Code

Nothing Is Impossible, It Just Takes Less Time

How agentic coding made a fuzzy growth idea into a working Harness translation POC in an afternoon, and why proving the future got a lot cheaper.

In a world where agentic coding is part of the everyday toolkit, future thinking and proof of concepts are a lot more straightforward than they used to be, especially when the request from product is still fuzzy and nobody is sure whether the idea even fits inside the system we have today. I have always liked saying that nothing is impossible in software, it just takes time to build. The part that has changed is that the time to find out whether something is doable has gotten a lot shorter once you can sit down with an agent and push through the ambiguity instead of waiting for a perfect brief.

I ran into a clean example of that recently on the growth team at ClickUp. We use a system called Harness for feature flags that can also carry configuration into the app, which is how we turn experiments on and off and how we pass the knobs that change what a user sees. One of the things we are always poking at is whether we can ship different copy through that config, so a string can change for a cohort without a full release cycle sitting between the idea and the test.

The catch was localization, and anyone who has shipped a product in more than one language already knows the limitation: when the browser is not in English, the English strings we usually hand-wave through a config do not get translated along with the rest of the product. That gap was a known blocker for the kind of future we wanted, where growth could experiment with messaging without painting non-English users into a corner, or waiting on a full translation pipeline before the first test could even run.

So I took it on myself to investigate quickly and build a POC with Claude that showed we could keep an English string in the Harness config and translate it on the fly inside the app. The point of the exercise was not a production-ready localization system so much as an honest answer to a product question that used to sit in the “someday, if we ever get the time” pile. With an agent in the loop, that answer showed up as working code instead of a speculative doc.

The limits are still worth naming, and I chose a Chromium-only path because that is where the free capability we needed currently lives. In a different traffic mix that would have been a non-starter, but most of our users are already on Chrome, and a lot of this work lives in experimentation anyway, so the constraint was acceptable for proving the idea. The POC did not pretend every browser was covered, and it did not need to, since the job was to show that the config-driven string path could survive a language switch without inventing a whole new translation workflow first.

What I keep coming back to is less the specific trick and more the unlock underneath it. Ambiguous product asks used to die in the gap between curiosity and calendar space, when proving the future meant carving out enough engineering time to learn whether the system could bend at all. Agentic coding compresses that gap, which means I can spend an afternoon finding out whether the idea has legs, share a demo instead of a theory, and let the team decide from something concrete.

I still believe nothing is impossible, it just takes time to build. The difference now is that the time to understand whether we should build it has gotten short enough that future thinking can keep up with the questions product is already asking.