Agents are becoming table stakes in IT and network operations. What separates the leaders now is trust, and trust has three parts.
Every major operations platform now ships agents, and the analysts expect most enterprises to run them in production within three years. The question operators ask has changed. It is no longer “can it act?” It is “what happens on the day it’s wrong?“
We call the distance between what agents can do and what operators will let them do the trust gap. Three things close it: context, an understanding of the environment the agent is acting on; guardrails, so every action can be bounded, tested and explained; and proof, results from production rather than a demo. Six signals from the research firms show why the trust gap now matters more than the agents themselves.
Six signals shaping agentic AI in operations
1. Agents go mainstream. Gartner expects 60% of enterprises to deploy agentic AI in IT infrastructure operations by 2029, up from under 10%. When every platform has agents, agents stop being the differentiator.
2. Human approval collapses. Gartner expects human sign-off to cover 20% of AI-suggested actions by 2029, down from 80% in 2025, and credits deterministic guardrails. Autonomy is a climb, not a switch.
3. Risk rises with autonomy. Gartner warns that 40% of infrastructure and operations organizations running agentic AI at scale will face a business-critical disruption by 2028, and has recommended digital twins, real-time rollback and secure human override.
4. Hype meets proof. Gartner predicts over 40% of agentic AI projects will be canceled by the end of 2027, and estimates only about 130 vendors claiming agentic AI are the real thing. Buyers will ask for numbers, not adjectives.
5. Context beats more data. Appledore Research observes that agentic AI succeeds today mostly in narrow, well-bounded domains, and that cross-domain problems need deeper understanding, not more data.
6. Products beat custom builds. Gartner expects up to 70% of enterprises to abandon agentic AI built through vendor forward-deployed engineering by 2028. Packaged platforms that integrate with existing tools will outlast one-off builds.
Closing the trust gap: what it takes
Context. Data tells you what exists; knowledge tells you what depends on what. A container running on a host depends on it completely; two routers in a mesh depend on each other only partly. A topology graph draws both as the same line. A knowledge model that knows the difference finds the real root cause and the real blast radius, across domains.
Guardrails. Before any action is handed over, write down the boundary of what the system may do alone, and keep an audit trail that survives a bad day. Test before acting and verify after. And make every action carry a plain-language account of what it did, why, and what it checked first, so the person asked to approve it can understand it.
Proof. Report outcomes from production the same way every time: time to detect, noise removed, incidents resolved without a human touch, and service protected.
Where VIA AIOps fits
VIA AIOps is built around a knowledge layer for exactly this reason. Its Self-Evolving Knowledge Plane models how services, networks and infrastructure depend on one another, and keeps learning from telemetry, troubleshooting guides and closed tickets as the environment changes. Teams can start with a minimum viable knowledge model and grow it in 90- to 100-day steps, each with a measurable return.
In practice, a fiber cut at a North American operator that produced 28 automated tickets across eight departments became one ticket, with the root cause found and resolved in under ten minutes, as documented in Appledore Research’s recent paper. Vitria customers have reported catching and triaging problems before customer impact 95% of the time, and NOC productivity gains of 20 to 30% a year.
What to ask in your evaluation
- Which use cases run with approval today, which run alone, and what moved them?
- What may the system do without a human, and can every decision be reconstructed and explained in plain language?
- How does the platform learn your environment, and how does it keep that knowledge current?
- How does it reason across domains, not just within one?
- Which outcomes can it show from production, measured the same way every quarter?
Where to go next
- The self-healing journey: from event intelligence to autonomy
- Demo: Self-Evolving Knowledge Plane, bridging the trust gap
- Appledore Research: Agentic AI in Mission-Critical Operations
- Appledore podcast: Transformation with Agentic AI
- Agentic AI Scorecard: six questions to assess your system
- VIA Results: proven event intelligence outcomes
- From Event Intelligence to Self-Healing: Where VIA AIOps Fits
FAQ
What is the trust gap in agentic AI?
The distance between what AI agents can do and what operators will let them do. Context, guardrails and proof close it.
Why does knowledge matter more than data in AIOps?
Data tells you what exists; knowledge tells you what depends on what. That is what lets a platform find the real root cause and blast radius across domains.
What does Gartner expect for agentic AI in IT operations?
Gartner expects 60% of enterprises to deploy agentic AI in IT infrastructure operations by 2029, and human sign-off to fall from 80% of AI-suggested actions in 2025 to 20% by 2029.
How should an operations team start?
With a minimum viable knowledge model on a well-understood use case with a small blast radius, grown in 90- to 100-day steps with a measurable return at each step.
Sources: Gartner, Hype Cycle for AI in IT Operations, 2026 (July 2026, as reported by The Register); Gartner press releases of June 25, 2025, February 12, 2026 and September 29, 2026; Appledore Research, Talking Transformation, Episode 61, ‘Agentic AI and the Semantic Knowledge Plane’; Appledore Research, Agentic AI in Mission-Critical Operations: Setting a New Standard (2026).
