Share this
Table of Contents
Twice a year, Itential holds what we call Product Celebration Week — a dedicated five days where the whole company comes together around the product. For me, this February was my first PCW, and I’ll be honest: I didn’t know what to expect. What I found was an entire organization — engineering, product, and beyond — deeply invested in where this platform is going and excited to push it forward together. This edition added a new dimension to that energy: an AI Agent Hackathon.
But we didn’t just talk about agents.
We built them.
In one week, teams from across the company designed, shipped, and demonstrated 17 working AI agents, spanning infrastructure automation, compliance, DNS, vulnerability management, lifecycle onboarding, real-time observability, checkpoint recovery, and multi-system incident correlation.
This wasn’t experimentation in a sandbox. These were real product extensions. Real integrations. Real road map contributions.
By the Numbers
17
agents created
31
total tool bindings
10+
projects and deterministic workflows leveraged
1
IAG5 task
2
MCP integrations (NetBox; OpenVuln)
4
direct product road map contributions
This week wasn’t about theory. It was about operationalizing AI inside Itential.
The Agents We Built
🛡 Compliance & Security
Compliance Manager Agent
A creative CLI-driven compliance extension that goes beyond Golden Config. It executes arbitrary CLI commands across inventory, parses operational state, evaluates compliance thresholds, and generates structured audit reports — covering real-time operational posture, not just config drift.
Security Vulnerability Assessment Agent
This agent correlates live device software versions with Cisco’s OpenVuln advisories. It pulls current advisories, collects real running versions from devices, cross-references affected releases, and produces severity-ranked remediation recommendations. Manual CVE spreadsheet cross-referencing? Eliminated.
PSIRT Placeholder Agent
This agent fetches Cisco PSIRT advisories by CVE, logs into devices to verify running software and enabled features, and produces a severity-ranked vulnerability report. What sets it apart: a dedicated Comms sub-agent handles all notifications, keeping tooling clean and the orchestration pattern reusable across teams.
🌐 Brownfield & Inventory Intelligence
Service Discovery Agent — Brownfield LCM Onboarding
One of the most important automation gaps in the industry is brownfield onboarding. This agent retrieves live configurations, identifies structured service patterns (starting with L3 sub-interfaces), constructs JSON models, and registers them in Lifecycle Manager — without reprovisioning.
Provisioning is easy. Discovering what already exists at scale is hard. This agent solves that.
Inventory Sync Agent (NetBox MCP)
A fully idempotent synchronization agent that keeps Itential inventory aligned with an external source of truth. It was validated against NetBox but architected to extend to any REST-based CMDB.
Inventory Build & Host Connection Verification Agent
A hands-on exploration of Agent Builder capabilities that:
- creates new inventories
- populates nodes
- retrieves secrets from HashiCorp
- executes Gateway 5 connectivity checks
- produces reachability reports.
It even gracefully handled a failed GitLab retrieval and continued execution with fallback payloads — making it both a practical tool and a learning blueprint for FlowAI newcomers.
🚨 Incident Intelligence & Cross-System Correlation
Network Incident Triage Analyst
A cross-platform reasoning agent integrating ServiceNow and Cisco Meraki. It retrieves active P1 incidents, pulls live infrastructure state, correlates symptoms with device health, and identifies:
- priority inflation
- stale assignments
- misclassified tickets
- silent failures (infrastructure issues with no ticket)
When silent failures are found, it auto-creates P2 incidents and sends a structured executive-ready triage report.
This replaces hours of manual context switching between ITSM and monitoring systems.
🌍 DNS & Idempotent Change Management
Infoblox DNS A Record Creation Agent
Built as a strict ReAct-style goal agent enforcing a four-phase lifecycle:
- Verify (idempotency / conflict detection)
- Act (create record)
- Validate (post-check & optional rollback)
- Notify (HTML email summary)
No blind writes. No duplicate records. No unsafe state transitions.
This is how DNS automation should behave in regulated environments.
📡 Observability, Real-Time Execution & Mission Control
Mission Control for FlowAI
Agents previously ran in the dark. You saw results only after completion.
Mission Control introduced:
- Live mission cards
- Real-time tool tracking
- Token usage monitoring
- Parallel tool visualization
- Cancel / Retry controls
- Animated state transitions
- 3-second polling synchronization
- Mongo-backed mission state tracking
The infrastructure for SSE streaming is already built – pending routing updates.
