Network Automagic

Ivan Pepelnjak joins Steinn and Urs to talk about netlab, why multi-vendor ACLs are a hellscape, and what AI can and cannot do with network configuration.
Ivan is a CCIE Emeritus, the author of more Cisco Press books than most of us have read, the voice behind ipSpace.net, and the creator of netlab — an open source tool that builds and configures virtual labs across roughly twenty platforms. He describes it himself as an "intent based digital twin lifecycle management system", then immediately admits that's a bullshit bingo winner if you have the right card.
The conversation digs into why LLMs write decent Python and bad router configs (vendor CLIs are almost identical but not quite, which is the worst possible training data), the ChatGPT doom loop Ivan hit trying to configure EVPN on Nokia SR OS, and the endless device quirks behind multi-vendor abstraction — Juniper loopbacks, Nokia prefix lengths, and one vendor with roughly 400 keywords you are not allowed to use as a VLAN name.
Plus: why every layer 3 switch is really a bridge in disguise, why ACLs are containable and QoS never will be, what a real CCIE engagement looks like when you have twenty years of customer mistakes and exactly one command to fix them, and Ivan's controversial opinion on conferences.
What we cover
What netlab actually is, and the five years behind it
Whether netlab is a crutch for beginners
Splitting a lab across multiple servers with automatic VXLAN tunnels
Netlab vs containerlab and Kubernetes-based labs
Why AI generates good Python and bad router configs
The ChatGPT doom loop on Nokia SR OS EVPN
Cloning production into a lab: control plane works, data plane is a lottery
ACL support in netlab, Cartesian products and resequencing
AI code review vs AI code — technically correct and five times too long
Vibe coders, 1982 edition, on a ZX Spectrum
Leaky abstractions and worst common denominators
Device quirks: Juniper, Nokia, Cisco, and 400 forbidden VLAN names on EXOS
Why every layer 3 switch is a bridge with a router attached
ACLs are containable. QoS is not.
BGP labs designed like gym machines
Why CCIE lab scenarios are so insane — and why they're accurate
The power of saying no, and the customer who said thank you
Controversial opinion: the hallway track is the only track
Links from the episode
netlab: netlab.tools
netlab on GitHub: github.com/ipspace/netlab
netlab examples: github.com/ipspace/netlab-examples
Ivan's blog: blog.ipspace.net
Free BGP labs: bgplabs.net
Free IS-IS labs: isis.bgplabs.net
Free VXLAN/EVPN labs: evpn.bgplabs.net
Batfish: github.com/batfish/batfish
Aerleon: github.com/aerleon/aerleon
Guest
Ivan Pepelnjak — ipspace.net | netlab.tools
Hosted by Steinn "Steinzi" Bjarnarson and Urs Baumann.
Subscribe and follow
Watch on YouTube: youtu.be/odcwCByRj6g
Channel: youtube.com/@networkautomagic
Web: networkautomagic.net
#networkautomation #networkautomagic #netlab #ipspace #AI #networkengineering

NA010 - The Future of Platform Engineering with Stuart Clark

Hairdresser to CCIE Automation. Cisco DevNet to AWS to Spotify. Stuart Clark joins Steinn and Urs for a wide-ranging conversation on internal developer portals, AI in production, and what 8 years across DevRel actually taught him about driving revenue instead of vanity metrics.

The conversation digs into Spotify's Backstage and the new Portal SaaS offering, why IDPs become a "voice to chaos" the moment AI agents start checking in code, what really happened to Cisco DevNet after the leadership changes, and Stuart's signature Ford RS Turbo metaphor for organizations bolting on horsepower without ever upgrading their brakes.

Plus: Claude Code knocking 221 broken links to zero in an hour, the AI mania trap and why Stuart signs out of work apps after 6pm, the conference-room reality that 99% of devs use AI but only 2% run it in production, and Stuart's closing hot take on not letting your job define you.

