Reykjavík, Iceland · online & opinionated

I help AI survive contact with the business for money.

I’m Steinn Örvar Bjarnarson — Steinzi is easier. I turn messy processes, legacy systems, and ambitious AI ideas into useful automation that real teams can trust.

Portrait of Steinn Örvar Bjarnarson, smiling with two thumbs upCCIE #60715less AI theatre,
more working systems
15+years in real systems
#60715CCIE, earned at 27
5AutoCon appearances
11podcast episodes

The Steinzi pitch, since you asked

I know what happens after the clever demo.

I am at my most useful when the process is messy, the systems are old, and somebody has already said, “couldn’t we just use AI?”

Fifteen-plus years in infrastructure taught me to care about failure, access, operations, and the people on call. Product development taught me to make all of that usable. AI is the newest tool in the box, not an exemption from reality.

01

Find the real problem

Trace the actual process across people, spreadsheets, APIs, ticket queues, and legacy systems before anyone spends six months automating the wrong bit.

Process archaeology, but with fewer little brushes.
02

Build the useful version

Turn the good idea into an operable system: agents, MCP tools, APIs, workflows, tests, approvals, observability, and very clear places where the AI is not allowed to improvise.

The demo is allowed to become infrastructure.
03

Make it safe enough to respect, not trust

Keep credentials out of model context, gate access by session and role, put humans on the consequential decisions, and design failure as a normal operating condition.

Useful autonomy. Adult supervision included.
04

Get humans to use it

Polish the interface, fit the tool to the team, teach the why, document the awkward bits, and listen when the people doing the work say the clever solution is annoying.

Adoption is a feature. Apparently.

The long version, still aggressively abridged

From soldering irons to AI agents.

The through-line was never “networking.” It was taking complicated systems, making them safer and easier to operate, and then convincing humans the new way is less painful.

Act 1

Hardware first. Consequences included.

Started in electronics, physical security, CCTV, and IT services before moving deep into network operations. The machines were physical, the customers were real, and a bad change could ruin everyone’s afternoon.

Act 2

Expertise, then a deliberate escape from the box.

Joined Advania as a principal network engineer, became Iceland’s youngest CCIE at the time, taught networking, earned a computer-science degree at night, and learned enough software to stop accepting repetitive work as a law of nature.

Act 3

Automation became a product, not a folder of scripts.

Moved into lead product development, built cross-vendor automation and source-of-truth systems, replaced legacy tooling, and learned that the hard part is rarely the API. It is the process, the ownership, and Gary’s spreadsheet.

Act 4

AI met the business. Both survived.

Started speaking publicly about generative AI in 2023 and now builds controlled AI automation across infrastructure and business processes: internal MCPs, authorized sessions, human approval, secret-safe tooling, and systems people will actually use.

Selected work

The public bits
I can show you.

Open-source tools, a podcast, community building, and enough public material to prove this is not a very elaborate LinkedIn phase — plus one suspiciously blank file.

01 / Podcast2025—now

Network Automagic

A long-form show with Urs Baumann about AI, automation, infrastructure, the people building the tools, and what survives outside a demo environment.

11 episodesvideo + audiointerviews
Explore the show
02 / Community2025—now

ISNOG

I founded the Icelandic Network Operators Group to give the local infrastructure community a room, a stage, and a reason to meet. Event zero brought together roughly 100 people.

Icelandcommunity~100 at ISNOG #0
Visit ISNOG
03 / Open source2024—now

Network Automation Landscape

A community-maintained map of the sprawling tools, products, and projects that make up modern network automation.

50+ GitHub starscommunity-maintainedlandscape2
Open the landscape
04 / Open source2026

netlab-mcp

An MCP server that wraps netlab so language models can work from rendered, lab-tested network configurations instead of confident guesses.

PythonMCPcontainerlab
View on GitHub
05 / Classified[redacted]—now

The actually cool stuff.

I wish I could tell you about the coolest stuff I work on, but the lawyers made me sign an NDA and took my firstborn son. I miss him.

Approved by absolutely nobody

Network Automagic

AI, automation, and infrastructure —
with the varnish removed.

Urs Baumann and I talk to the people building the tools — from Temporal and Pulumi to agents, MCP, and everything between “nice demo” and production. Occasionally somebody says “agentic” and we make them explain what they mean.

Watch on YouTube

2018 — now · independently clickable

The public record.

Talks, workshops, releases, guest appearances, articles, and community work — each linked to a public source. Not every commit or LinkedIn thought; the durable things worth making another human click. Because vibes are not a citation.

Build

Released netlab-mcp

Published an Apache-2.0 MCP server that turns network intent into rendered configurations and can validate them in a netlab + containerlab environment.

View Released netlab-mcp
Podcast

Network Automagic 008 — MCP vs CLI

Hosted Calvin Remsburg to examine deterministic CLIs, probabilistic tool use, agent security, and how network engineers should prepare.

