations and remediation tasks. Your IT team is triaging integrations instead of planning the next capability. Your budget is funding maintenance rather than improvement.
The Freshworks findings are consistent with a broader 2026 theme: organisations that moved fast on AI adoption are now discovering that speed without structure generates hidden costs. Those costs do not appear on a model vendor's invoice. They accumulate in people's time, customer patience, and leadership bandwidth.
Why Hybrid Intelligence Is the Operational Answer
The instinct to respond to botsitting by adding more automation is understandable but usually wrong. A second layer of AI monitoring the first layer is simply a more expensive version of the same problem. The smarter operational response is to build the human layer deliberately — not as a fallback, but as a designed component of your service architecture.
In a well-constructed hybrid model, skilled human agents are not there to catch AI failures reactively. They are integrated into the workflow proactively: handling the interaction types where nuance, empathy, or regulatory sensitivity genuinely require a person; reviewing AI outputs on a sampled, structured basis; and feeding those quality signals back into model improvement. This is not botsitting. It is human intelligence doing what it does best — exercising judgment — while AI handles volume, routing, and first-contact resolution at speed.
Critically, this model also produces better governance. When humans are part of the loop by design, escalation paths are clear, accountability is traceable, and quality assurance is continuous rather than crisis-driven. The AI gets better over time because the feedback mechanism actually exists.
The Operational Takeaway
If your AI deployment is generating more oversight work than it is eliminating, the answer is not patience — it is architecture. Audit where human intervention is currently happening informally and make it formal. Define which interaction types belong to AI, which belong to humans, and which require both in sequence. Treat your multilingual, culturally fluent human talent as a first-class operational asset, not a contingency plan. AI complexity becomes a CX problem when humans are an afterthought. Make them the backbone, and complexity becomes manageable.