What we cover:
• Stuart's hairdresser-to-CCIE-Automation career arc
• What happened to Cisco DevNet after Susie Wee and Mandy Whaley left
• Working faster, not harder, with AI assistants
• Setting boundaries in the always-on AI era
• DevRel as revenue pipeline, not vanity metrics
• Spotify's Backstage and the Portal SaaS offering
• Why IDPs are a "voice to chaos" once AI agents are committing code
• Skills files, golden paths, and CI/CD
• The 99/2 gap between AI use and AI in production
• The Ford RS Turbo brakes metaphor for platform engineering
• Why a strong fundamentals CCIE still beats chasing every shiny thing
• Why AI companies still ask candidates to write Python by hand

Guest:
Stuart Clark, Senior Developer Advocate at Spotify — linkedin.com/in/stuarteclark

Hosted by Steinn "Steinzi" Bjarnarson and Urs Baumann.

Subscribe and follow:
YouTube: youtube.com/@networkautomagic
Spotify: Network AutoMagic
Apple Podcasts: Network AutoMagic

#networkautomation #networkautomagic #platformengineering #backstage #spotify #devnet #ai #devrel #podcast

Seven network automation engineers. One table. Zero filter. In our first-ever roundtable episode, Steinn and Urs are joined by Marco Martinez (Swisscom), Bart Dorlandt (Dream Networking), John Howard (ProtonMail), Joseph Nicholson (NTT DATA), Mark Prosser (Nokia), William Collins(Itential), and Ryan Shaw for an unfiltered conversation about AI's real impact on network engineering. The panel digs into what AI has actually changed day-to-day, whether the "force multiplier" framing holds up, the danger of vibe coding without validation, AI mandates from management who read one LinkedIn post, knowledge atrophy vs getting dumber, why critical thinking is the meta-skill of the AI era, and whether code quality is a moral virtue or just a pricing tier. Expect sharp disagreements on OSPF vs IS-IS, a 1910 engineering quote that predicted all of this, and Marco's call to action: if AI makes you 30% faster, invest that time in quality — not just more output. What we cover:• AI as a force multiplier — the xkcd ROI chart has been rewritten• The cognitive shift from code generation to code verification• AI mandates and the unhealthy pressure of inflated expectations• Are we getting dumber or just shifting context?• Vibe coding, neural fatigue, and the ownership problem• Code quality as a pricing tier — script vs service• Critical thinking in the age of information overload• The next generation of engineers and how they learn differently Guest LinkedIn profiles:Marco Martinez — linkedin.com/in/marcom4rtinezBart Dorlandt — linkedin.com/in/bartdorlandtJohn Howard — linkedin.com/in/jonnyhowardJoseph Nicholson — linkedin.com/in/joseph-nicholson-b8413533aMark Prosser — linkedin.com/in/mark-prosserWilliam Collins — linkedin.com/in/william-collinsRyan Shaw — linkedin.com/in/ryan-shaw-3a474169 Hosted by Steinn "Steinzi" Bjarnarson and Urs Baumann. Subscribe and follow:YouTube: youtube.com/@networkautomagicSpotify: Network AutoMagicApple Podcasts: Network AutoMagic #networkautomation #networkautomagic #ai #networkengineering #automation #podcast

Steinn and Urs are joined by Calvin Remsburg — network engineer turned AI security specialist at Palo Alto Networks, formerly at Juniper — to break down the MCP vs CLI debate for AI agents, why deterministic tooling is beating probabilistic tool invocation, and what network engineers should be doing right now to stay ahead.

Network Automagic episode 007

Scott Lowe (Cisco/Isovalent, former Pulumi DevRel) joins Steinn and Urs to break down Pulumi — the infrastructure-as-code tool that lets engineers ditch HCL for real programming languages. The conversation covers what makes Pulumi tick, how it bridges Terraform providers, the honest state of the company, and why network automation remains fundamentally harder than cloud or server automation. Scott and Urs dig into the idempotency trap of mixing imperative code with declarative engines, rollback pitfalls, and provider quality across vendors. Scott closes with a controversial opinion: network engineers who stop caring at the server boundary are setting themselves up to get left behind.