77 % des entreprises ont payé au prix fort l'absence de gestion des agents d'IA

Redwood Software's survey of 300 technology, risk and AI leaders found that 77% of companies experienced negative consequences from an unmanaged AI agent, most commonly unexpected cloud costs, operational disruption or compliance issues. Agent deployment is advancing faster than integration and governance: 89% increased agent deployment over the past year, but only 42% report strong capability to connect agents to legacy or core systems, and 42% maintain a centralized agent registry. Respondents cited real-time observability (62%) and proven integration with legacy systems (58%) as priorities for scaling agentic AI confidently.

Analysis

The investable signal is a potential shift in enterprise AI budgets from experimentation toward integration, observability, and runtime controls—not evidence yet of a broad increase in software spend. Redwood has a commercial interest in this conclusion, and the survey’s 300 respondents do not establish realized market-wide losses or willingness to pay. Treat the percentages as a demand hypothesis, not a revenue forecast.

Over the next 1–3 months, vendors that can demonstrate secure connections to legacy systems, centralized agent inventories, and real-time monitoring may gain budget priority. That could benefit established workflow and IT-operations platforms such as ServiceNow and IBM, alongside observability providers such as Datadog; it may also support specialists in automation and governance. Conversely, cloud providers could face pressure to justify unexpected consumption costs, though agent growth could still lift usage overall. The second-order risk is that stronger controls slow agent rollouts, delaying software vendors’ AI monetization rather than accelerating it.

Over 6–18 months, the differentiator is likely measurable reliability and integration into production workflows, not agent demos. Redwood’s customer comparisons are not causal evidence of product advantage. The contrarian view: this may be ordinary implementation friction amplified by a vendor-funded survey, while cloud-cost overruns could reflect immature usage rather than a durable governance market. Reassess if major software buyers disclose specific control-related budget increases, or if agent deployments stall despite those investments. Falsifiers include enterprise guidance showing continued AI budget concentration in model access without orchestration spend, and evidence that agent incidents or cloud-cost overruns are not recurring.

Key Decisions for Investors

• No immediate trade on this release alone: low-impact, vendor-promotional survey evidence is insufficient to underwrite earnings revisions.
• Put ServiceNow, IBM, and Datadog on a 1–3 month watchlist for evidence that orchestration, integration, and observability demand is converting into bookings or guidance—not just product announcements.
• Monitor cloud consumption commentary and FinOps demand. A durable rise in cost-control spending alongside agent adoption would support the orchestration thesis; slowing consumption without corresponding control budgets would weaken it.
• Before taking a relative-value position between workflow platforms and AI-agent software vendors, verify customer renewal and expansion data, production deployments, and whether governance features are separately monetized.
• Revisit the thesis over 6–18 months if enterprise disclosures show production agent deployments accompanied by measurable spend on monitoring and legacy-system integration; stand down if rollouts remain pilots or buyers defer those projects.