Sample Page Title

“The shift for builders is extra radical than we predict.”

— Jeetu Patel, President and CPO, Cisco

AI is working full throttle, leaving a wake of radical modifications for software program builders. We’re getting into a time the place AI can write code, name instruments, and execute advanced workflows—all from a single immediate. This shift has huge implications. Learn Jeetu’s weblog to study extra.

Radical shifts are on the horizon for extra than simply builders.

What about AI’s affect on community engineers?

In my earlier weblog, I described MCP—Mannequin Context Protocol—and the way agentic AI may lastly converse our language, perceive our networks, and take significant motion. Now I wish to present you what occurs when that dialog goes one step additional: when the agent doesn’t simply perceive what’s damaged however fixes it with out being instructed how.

From detection to motion: a self-healing community

A self-healing community isn’t a hypothetical “what-if.” That is agentic AI dealing with one of the vital irritating points in community operations: configuration drift.

Let’s break down the method of constructing self-healing occur.

On this setup, I deployed two MCP servers—one integrates with my occasion of Splunk Enterprise, and the opposite integrates with my Meraki dashboard. What makes this particular isn’t simply the device integration—it’s that the agent could make autonomous selections throughout each instruments primarily based on a easy immediate.

Server 1: Splunk MCP

On the Splunk MCP server, we begin with a pure language immediate like this:

• “What are my Splunk Indexes?”
• “Search the meraki_index for any gadget connectivity points or standing modifications within the final 4 hours.”
• “Present me all community configuration modifications and group them by the one who made them.“
• “Analyze community visitors patterns and establish any anomalies that would point out safety threats.“

The agent processes the request—below the hood— like this:

The Splunk MCP server makes use of the Splunk SDK to question actual log information. Its job is to detect if one thing within the surroundings—like a configuration change—has occurred from what we anticipate, examine it to our supply of reality, and treatment it.

Let’s attempt self-healing with an agent we’ll name “Community Pharaoh.”

Right here’s a superb immediate to start out:

Immediate > I must see what’s occurring with my Meraki community. Are you able to present me the most recent Splunk alerts, particularly within the meraki_index? I would like you to look within the spath the place the supply is Meraki Community solely inside the previous 3 weeks.

You’ll discover that I didn’t should be particular about how one can search. I simply wanted to inform the agent what I used to be after.

Spectacular, proper? Right here’s what Community Pharaoh did:

1. Preliminary search try – Tried looking out meraki_index with supply= “Meraki Community” however obtained a syntax error.
2. Question format repair – Adjusted Splunk search syntax to correct format.
3. Broad reconnaissance – Searched the complete Meraki Index to know the information construction and obtainable sources.
4. Knowledge evaluation – Discovered two essential sources: “Dwelling Community” (port occasions) and “Meraki Community” (config modifications).
5. Focused extraction – Centered particularly on “Meraki Community” supply as requested.
6. Alert parsing – Extracted key fields: alert sorts, ranges, gadgets, and alter particulars from JSON payload.

This was all by itself, together with the self-correction and retries to get the consequence I requested.

Server 2: Meraki MCP

The second MCP server is the place the “self-healing comes into play.”

After the Meraki MCP server receives the detected change (for instance, gadget IP tackle change), it makes use of the Meraki dashboard API to reverse that change. No handbook directions, no pre-programmed response chain. The agent understood that the change represented drift and took motion to revive alignment.

Key takeaways

• I didn’t have to jot down a hardcoded if-then circulation between the 2. I simply outlined the instruments and gave the agent context. The agent selected the suitable device, chosen the right operate, and acted completely autonomously.
• I outlined the instruments decorators to make obtainable in my Meraki MCP–nothing earth-shattering–easy features that execute one factor and one factor solely—get a listing of my gadgets, replace my gadgets, and so forth— all of which community engineers have probably used and coded.

That is what occurs while you let intent drive the motion and let the agent do the orchestration. It’s easy, scalable, and highly effective.

Now, let’s have a look at how the agent self-heals our community with Meraki MCP [that includes actual output].

