I recently came across an article about a well-meaning but poorly thought-out infrastructure experiment in Malaysia. The idea? “Illuminated roads” — glow-in-the-dark highways designed to reduce electricity use while increasing safety. It sounded like a cool tech solution. Low energy, futuristic aesthetics, public benefit. What’s not to love?

Until it rained. The glowing material started to wear out. Maintenance became expensive. The glow barely lasted a few months. And eventually, the whole thing was abandoned.

This story, while about roads, mirrors something I’ve seen over and over again in tech: people obsessed with building, but not with running.

The “Run” Is the Real Test

In my years leading large platforms — whether it was launching Visible, scaling Amazon Rentals, building data platforms for 2.5 Billion users at Meta, or modernizing systems at LinkedIn — I’ve learned this one truth:

It’s not hard to ship something once. It’s hard to keep it running, reliable, and valuable over time.

That’s where the real leadership muscle is built.

Too many leaders (and teams) today — especially in the AI era — are caught in the trap of the “cool build.” They roll out some new LLM-powered feature, automate a process, launch a GenAI assistant… …but don’t pause to ask:

  • How do we support this in production?

  • What happens when the model degrades?

  • Can we retrain efficiently?

  • Is the infra cost sustainable at scale?

  • Is the product’s usefulness durable or just a novelty?

What Malaysia’s Roads Remind Us

The glowing road failed not because it was a bad idea — but because no one thought deeply about operational viability. Durability. Maintenance. TCO. Real-world friction.

The same applies to software platforms and AI systems. The things we build — recommendation engines, AI chatbots, prediction models, payment infra — all have a run cost that’s invisible at first but always arrives.

At Visible, I turned down flashy AI ops that looked good in pitch decks but didn’t scale with our cost structure. Instead, I doubled down on human-powered AI that actually reduced customer service costs by 73%, while staying maintainable and trustworthy.

In every new initiative, I walk the same line again — shipping fast but always balancing with platform sustainability. You can't scale something if you can’t operate it well.

The Leadership We Need

This is where seasoned leadership matters.

It’s easy to say yes to shiny tech. Harder to say: “Let’s test the run cost first.” Much harder to say: “We’re not building this yet — it’ll break us at scale.”

In my playbook, product and platform leaders must ask:

  • What’s the MTBF (mean time between failures) here?

  • How do we recover gracefully?

  • What’s the support cost per user 12 months from now?

  • What happens when scale hits?

These aren’t sexy questions. But they’re the ones that create resilient systems.

Final Thought

You don’t get praise for preventing a fire. But that’s exactly what great platform leaders do.

So before you celebrate the next big AI rollout or flashy launch, pause and ask:

Are we building illuminated roads that won’t survive the first rain? Or platforms that will stand the test of scale, time, and customers?

Choose wisely.