AI Agents for Network and Security: Expectations vs Reality

[ad_1] On this article, we examined engineers’ expectations relating to AI brokers for networking and safety, in addition to the provision of business and open-source AI brokers or options.

What Is an AI Agent? An AI agent is an autonomous software program system that independently perceives, causes, and acts to realize particular targets. In contrast to conventional software program that follows fastened directions, brokers repeatedly monitor their setting, make dynamic choices based mostly on accessible info, and adapt their methods by way of studying and expertise. Trendy AI brokers make the most of instruments corresponding to net search, APIs, code execution, and file manipulation to work together with exterior methods, enabling them to carry out complicated duties—from community monitoring and troubleshooting to safety evaluation and menace response.

Design and Structure On the design and structure stage, AI agent creators ought to resolve the next:

Select the AI framework. Select associated AI mannequin(s). Design AI agent instruments. Collect and check details about databases, APIs, and MCP servers which are deliberate for use.

Fashions for AI Brokers The LLM is a central controller, or “mind,” for brokers. It’s a vital ingredient for the brokers. So, which mannequin must you select? One of many methods to tell apart between the fashions is by utilizing benchmarks. Benchmarks in AI are standardized assessments designed to measure and evaluate the skills of various fashions on particular duties or a set of duties.

Most of the hottest benchmarks usually chosen for mannequin comparability don’t embody networking- or security-related questions.

One of many closest networking benchmarks that we’ve discovered is Community Operational Data (NOK) benchmarking for LLMs (Dec 15, 2023). For safety, there may be the CTIBench benchmark. You may also discover Basis-Sec-8B, a mannequin developed by Basis AI at Cisco. Basis-Sec-8B is an open-weight language mannequin specialised for cybersecurity functions.

There are pondering fashions that ideally match the necessities for brokers.

AI Reasoning and Pondering Fashions

Gemini 2.5 Professional (Deep Suppose). GPT-5 Pondering. Claude Sonnet 4.0 Pondering. Claude Opus 4.1 Pondering. Qwen3-235B-A22B-Pondering.

AI Brokers: Expectations Listed below are the outcomes of surveys relating to “Which AI brokers would you most need to discover on Cisco DevNet Code Alternate?”:

AI Brokers on Cisco DevNet Code Alternate: Survey Outcomes (September 2025).

Mixed survey outcomes (100 complete votes):

Configuration automation: 37% — The best choice among the many DevNet group, indicating robust demand for AI brokers that may automate community configuration duties and infrastructure administration. Community monitoring brokers: 32% — Shut second precedence, displaying engineers need AI-powered instruments for community visibility, efficiency monitoring, and operational insights. Risk and vulnerability brokers: 22% — Vital curiosity in AI brokers targeted on safety monitoring, menace detection, and vulnerability evaluation capabilities. Code technology brokers: 9% — Lowest precedence, suggesting engineers are much less all in favour of AI instruments for writing code in comparison with operational automation.

The mixed outcomes from Cisco DevNet, LinkedIn, and X/Twitter surveys display a transparent desire for operational AI brokers over development-focused instruments. With practically 70% of votes going to configuration automation and community monitoring, the Cisco technical group prioritizes AI brokers that remedy real-world infrastructure challenges fairly than code technology utilities. Twenty-two % of votes went to AI brokers targeted on safety.

Listed below are duties that Community and Infrastructure engineers, SREs, and DevOps groups need to make the most of an AI agent for inside their workflows:

Common outcomes of the offline surveys at Cisco DevNet Share Your Expertise Zone (2024–2025).

AI Brokers for Networking and Safety: Open-Supply and Industrial There are open-source AI brokers for networking that allow you to handle and troubleshoot community gadgets through pure language, robotically changing requests into protected CLI actions or gRPC/REST API calls. There are additionally brokers that automate complicated diagnostics and troubleshooting finish to finish. These brokers validate intent, plan per-device investigations, execute instructions throughout a number of gadgets, assess findings, and generate clear, memory-augmented studies.

We in Cisco DevNet collect open-source networking and safety AI brokers right here: https://developer.cisco.com/codeexchange/search/?q=ai+agent. Quickly, it is possible for you to to discover a separate part for MCP servers at DevNet Code Alternate.

And what can industrial firms supply?

Cisco has a Cisco AI Assistant that may work with varied merchandise and assist analyze insurance policies, robotically generate studies, and ship notifications, amongst different duties. Cisco AI Assistant is conscious of the newest documentation and guides associated to the merchandise. Cisco can also be creating AI Canvas, the primary generative UI that unifies real-time telemetry, AI insights, and group collaboration throughout all IT domains in a single clever workspace.

Here's a listing of business firms that develop AI brokers as merchandise: AI Community Engineers from Nanites AI, DevAI, Copilot for Community Automation by Selector AI, and Aviz AI Brokers.

Many industrial firms declare to supply multivendor help. The AI brokers developed by industrial firms can carry out on a regular basis duties, corresponding to robotically verifying end-to-end pings, checking community compliance, producing audit studies on demand, validating firewall guidelines, and offering Degree 1 help automation. Additionally they supply superior capabilities, together with stock insights that immediately entry full community knowledge (gadgets, OS variations, {hardware} SKUs, ASICs, MAC addresses, and transceivers), in addition to correlation and prediction for capability and efficiency planning.

AI Brokers: Safety Issues As fashions evolve, considerations about accuracy and reliability lower over time; in the meantime, safety and compliance considerations improve. That is related to widespread media protection relating to mannequin jailbreaking, immediate injection, and mannequin poisoning.

AI brokers may cause points, present incorrect directions, or apply non-optimal configurations.

Doable AI agent errors might be related with the next:

Incorrect agent ideas based mostly on info/errors obtained from providers by way of REST APIs/WebSockets/CLI. Hallucination. Selecting the fallacious device.

Using associated AI safety instruments, guardrails, and deciding on the suitable fashions can assist mitigate dangers.

On the similar time, AI brokers can present clever automated protection. Typically safety specialists want an assistant to assist react to incidents or collect and apply associated insurance policies and updates. Safety engineers usually lack ample time to assessment alerts, acquire knowledge from varied sources, think about particular and historic context, and take acceptable motion.