Indian companies are putting AI into production faster than they can see what it is doing. That is the thread running through Dynatrace’s State of Observability 2026 report, which finds that 53% of Indian organisations name the spread of AI, large language model (LLM) and agentic workloads as a leading source of technology complexity, up nine percentage points from last year and ten points above the global figure of 43%.

The report is based on a global survey of 1,202 technology leaders. Dynatrace, the observability vendor, reads the results as a widening gap between AI ambition and “operational proof”: the ability to understand, validate and demonstrate what AI systems are doing, whether they behave as intended and what value they deliver.

Unpredictable AI behaviour worries half of Indian organisations (50%), and 51% point to the complexity of their technology environment more broadly. The data these systems throw off is a growing burden too. Some 45% of Indian organisations cite telemetry volume as a source of complexity, up ten points, and the pressure shows up in how they collect it: 53% call instrumentation complexity a key data-ingestion challenge, a jump of 21 points, while 48% cite integration problems, up 13 points, and 44% name cost.

The global numbers show why visibility is the sticking point. While 43% of organisations say understanding how AI agents reach decisions is important, only 27% currently address it. Agents are already widely used: 87% of organisations use them moderately or extensively in observability and IT operations. But only 12% run fully autonomous operations today, though 44% are aiming for that level within five years, and humans still check an average of 66% of AI-powered decisions. Nearly half of organisations (49%) now run between six and 25 AI projects, up from 42% a year ago.

AI is also raising the bill. Globally, 81% report higher compute and infrastructure spending because of AI workloads, 78% report higher energy costs, and 77% say AI has increased their need for visibility into cost allocation.

The Indian respondents report real gains from observability. Reliability and uptime are cited as an important outcome by 68%, up 16 points, and better-informed decision-making by 57%, up 11 points. Capacity planning, named by 41%, is up 12 points. AI’s value in wider operations is less established: only 28% see security remediation as a key area and 25% cite reduction of repetitive “toil”.

One finding on security stands out. While 29% of Indian organisations say they fix vulnerabilities in under 24 hours, only 21% have a target service-level agreement of under 24 hours, which means actual performance is running ahead of formal goals at some firms. Dynatrace argues that remediation processes will need to keep pace as autonomous systems become more embedded.