First, we’ll get a diff of what was modified.  

Immediate > This Kareem Iskander dude shouldn’t have made any modifications to the community. Unacceptable! Are you able to present me side-by-side what was modified?

As soon as once more, spectacular! Discover that the data is being pulled from Splunk by the Splunk MCP server. Additionally, discover how our agent gave us ideas on how one can revert the modifications. As soon as once more, spectacular! Discover that the data is being pulled from Splunk utilizing the Splunk MCP server.

Additionally, I’d prefer to level out how our agent gave us ideas on how one can revert the modifications mechanically utilizing the obtainable API endpoints within the Meraki MCP! I didn’t need to specify which Meraki group or community the gadgets belong to, nor did I’ve to specify the gadget kind. Community Pharaoh knew the hierarchy of the Meraki dashboard and traversed it!

Now, it’s time to heal the community!

Immediate > NetP Let’s revert the configuration to its unique state for all of the modifications you have got detected!

Why this issues

This isn’t only a enjoyable facet challenge. It addresses an actual ache level for all community engineers: configuration drift!

Whether or not it’s unintentional modifications, unauthorized edits, or misalignment with the supply of reality, config drift results in downtime, compliance points, and infinite handbook cleanup. Agentic AI affords a greater mannequin: detect, perceive, and repair mechanically.

I simply took two steps and let the agent run with it:

• Outline the device interfaces (Splunk SDK + Meraki API)
• Register these instruments with MCP

That is the ability of constructing agentic programs on prime of the workflows we already know.

What abilities do you really want?

Let’s preserve it actual. Listed here are the talents required:

• Coding with Python
• Understanding SDKs and how one can use them
• Community automation and programmability with APIs
• MCP framework to construction device entry and execution
• Networking abilities

The place that is all going

Let’s zoom out for a second to higher perceive the large image.

What I’ve constructed right here—a self-healing community utilizing two MCP brokers—isn’t a prototype. It’s a sensible preview of Cisco’s broader imaginative and prescient.

Within the AI Canvas announcement, Cisco laid the inspiration for the agentic period: modular brokers that work with our instruments, perceive our intent, and take autonomous motion.  This demo suits proper in.  One agent detects drift by Splunk, one other acts by Meraki—all with only a immediate and some registered device features.

Now think about layering in Cisco’s Deep Community Mannequin—a complete, machine-readable understanding of your complete community, skilled on years of CCIE-level Cisco experience and telemetry, and a set of pre-built brokers prepared out of the field.

As a substitute of merely reversing a misconfigured VLAN, the agent understands:

• Which purposes depend upon that VLAN throughout hybrid environments
• Whether or not the change launched a segmentation violation or efficiency regression
• The way to resolve the difficulty with out disrupting essential enterprise visitors
• The way to replace the supply of reality to mirror any reputable intent behind the change

That is the place the idea turns into a actuality:

• AI Canvas offers us the surroundings and brokers.
• The Cisco Deep Community Mannequin offers brokers the situational intelligence to behave with context.
• MCP offers us the extendibility to BYOA (deliver/construct your personal agent, which is a future function).

And that’s what community engineers want—not one other platform, however an assistant that will get it; one that may motive like us, function sooner than us, and make selections we belief.

It’s time to sit down within the driver’s seat

This isn’t a one-off. It’s a multiplier. Collectively, AI Canvas, Cisco Deep Community Mannequin, and MCP put community engineers within the driver’s seat of this new agentic AI period. As Jeetu additionally mentioned, “The long run is coming sooner than you assume.”

Keep forward of the curve and be a part of the extraordinary.

For a totally working code of this demo, take a look at my GitHub repository.

Unlock the way forward for expertise with synthetic intelligence coaching in Cisco U.

Discover AI studying and begin constructing your abilities at present.

Learn extra from the AI Break sequence:

Join Cisco U. | Be part of the  Cisco Studying Community at present totally free.

Study with Cisco

X | Threads | Fb | LinkedIn | Instagram | YouTube

Use  #CiscoU and #CiscoCert to affix the dialog.

Share: