What if the most valuable thing a founding engineer ships is the thing they delete?
Before going further, a disclosure that matters more than usual on this one. I went looking for primary sourcing on the story making the rounds — ex-Ramp engineers raising $20M for a platform called Melius after scrapping their first product — and I came up empty. What I found instead was a business and finance podcast feed, an aerospace earnings call transcript where an analyst named Scott Mikus from Melius Research asks a question about F-35 utilization, and a Reddit bot-check page telling me to prove I’m human. None of that confirms a funding round. None of it confirms a product pivot. So I’m not going to write you a fake scoop with invented numbers and quotes that sound like they came from a press release.
Worth flagging the name collision too, because it’s the kind of thing that trips up anyone building research automation. Melius Research is an established equity research shop whose analysts show up in earnings call transcripts constantly. If you’re scraping for mentions of a startup called Melius, your pipeline is going to drown in sell-side analyst Q&A. That alone is a useful lesson for the bot builders reading this site.
Why the pivot story keeps resonating anyway
The reason this headline travels — verified or not — is that it describes something almost every builder has lived. You ship v1. It works. Nobody wants it. And then you have to decide whether to defend it or bury it.
I’ve built enough bots to know which choice is harder. Defending v1 feels like diligence. You add integrations, you tune prompts, you write better docs, you convince yourself the problem is distribution. Burying it feels like failure, right up until the moment you realize how much faster you move without carrying it.
The signals that your first bot should die
Here’s what I watch for in my own projects, and what I’d watch for in anyone else’s:
- Usage is demo-shaped. People try it once, say something nice, and never return. Retention is the only honest metric for conversational products, and it’s brutally easy to ignore when your Discord is full of compliments.
- The prompt keeps growing. If every new edge case gets patched with another paragraph of instructions, you don’t have a product. You have a negotiation with a model, and the model is winning.
- You’re building around a capability instead of toward a job. “We have an agent that can do X” is not the same as “someone pays us to stop doing X themselves.” A lot of 2024-era agent demos died in this gap.
- Your own team doesn’t use it. Nothing damns an internal-tooling bot faster than the engineers who built it quietly going back to doing the task by hand.
What carries over when you scrap the product
Scrapping a product is not the same as scrapping the work. If you built a real bot, you probably have an evaluation use, a set of traces showing where your model fails, a retrieval layer you had to fix three times, and a hard-won sense of how your users actually phrase things. All of that survives a pivot. The pretty frontend and the clever positioning do not.
That’s why engineers who come out of a company known for operational rigor can credibly reset. Their advantage was never the specific product. It was the ability to tell quickly whether a thing works, which is a muscle you build by instrumenting your own systems and then being willing to read the numbers honestly.
The practical version for your own project
If you’re sitting on a bot you suspect is dead, give yourself a two-week test. Define one job, one user, one measurable outcome. Instrument it so you can see completions and abandonments, not just requests. Then stop adding features entirely and watch what happens. If the number doesn’t move with zero new development, the features were never the problem.
And if you want a solid version of this habit, keep a short written record of what you believed before you shipped and what you believe after. Pivot decisions get mythologized after the fact. Your notes won’t let you.
As for Melius, I’ll update this piece if sourcing I can actually stand behind shows up. Until then, treat the headline as a prompt rather than a fact — and treat your own first product with the same skepticism.
🕒 Published: