NA012 - The Oblong Table

Posted on October 19, 2026 • 6 min read • 1,264 words
Podcast   Network-Automation   Ai   Ai-Native   Community   Autocon   Conferences   Security   Ai-Policy  
Podcast   Network-Automation   Ai   Ai-Native   Community   Autocon   Conferences   Security   Ai-Policy  
Share via
John Howard, Brian Jeggesen and Bart Dorlandt join Steinn and Urs at the oblong table for an unstructured conversation about why network automation adoption is still stuck, what AI actually changes for infrastructure teams, and what conferences and communities owe the people who show up.
Episode Overview   At the table   Listen to the show on YouTube:   Listen to the show anywhere:   Show notes resources:   What we cover:   Community Events and the Adoption Problem   Systems Thinking and Silos   AI, Salaries and Who Adopts First   Security in the AI Era   AI Assisted, Augmented, Native   AI Policy and Guardrails   Conferences, Talks and Sharing   AI Reality Check   Key Takeaways  
NA012 - The Oblong Table
Photo by Steinzi

Network Auto Magic Podcast  

The Oblong Table: AI, Automation Adoption and Community

Episode Overview  

After a hiatus spent trying to keep up with the AI-fuelled madness, Steinn and Urs pull up chairs at the oblong table — not a round table, because nobody here is unbiased. John Howard, Brian Jeggesen and Bart Dorlandt join for an unstructured conversation that starts with Brian’s first CPH^Automate in Copenhagen and the question nobody can quite answer: after fifteen years, why is network automation still the niche of the niche? From there it goes everywhere. John pushes back on the idea that network engineers can’t think in systems, the panel works through the Trail of Bits model of AI assisted, augmented and native organizations, and Urs makes the case that you could replace the word “AI” with “automation” and lose nothing. Along the way: vendors packing a hundred CVEs into a single advisory, port knocking with MFA, tiered AI policies for one-person shops, why conference talks should show the how and not just the result, and John’s verdict on the “moving goalposts” argument — the goalposts haven’t moved, the ball has.

At the table  

  • John Howard: Head of Network Infrastructure at Proton, AutoCon advisory board member, and — by his own admission — the person who writes a lot of text in Slack.
  • Brian Jeggesen: Organizer of CPH^Automate, the first network automation day in Copenhagen, and a self-described professional lurker.
  • Bart Dorlandt: Organizer of NLNAM, the Netherlands Network Automation Meetup, and one of the many Barts of the network automation community.

Listen to the show on YouTube:  

Listen to the show anywhere:  

Listen now!

Show notes resources:  

What we cover:  

Community Events and the Adoption Problem  

  • CPH^Automate: Brian on running the first network automation day in Copenhagen, filling up in the last month, and keeping the organic community feel
  • Automation Is Still the Niche of the Niche: Fifteen years in, and the network automation survey is a few hundred people out of hundreds of thousands of engineers
  • “Generate Configs Faster”: Why so many engineers still think that’s all network automation means
  • Git Is a Hurdle: Brian on missing software development foundations, and in-house development still being a swear word in telecom

Systems Thinking and Silos  

  • Mean Time to Innocence: John’s pushback — years of “is it the network?” turn network engineers into systems thinkers whether they know it or not
  • The Organizational Problem: You can’t diagnose a server you’re not allowed to SSH into
  • Complacency and the XKCD Trap: CLI heroes, ROI calculations that always say “not yet”, and the foundational cost that only gets pushed further down the road

AI, Salaries and Who Adopts First  

  • Investor Math: Promised 200%, willing to settle for 30% — and what that means for dev vs ops salaries
  • Security, Then Coding, Then Infra: John’s adoption order, and why infrastructure tooling only became good enough recently
  • You Can’t Sandbox a Data Center: Urs on why infra will always be more careful — code you can re-run, a data center you might have to drive to

Security in the AI Era  

  • CVE Packing: Single vendor advisories with a hundred-plus vulnerabilities, no details and no triage options
  • Close Your Border: John on Knocknoc and just-in-time access — authenticate, click, and the firewall opens for your IP only while you’re logged in
  • Patch Everything Anyway: Urs on why securing the border alone isn’t enough when the problems are already inside

AI Assisted, Augmented, Native  

  • Three Levels: Assisted (tools to do your job faster), augmented (workflows redesigned around AI), native (the organization built assuming agents are part of the process)
  • Swap the Word: Replace “AI” with “automation” and the model still holds — assisted is a Python script, native is telemetry kicking off a workflow on its own
  • The Maturity Matrix: Not engaged, capable, adoptive, transformative — and why sharing is the step that matters

AI Policy and Guardrails  

  • Incident Summaries for Managers: An AI turning a fifty-engineer chat thread into a status PDF every thirty minutes, so nobody fixing the problem has to do the communication
  • Privacy for Small Shops: Steinn’s tiered approach — an agent gateway with PII detection on MCP calls, routing flagged requests to a model you control
  • Green, Yellow, Red: John on why one approved tool and a clear grid of what’s allowed makes people adopt faster, not slower
  • The System Prompt as a Last Line of Defense: Catching accidental exposure of customer data and routing it to the right team

Conferences, Talks and Sharing  

  • Show the How: John on talks that inspire vs talks you can apply — Naveen’s homebrew Temporal setup vs “I went from 30% to 10%”
  • “What’s in It for Me?”: Building a deck for the audience rather than for your next promotion
  • Who Are Conferences For?: Urs and Bart on beginners vs fifteen-year veterans, and why small events can’t run four parallel tracks
  • The Hallway Track: Would you still go to a conference in the age of AI? (Yes.)
  • We All Suck at Sharing: Too many secrets shared in hallways, not enough blogs, and a landscape that overwhelms newcomers

AI Reality Check  

  • Late to AI? You’re Better Off: Most of what enterprises built from 2022 to 2024 is table stakes today
  • The Goalposts Haven’t Moved, the Ball Has: John on why older models still wouldn’t be good enough for production network engineering
  • The Mr. Manager Agent: Steinn’s pragmatic-boss persona that asks who pays for it, who supports it, and points out it could have been a cron job in 1997
  • Linus and the Whole Thread: Why quoting one block of text out of context gets the AI debate wrong
  • Spotting Claude Slides: Three boxes, a slim bar, and that shade of orange

Key Takeaways  

  1. Automation Is Still a Niche: The tools exist. The foundations — git, software practices, systems thinking at the organizational level — are what’s missing
  2. AI Is Just Different Automation: Assisted, augmented, native works just as well with the word “automation” in it
  3. Clarity Speeds Adoption: One approved tool and a clear list of what’s allowed beats a vague “be careful”
  4. Layer Your Defenses: Close the border, patch everything, and put guardrails in the system prompt too
  5. Give the Audience Something to Take Home: Show the how, not just the result
  6. Starting Late Isn’t Losing: Today’s tools are better and cheaper than what early adopters paid for
  7. Be Local: Go to your local community, share what you’ve built, and make it work
On this page: