REPORTED DEVELOPMENT

Redwood's Enterprise Benchmarks in AI and Agentic Orchestration 2026 draws on Leger Opinion's survey of 300 technology, AI and risk leaders in the US, Canada and UK. Seventy-seven per cent report negative consequences from ungoverned agents; 54% call their organisations fully prepared, 42% have central agent registries and 51% real-time monitoring. Unexpected cloud costs are among frequent consequences.

The supplier-commissioned, self-reported study does not independently establish economic losses or causal relationships between controls and outcomes.

WHY IT MATTERS TO THE MARKET

Agent inventories, responsibilities, permissions, actual activity, observed costs and demonstrable ROI must remain distinct.

MARKET EVIDENCE

SOURCE-LINKED INDICATORS

The study distinguishes a survey population, perceived consequences and reported control readiness.

300Senior technology, AI and risk leaders surveyedSPONSORED SURVEY SAMPLE2026 / US, Canada and UK / Leger Opinion sampleLeger-panel study of 300 senior leaders in the US, Canada and UK; predominantly large organisations. Self-reported controls and consequences, not independently audited incidence.REDWOOD / LEGER / VIEW SOURCE
77%Respondents reporting adverse consequences of ungoverned agentsSELF-REPORTED SURVEY2026 / Surveyed leaders / perception, not audited incident recordsLeger-panel study of 300 senior leaders in the US, Canada and UK; predominantly large organisations. Self-reported controls and consequences, not independently audited incidence.REDWOOD / VIEW SOURCE
42%Respondents reporting a centralised agent registrySELF-REPORTED SURVEY2026 / Surveyed organisations, according to leadersLeger-panel study of 300 senior leaders in the US, Canada and UK; predominantly large organisations. Self-reported controls and consequences, not independently audited incidence.REDWOOD / VIEW SOURCE
51%Respondents reporting real-time agent monitoringSELF-REPORTED SURVEY2026 / Surveyed organisations / control descriptions unverified77%, 42% and 51% are respondent shares, with different questions. Adverse consequences, central registry and real-time monitoring are distinct measures.REDWOOD / VIEW SOURCE
WHAT THE DATA SUGGESTS

The relevant operational benchmark is whether an organisation can inventory deployed agents, revoke their rights, observe their actions and reconcile reported costs against incident evidence.

INTERPRETATION BOUNDARY

Supplier-commissioned, self-reported answers do not prove actual economic loss, causal relationships or the effectiveness of monitoring and registration tools.

ORIGINAL REFERENCES