The Post-Hype Year in Tech | ✉️ #82
Hey! 👋
Looking back at the last 12 months of 2025 with brutal honesty, the IT industry finally stopped selling revolutions and started getting paid for things that actually work. Less hype, fewer futuristic slides, and more “this is running in production and not falling over.” The narrative shifted: scaling, cost control, and systems that don’t depend on three burned-out engineers and a forgotten README started to matter again.
AI stopped being a fun experiment and turned into uncomfortable infrastructure. Expensive if you don’t know what you’re doing, risky if you deploy it carelessly, and impossible to ignore if you want to stay competitive. The conversation moved from “let’s add an LLM here” to “who’s paying for this, who’s maintaining it, and what happens when it gets things wrong?” The companies that made it through were the ones that integrated AI with intent, not the ones that slapped it everywhere as a marketing badge.
At the same time, cloud went through its own reality check. The era of “everything is serverless by default” started to smell like lazy architecture and unnecessary spend. Efficiency, clear boundaries, predictable costs, and real engineering trade-offs were back on the table. Kubernetes didn’t disappear (it won’t), but it stopped being the automatic answer to every problem. Architecture became a verb again.
And maybe most importantly, IT work itself matured. Fewer shiny titles, more real accountability. Less obsession with “rockstars,” more teams that function without heroes. The market made one thing clear: knowing tools is not enough. Understanding systems, business impact, and trade-offs is the new baseline. 2025 wasn’t the loudest year in tech — but it may have been one of the most honest. And for people who actually build, that’s a very good sign.
What We've Shared
DevOps Accents, episode 65: How Tech Hiring Really Works Now and Bits & Pretzels 2025
Explaining AI Explainability: Vision, Reality and Regulation. AI can feel magical, but when decisions affect health, justice, or safety, we need more than magic—we need explanations. Paul Larsen breaks down what “explainable AI” really means, why different stakeholders need different kinds of “why,” and how this series demystifies the promises, limits, and regulations of AI explainability.
What We've Discovered
Why Signal’s post-quantum makeover is an amazing engineering achievement: This almost made one of our co-founders re-consider using Signal. Easily digestable explanation of how Signal makes their protocol quantum-safe - and how hard it actually is. At least good to know that quantum-safe protection is possible!
Pipelining in psql (PostgreSQL 18): This new pipeline feature unlocks some real performance benefits for certain scripts - and who didn't have to script around psql to perform important data migrations?
Kubernetes Gateway API in action: Please tell us if you are using (and how) Gateway API and whether it replaced Ingress for you.
Why didn't ngrok go down in last week's AWS outage? "Just don't use us-east-1" - this! Even though some services will go down in other regions if use1 is down, you will be in a much bigger trouble if all of your infra is in this region. This region is a huge beast and you are better off choosing another one for your workloads.
AWS to Bare Metal Two Years Later: Answering Your Toughest Questions About Leaving AWS. Good to see more companies share their successful cloud repatriation journey, with real numbers on cost and availability included.
The 83rd mkdev dispatch will arrive on Friday, December 12th. See you next time!