This was not just a feature demo, it’s a foundational observability layer for agent execution.
⏸ Checkpointing, Pause & Resume
Long-running agents are risky without recovery controls.
The checkpointing contribution introduced:
- Automatic state persistence at tool boundaries
- Pause/resume control
- Cancel mid-execution
- Restart from last stable checkpoint
- Global enable/disable configuration
This fundamentally changes how production agents can be trusted.
🧠 Dynamic Knowledge Base for Agents
Agents hallucinate when they lack authoritative context.
The Knowledge Base feature allows:
- Operator-authored documents
- Category & tag organization
- Agent-level document pinning
- Automatic injection into system prompts
No redeployments. No prompt editing. No brittle system-message hacks.
Agents now reason from enterprise truth, not just training data.
☎️ Twilio Escalation Agent
Automation shouldn’t stop at dashboards.
A Twilio-powered agent enables FlowAI to place phone calls – escalating incidents to NOC or on-call engineers directly.
When an agent detects something critical, it doesn’t just log it. It calls.
Architectural Themes That Emerged
Across all 17 agents, five patterns became unmistakably clear:
Deterministic + Agentic = Powerful.
Many agents were built entirely on deterministic workflows. The LLM orchestrated them — but execution was grounded in structured tasks.
Sequential Tool Discipline Matters.
Explicit ordering of tool calls dramatically improved reliability.
Idempotency Is Non-Negotiable.
The best agents reasoned before writing – pre-checks, validation steps, rollback logic.
Observability Is a Requirement, Not a Nice-to-Have.
Mission Control and checkpointing show that production agents need live visibility, interruptibility, state recovery, and auditable history.
Agents Extend the Product.
Four hackathon submissions directly contribute to the product road map. This was not a side project week, it was forward motion.
What 31 Tool Bindings Really Means
31 tool bindings across 17 agents signals something important:
FlowAI is not a chatbot wrapper.
It is a reasoning layer over:
- Workflows
- Adapters
- IAG services
- Inventory
- MCP integrations
- External APIs
The agents built during this week weren’t toy demos — they interacted with real systems and executed meaningful actions.
What This Hackathon Proved
- Agents can safely operate in enterprise infrastructure environments.
- Observability and checkpointing are critical for trust.
- Brownfield automation is solvable with pattern-based reasoning.
- Cross-platform correlation creates new operational visibility.
- Knowledge injection dramatically improves reliability.
- Deterministic workflows remain foundational.
Most importantly: the team didn’t just experiment with AI. They operationalized it.
What Comes Next
The hackathon surfaced a clear direction:
- Human-in-the-loop approvals
- Parent/child agent visualization
- Multi-agent chaining,
- Slack/Teams notifications
- Tool call inspectors
- Agent thought streaming
The foundation is there.
Final Takeaway
In one week:
17
agents created
31
total tool bindings
10+
projects and deterministic workflows leveraged
4
product contributions
Real-time observability. Checkpoint recovery. Knowledge-grounded reasoning.
This wasn’t just a Product Celebration Week.
It was proof that Itential isn’t just talking about AI agents.
We’re building them – safely, visibly, and in ways that extend the platform in production-ready directions.
And this is just the beginning.
John Capobianco
Head of AI & Developer Relations ‐ Itential
John Capobianco is the Head of AI & Developer Relations at Itential, and a technology leader, developer advocate, and builder at the intersection of AI and network automation. With a career spanning enterprise, government, and cloud networking, John has held roles including Head of Developer Relations at Selector AI, where he focused on AI-driven observability, configuration intelligence, and autonomous network operations, as well as Cisco AI Technical Leader and Senior Network Architect for the Parliament of Canada / House of Commons. He brings deep, hands-on experience applying automation and AI in highly regulated, mission-critical environments. His work centers on helping large organizations adopt AI safely while maintaining reliability, security, and operational trust. John is a former professor at St. Lawrence College, an author, speaker, and educator. He is the author of Automate Your Network (self-published, 2019) and the Cisco Press pyATS book (2024). He regularly shares insights through talks, workshops, and the Automate Your Network brand, with a focus on practical, production-ready AI, developer empowerment, and the evolution of network engineering in an AI-first world.