View Network Automagic 008 — MCP vs CLI
Podcast

Network Automagic 007 — Pulumi

Hosted Scott Lowe for a frank look at Pulumi, infrastructure as code, provider quality, and why network automation remains harder than cloud automation.

View Network Automagic 007 — Pulumi
Podcast

Network Automagic 006 — AI Hype

Hosted Itential chief architect Peter Sprygada to separate practical AI, MCP, and deterministic execution from the industry hype cycle.

View Network Automagic 006 — AI Hype
Community

Founded and ran ISNOG #0

Brought roughly 100 Icelandic network professionals together for the country’s first Network Operators Group meeting, with speakers from Nokia, NetBox Labs, academia, local operators, and vendors.

View Founded and ran ISNOG #0
Speaking

Network Testing with NUTS at AutoCon 3

Co-proctored a sold-out four-hour workshop in Prague with Urs Baumann, teaching NUTS, pytest, custom reporting, and the INPG stack.

View Network Testing with NUTS at AutoCon 3
Podcast

Network Automagic 003 — NUTS

Hosted Marco Martinez for a deep dive into the Network Unit Testing System, testing architecture, integrations, and pre/post-change validation.

View Network Automagic 003 — NUTS
Podcast

Network Automagic 002 — Infrahub

Hosted OpsMill CEO Damien Garros to discuss flexible schemas, source-of-truth architecture, and whether Infrahub replaces NetBox.

View Network Automagic 002 — Infrahub
Writing

The EX4000: Entry Level Never Looked So Good!

Published a practical review of Juniper’s EX4000 access switch: fast boot, perpetual PoE, Mist operations, and the feature trade-offs behind the price point.

View The EX4000: Entry Level Never Looked So Good!
Speaking

“MVP — The Keys to Automation” at AutoCon 1

Told the unvarnished story of replacing a legacy IPAM with NetBox: the architecture, the wrong turns, the useful rules, and why getting started beats waiting for the immaculate platform.

View “MVP — The Keys to Automation” at AutoCon 1
Speaking

Explained practical AI at Garðabær Library

Ran a public session on what AI is and how ordinary people can use it — part technology briefing, part community service, and unusually low on enterprise architecture diagrams.

View Explained practical AI at Garðabær Library
Speaking

MSP automation at Juniper’s Product Strategy Summit

Presented in Nice on using Mist APIs and AI-driven operations to make enterprise networks behave more like well-run data centers. The slides survived; the hotel coffee is unconfirmed.

View MSP automation at Juniper’s Product Strategy Summit
Podcast

The Hedge 169 — Network Address Translation

Joined Russ White to discuss the history, legitimate uses, creative misuses, and design trade-offs of NAT after researching rather more translation tables than is medically advisable.

View The Hedge 169 — Network Address Translation
Build

Published Hotsarch, the BSc project with a NAT problem

Co-built a secure web platform that normalized live NAT data from multiple firewalls and load balancers without giving end users direct access to the devices. Also wrote the long history of NAT hiding in the appendix.

View Published Hotsarch, the BSc project with a NAT problem
Milestone

Finished a BSc in computer science — at night

Went back to university while working full time to add proper software and computer-science foundations to years of infrastructure experience. “It works” was no longer a sufficient design document.

View Finished a BSc in computer science — at night
Milestone

Became Iceland’s youngest CCIE at the time

Earned Cisco CCIE Routing & Switching #60715 at 27, becoming the youngest person in Iceland to hold the certification at that point — useful proof that stubbornness can occasionally be productized.

View Became Iceland’s youngest CCIE at the time

About

Builder.
Teacher.
Professional skeptic.

I’m a Lead Product Development Engineer at Advania, working where AI, software, operations, legacy systems, and business processes collide — usually at speed.

I started in electronics and physical security, spent years designing and operating business-critical networks, taught networking on the side, and went back to university at night for a BSc in computer science. Along the way came CCIE #60715, CKAD, RHCSA, and a suspicious number of other certifications because apparently I used to enjoy exams.

Now I use models, agents, MCP, APIs, and dependable workflows to remove friction from infrastructure and business processes. Away from the confidential work, I host Network Automagic, founded ISNOG, maintain open-source projects, teach workshops, and occasionally explain that automating a bad process only makes the bad process faster.

Conferences, workshops & questionable panel decisions

Put me on a stage.
I’ll bring the useful skepticism.

I speak about AI automation that works beyond the demo, agents and MCP, business processes, infrastructure, integrating modern ideas with legacy systems, and the much harder problem of getting teams to actually use the tooling. If your audience likes practical ideas and honest failure stories, we should talk.

AI beyond the demoHow not to automate a messAgents, MCP & toolsLegacy systems, modern automationGetting teams to actually use the thingFrom networks to everything

Give me the useful bits.

No funnel. No mystery database. This opens a pre-filled email; nothing leaves until you send it.

Not a conference?

Something else entirely?
My inbox still works. Against all odds.

[email protected]