Nik Kale
Nik Kale
Building Intelligent Systems That Operate Themselves Securely
⌘K
Subscribe Sign in
Nik Kale
AI in Production

Dark Mode Isn’t a Theme: It’s a Survival Skill

Empathy isn’t just a design principle. For our users’ retinas, it turned out to be a survival skill.
Dark Mode Isn’t a Theme: It’s a Survival Skill

We thought dark mode was a nice-to-have. Network engineers violently disagreed.

When we launched our AI assistant into production, we were proud of the interface. Clean design, useful content, responsive layout. We’d tested thoroughly, on our monitors, during business hours, in our well-lit offices.

What we hadn’t tested was what happens when a support overlay renders at 2 AM in a dimly-lit Network Operations Center where engineers have been staring at dark-themed dashboards for hours.

The answer: it lit up NOCs around the world like a disco ball.

Picture it. An engineer is monitoring network health in a room designed for low light and sustained concentration. Every tool, every dashboard, every terminal is dark-themed. Then our assistant pops up, a blazing rectangle of white-backgrounded helpfulness, searing itself into their retinas and announcing its presence to everyone within a twenty-foot radius.

The feedback was immediate, colorful, and unanimously negative.

What we missed

The mistake wasn’t forgetting dark mode. The mistake was designing from our context instead of our users’ context.

We build software in offices with overhead lighting and adjustable monitors. Our users operate in environments where ambient light is controlled, where screen brightness matters for sustained attention, and where an unexpected flash of white UI is genuinely disruptive to work.

This isn’t about aesthetics. It’s about empathy. And empathy in product design means testing in the environments where your users actually work: not just the environments where your engineers develop.

The fix and the principle

We fixed the dark mode issue. But more importantly, we changed our process. Every new feature now gets tested under what we internally call “every shade of darkness”: not just light theme and dark theme, but actual simulated NOC conditions. Low ambient light. Extended session duration. Adjacent dark-themed tools.

The principle I took from this: dark mode isn’t a theme toggle. For network engineers working night shifts, it’s a work condition. Treating it as optional is treating your users’ working environment as optional.

If you’re building tools for operations teams, security analysts, or anyone who works in low-light environments: dark mode isn’t a backlog item. It’s a launch requirement. And if you haven’t tested your product at 2 AM in a dark room after staring at it for four hours, you haven’t tested it.

Empathy isn’t just a design principle. For our users’ retinas, it turned out to be a survival skill.

More like this

The npm Nightmare Just Repeated Itself in AI Agents. It’s Worse This Time.

The npm Nightmare Just Repeated Itself in AI Agents. It’s Worse This Time.

1,184 malicious skills infiltrated ClawHub, the plugin marketplace for the OpenClaw agent framework, reaching 20% of the ecosystem in weeks. Unlike npm packages, agent plugins run with full system permissions. The supply chain playbook broke because the blast radius changed.
What 170,000 Users Taught Me About AI Trust at Scale

What 170,000 Users Taught Me About AI Trust at Scale

What I learned over the next three years, serving 170,000 users across dozens of products, changed my understanding of what AI systems actually need to work.
What 3 AM War Rooms Taught Us About Designing Multi-Agent AI

What 3 AM War Rooms Taught Us About Designing Multi-Agent AI

We went from “trust me” AI to “show me” AI. And that distinction, it turns out, is what makes engineers willing to actually use the system instead of just running their own investigation in parallel.
From 50% to 95%: How We Taught AI to Read Relationships Instead of Documents

From 50% to 95%: How We Taught AI to Read Relationships Instead of Documents

We’d tuned embedding models. Our retrieval was pulling semantically relevant passages. The LLM was generating fluent, well-structured answers. And half of them were wrong.
Two Critical CVEs Just Blew Open the MCP Ecosystem, And Developers Were the Target

Two Critical CVEs Just Blew Open the MCP Ecosystem, And Developers Were the Target

Two critical vulnerabilities in MCP reference implementations confirm what the security community warned about: protocol-level design gaps become exploitable at scale. The CVEs aren't edge cases. They're structural consequences of shipping without security mandates.
The Feature Nobody Asked For That Customers Loved Most

The Feature Nobody Asked For That Customers Loved Most

If you’re building AI-powered experiences: don’t assume the most sophisticated feature will be the most valued one. Watch what users actually struggle with.