On this article, we examined engineers’ expectations concerning AI brokers for networking and safety, in addition to the supply of economic 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 attain particular targets. In contrast to conventional software program that follows mounted directions, brokers constantly monitor their surroundings, make dynamic choices primarily based on out there info, and adapt their methods by studying and expertise. Trendy AI brokers make the most of instruments resembling 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 risk 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 take a look at details about databases, APIs, and MCP servers which might be deliberate for use.

Fashions for AI Brokers

The LLM is a central controller, or “mind,” for brokers. It’s an important component for the brokers. So, which mannequin do you have to select? One of many methods to differentiate between the fashions is through the use of benchmarks. Benchmarks in AI are standardized checks designed to measure and examine the talents of various fashions on particular duties or a set of duties.

Lots 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 Information (NOK) benchmarking for LLMs (Dec 15, 2023). For safety, there’s the CTIBench benchmark. You can too 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 Assume).
• 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 concerning “Which AI brokers would you most need to discover on Cisco DevNet Code Change?”:

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

Mixed survey outcomes (100 whole votes):

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

The mixed outcomes from Cisco DevNet, LinkedIn, and X/Twitter surveys reveal a transparent desire for operational AI brokers over development-focused instruments. With almost 70% of votes going to configuration automation and community monitoring, the Cisco technical group prioritizes AI brokers that clear up real-world infrastructure challenges fairly than code era utilities. Twenty-two p.c 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 Business

There are open-source AI brokers for networking that allow you to handle and troubleshoot community units by way of pure language, robotically changing requests into secure 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 units, assess findings, and generate clear, memory-augmented stories.

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 Change.

And what can industrial corporations supply?

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

Here’s a record of economic corporations 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 corporations declare to supply multivendor help. The AI brokers developed by industrial corporations can carry out on a regular basis duties, resembling robotically verifying end-to-end pings, checking community compliance, producing audit stories on demand, validating firewall guidelines, and offering Stage 1 help automation. In addition they supply superior capabilities, together with stock insights that immediately entry full community information (units, OS variations, {hardware} SKUs, ASICs, MAC addresses, and transceivers), in addition to correlation and prediction for capability and efficiency planning.

AI Brokers: Safety Considerations

As fashions evolve, considerations about accuracy and reliability lower over time; in the meantime, safety and compliance considerations enhance. That is linked to widespread media protection concerning mannequin jailbreaking, immediate injection, and mannequin poisoning.

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

Attainable AI agent errors will be linked with the next:

• Incorrect agent ideas primarily based on info/errors obtained from companies by REST APIs/WebSockets/CLI.
• Hallucination.
• Selecting the incorrect software.

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

On the identical time, AI brokers can present clever automated protection. Generally safety specialists want an assistant to assist react to incidents or collect and apply associated insurance policies and updates. Safety engineers typically lack ample time to overview alerts, accumulate information from numerous sources, contemplate particular and historic context, and take applicable motion.

Need extra AI content material? Try Cisco DevNet AI Hub: https://developer.cisco.com/website/ai/