<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Systems-Engineering | Derek Armstrong — Payments Engineer · AI · Infrastructure</title><link>https://derekarmstrong.dev/tags/systems-engineering/</link><atom:link href="https://derekarmstrong.dev/tags/systems-engineering/index.xml" rel="self" type="application/rss+xml"/><description>Systems-Engineering</description><generator>Hugo Blox Builder (https://hugoblox.com)</generator><language>en-us</language><lastBuildDate>Fri, 20 Nov 2026 00:00:00 +0000</lastBuildDate><image><url>https://derekarmstrong.dev/media/sharing.png</url><title>Systems-Engineering</title><link>https://derekarmstrong.dev/tags/systems-engineering/</link></image><item><title>System Thinkers: The Scarce Role in the AI Era</title><link>https://derekarmstrong.dev/blog/system-thinkers-are-the-most-in-demand-role-in-the-ai-era/</link><pubDate>Fri, 20 Nov 2026 00:00:00 +0000</pubDate><guid>https://derekarmstrong.dev/blog/system-thinkers-are-the-most-in-demand-role-in-the-ai-era/</guid><description>&lt;p&gt;The thing I&amp;rsquo;m building in my homelab right now touches four different disciplines. A camera on a small Linux box, an ESP32 sensor node, a 3D-printed chassis, and a local LLM making the decisions that tie it together. Getting it off the ground meant producing artifacts in each of those domains: a mechanical drawing for the printed chassis, a wiring diagram, a network topology, and a software spec for the agent loop.&lt;/p&gt;
&lt;p&gt;I can code. All four of those domains? No.&lt;/p&gt;
&lt;p&gt;And that&amp;rsquo;s where I kept hitting the wall for years. The wall wasn&amp;rsquo;t &amp;ldquo;I can&amp;rsquo;t build this.&amp;rdquo; It was &amp;ldquo;I can&amp;rsquo;t even start the conversation.&amp;rdquo; A machinist or a contractor can&amp;rsquo;t work from a rambling description, and a specialist won&amp;rsquo;t sit down and learn your problem before producing a first draft. You have to hand over something — a drawing, a spec sheet, a diagram — that says &lt;em&gt;this is the shape of what I need&lt;/em&gt;. That first artifact is the barrier. It sits between an idea and a working system, and for a long time it was gated by domain expertise you didn&amp;rsquo;t have.&lt;/p&gt;
&lt;p&gt;That barrier is collapsing.&lt;/p&gt;
&lt;p&gt;
, a NeurIPS 2024 framework, generates parametric CAD models from text — &amp;ldquo;designer-friendly instructions for all skill levels,&amp;rdquo; with prompts that span from &amp;ldquo;two concentric cylinders&amp;rdquo; to a full parametric spec.
does the same for diagrams: &amp;ldquo;generate a diagram from a sentence, paste Mermaid, or let an agent do it.&amp;rdquo; And I run my own local text-to-CAD pipeline: describe a part, get a printable first-draft model in minutes. Not the final design — the first draft. The thing that starts the conversation.&lt;/p&gt;
&lt;p&gt;The demand data points the same direction. The World Economic Forum&amp;rsquo;s
puts analytical thinking as the top core skill — seven out of ten companies call it essential — with AI and big data among the fastest-growing skills. Systems thinking is among the skills the report says &amp;ldquo;solidify their importance, emphasizing the continued relevance of human-centric skills amid rapid technological advances&amp;rdquo; — and it ties that rise directly to autonomous systems: &amp;ldquo;skills essential for managing and optimizing interactions with autonomous technology.&amp;rdquo; Of the more than 2,800 skills they assessed, zero had a &amp;ldquo;very high capacity&amp;rdquo; to be replaced by current generative AI. And the Bureau of Labor Statistics
growing much faster than average through 2035, roughly 174,800 openings a year.&lt;/p&gt;
&lt;p&gt;What&amp;rsquo;s interesting to me is the combination. The skill that&amp;rsquo;s rising is the one that used to be gated by domain. A system thinker is someone who can hold a problem that spans software, hardware, networking, and data in their head at once. Ten years ago that meant either being lucky enough to have multi-domain experience or having a team. Now it means you can pull a basic prototype from each domain yourself, because the first artifact in each domain is cheap to produce.&lt;/p&gt;
&lt;p&gt;That&amp;rsquo;s the part nobody talks about: cross-discipline system design is now doable by one person. A payments engineer can prototype a soil-moisture sensor, a camera streaming service, and an automation pipeline, because the first-draft artifact no longer requires years of domain study. And it&amp;rsquo;s not just the non-specialist winning — experts get the same tooling. AI translates spec to diagram, codebase to architecture diagram, domain jargon to plain language. Excalidraw&amp;rsquo;s &amp;ldquo;let an agent do it&amp;rdquo; mode is that in one sentence. Conversations start earlier, loops get shorter, and the whole workflow gets simpler on both sides of the table.&lt;/p&gt;
&lt;p&gt;There&amp;rsquo;s a number worth knowing:
, one of the bigger &amp;ldquo;vibe coding&amp;rdquo; app builders, reports that 80% of its builders are non-technical. Self-reported, unaudited — take it with that. But the direction is hard to argue with.&lt;/p&gt;
&lt;p&gt;Here&amp;rsquo;s the part I&amp;rsquo;d say to the people who think this is hype: AI is not doing everything for you. It&amp;rsquo;s not doing the last mile.&lt;/p&gt;
&lt;p&gt;
ran a randomized controlled trial with experienced open-source developers in 2025 and found that giving them early-2025 AI tooling actually made them 19% &lt;em&gt;slower&lt;/em&gt; on their own codebase — while the developers themselves felt about 20% faster. Perception and reality were nearly forty points apart.
at least 30% of generative AI projects get abandoned after proof of concept by the end of 2025. The first draft is not the answer. Verification, integration, knowing where the seams are and whether the model is actually right — that&amp;rsquo;s still human work, and it doesn&amp;rsquo;t get delegated.&lt;/p&gt;
&lt;p&gt;And that&amp;rsquo;s exactly why the system thinker is the scarce role. The last mile is where the system shows itself — where the CAD drawing fights the enclosure, where the spec fights the reality of the data. Only the person holding the whole picture can see it, because it only exists at the intersections.&lt;/p&gt;
&lt;p&gt;Here&amp;rsquo;s where I land. If you can think in systems, you can basically build anything you can imagine — not because AI does it for you, but because one thing leads to the next, and the tooling gets you moving. You don&amp;rsquo;t need to be the best in every domain. You need to be the person who can hold the entire system in their head and get a first artifact into every domain it touches — and be deep enough in at least one of them to know when the first draft is wrong.&lt;/p&gt;
&lt;p&gt;We&amp;rsquo;re in an era of systems engineering, and it&amp;rsquo;s a little crazy. For decades the discipline lived in the people who got paid to run programs. Now the leverage is in the person who can see the whole system and start the conversation in every domain at once — with a tool that will do the first draft of the drawing, the spec, and the diagram for them.&lt;/p&gt;
&lt;p&gt;How AI as a barrier-lowering, generalist tool changes the way we interact with technology is a whole different conversation — that&amp;rsquo;s a post for another day. This one is about why the person who can hold the whole system is the scarce one in the room.&lt;/p&gt;
&lt;h2 id="next"&gt;Next&lt;/h2&gt;
&lt;ul&gt;
&lt;li&gt;
— the first artifact is a communication problem more than a technical one.&lt;/li&gt;
&lt;li&gt;
— the personal-adaptation side of the same shift.&lt;/li&gt;
&lt;/ul&gt;</description></item></channel></rss>