You know what keeps me up at night? Not the usual product manager worries about feature deadlines or user retention metrics. No, I’m lying awake thinking about how our entire industry is building AI systems with the permanence mindset of cathedral builders—when we should be thinking more like nomadic tribes.
Let me explain what I mean by 「embracing impermanence」 in AI. It’s not about building fragile systems or accepting failure as inevitable. Rather, it’s about designing AI solutions that acknowledge their own temporary nature—products that are built to evolve, adapt, and even gracefully dissolve when their purpose is served.
Remember when Google sunset Google Reader in 2013? The outrage was palpable. But here’s the uncomfortable truth: sometimes products need to die. In AI, this is even more critical. The models we train today will be obsolete tomorrow. The data pipelines we build will need complete overhauls in six months. The user behaviors we optimize for will shift beneath our feet.
I’ve been applying the principles from The Qgenius Golden Rules of Product Development to this problem, and it’s revealing some fascinating insights. The rule about 「starting from niche markets with strong user pain points」 becomes even more powerful when you acknowledge that both the niche and the pain points might be temporary. You’re not building for eternity—you’re solving today’s urgent problem while keeping one eye on the exit strategy.
Take the current generative AI explosion. How many companies are building elaborate infrastructure around models that will be outdated in twelve months? They’re pouring concrete foundations on shifting sand. Meanwhile, the smart teams are building modular systems where components can be swapped out as better alternatives emerge.
Here’s where it gets counterintuitive: embracing impermanence actually creates more durable value. When you design systems that expect change, you build in resilience. When you acknowledge that today’s breakthrough will be tomorrow’s baseline, you focus on creating workflows rather than fixed solutions.
I saw this beautifully executed by a startup building AI-powered customer support tools. Instead of betting everything on a single large language model, they created an abstraction layer that could route queries to different AI services based on cost, performance, and availability. When OpenAI’s API had outages, their system automatically shifted to alternatives. When new models emerged with better capabilities for specific tasks, they could integrate them without rebuilding their entire architecture.
The psychological barrier here is real. We’re trained to think that good engineering means building things that last. But in AI, the half-life of technical advantages is shrinking dramatically. What feels like solid engineering today might be technical debt tomorrow.
So what does this mean for product leaders? First, stop asking 「how long will this last?」 and start asking 「how easily can we replace this?」 Second, build measurement systems that constantly evaluate whether your AI solutions are still delivering disproportionate value. Third, create team cultures that celebrate sunsetting outdated approaches as much as launching new features.
The most innovative AI products I’ve seen recently aren’t the ones with the most advanced algorithms—they’re the ones designed with their own obsolescence in mind. They’re temporary solutions to permanent problems, and they’re honest about it.
After all, isn’t that what innovation really is? A series of temporary bridges across gaps in our understanding and capability. The bridge matters less than maintaining the capacity to build new bridges when the landscape changes.
So tell me—when was the last time you designed something specifically to be replaced?