They Scrapped the Entire Codebase: How Ex-Ramp Engineers Raised $20M After Their First Product Failed

The most honest moment in a startup's life is the one founders rarely share publicly: the decision to kill what you built and start over. On October 6, 2026, Melius announced a $20 million Series A led by CRV on top of a $5 million seed from General Catalyst, bringing total funding to $25 million. The headline is a funding round. The story underneath is a scorched-earth pivot by three ex-Ramp engineers who launched an AI product, watched it fail to gain traction, scrapped the entire codebase, and rebuilt something fundamentally different.

Founders Young Kim, Arnav Ramu, and CEO Jae Kim knew each other from engineering roles at Ramp, the corporate spending and finance platform that became one of fintech's standout growth stories. They left with deep technical skills, shared working chemistry, and a thesis about AI-powered performance marketing. Their first product attempted to optimize ad spend using machine learning — a logical bet in a market where every marketing team wants better ROI from paid channels.

It did not work. Not in the way that matters for a venture-backed startup.

"We Burned All of It"

Jae Kim's description of what happened next is blunt: "We scrapped the entire codebase; we burned all of it." Not a refactor. Not a pivot within the existing architecture. A complete destruction of the first product's technical foundation, followed by nearly a year of building something new from zero.

This is rarer than startup mythology suggests. Founders are trained to iterate, to pivot gradually, to preserve what works while changing direction. Full codebase destruction feels like admitting total failure. Investors often pressure founders to find salvageable elements rather than starting fresh. Technical debt advocates warn against throwing away working code.

But Kim and his co-founders concluded that their first product's architecture encoded assumptions that were wrong — not just features that needed adjustment but foundational design decisions that could not be incrementally fixed. The AI performance marketing tool they built was solving a problem that customers experienced differently than the founders hypothesized. The technical approach reflected that misunderstanding at every layer.

What They Built Instead

Nearly a year after the initial launch, Melius revealed its new platform: an "agents lab for creative work." Instead of optimizing ad spend algorithmically, the product focuses on generating ad campaigns, images, and videos using AI agents. The shift moves Melius from the analytics and optimization layer of marketing — where incumbents like Google, Meta, and dozens of startups already compete — to the creative production layer, where AI is genuinely transforming workflows.

The new product positions Melius as infrastructure for creative teams rather than a dashboard for media buyers. AI agents generate campaign concepts, produce visual assets, and iterate on creative variations — tasks that traditionally require agencies, designers, and copywriters working in sequence over days or weeks.

Whether this positioning succeeds remains to be proven. $25 million in funding buys time to find out, but it does not guarantee product-market fit. What is instructive for other founders is the decision process that led to the pivot, not the outcome of the second product.

Lessons From the Melius Pivot

Speed of recognition matters. The Melius founders did not spend years defending a failing product. They launched, assessed traction honestly, and made the kill decision while they still had runway and investor relationships intact. Founders who delay this recognition burn through capital, team morale, and investor patience on a product that will never work.

Codebase destruction is sometimes cheaper than codebase rescue. When foundational assumptions are wrong, incremental fixes compound technical debt without addressing the core problem. A year spent rebuilding from zero on correct assumptions can be faster than two years spent patching a wrong architecture. The Melius team's Ramp engineering background likely gave them confidence that they could rebuild quickly — a capability that less experienced teams might lack.

Shared founder history enables harder conversations. Kim, Young Kim, and Ramu worked together at Ramp. They had trust built through prior collaboration, which made it possible to have the conversation about scrapping everything without fracturing the team. Solo founders or newly formed teams without shared history often struggle with the interpersonal dynamics of admitting collective failure.

Investor relationships survived the pivot. CRV led the Series A and General Catalyst led the seed — both firms invested before and after the pivot. This suggests the founders communicated transparently with investors throughout the process rather than hiding poor traction until a crisis forced disclosure. Investors who understand why v1 failed are more likely to fund v2 than investors who discover problems during due diligence for a bridge round.

Domain expertise transfers even when products do not. The Melius team's Ramp experience was in fintech infrastructure, not creative marketing. But the engineering skills — building reliable systems, designing agent workflows, shipping product quickly — transferred directly. Founders often worry that pivoting means wasting their domain expertise. Melius demonstrates that engineering excellence and product judgment are portable even when industry focus changes.

The Broader Context: AI Startup Pivots in 2026

Melius is not alone in navigating the gap between AI ambition and market reality. October 2026 has seen a wave of AI startup funding — Hadrian raised $40 million for AI penetration testing, Avarra raised $17 million for AI sales training, Siena raised $17 million for customer experience agents, Ghost AI raised $11 million for local AI hardware. Each of these companies arrived at their current product through some combination of iteration, pivot, and market feedback.

The AI startup landscape in 2026 rewards speed and punishes attachment. Models improve quarterly. Customer expectations shift as AI capabilities become mainstream. A product that was novel twelve months ago may be a feature of a larger platform today. Founders who cannot kill underperforming products quickly will be outrun by competitors who can.

At the same time, the bar for AI startups has risen. "We use AI" is no longer a differentiator. Investors want to see specific workflows transformed, measurable customer outcomes, and technical moats that survive model improvements. Melius's pivot from generic performance marketing optimization to a specific creative agents lab reflects this pressure toward specificity.

What Founders Should Take Away

If you are a founder staring at a product that is not gaining traction, the Melius story offers permission and a framework.

First, measure honestly. Are you iterating on a product with genuine early signals — retained users, organic growth, customer requests for more — or are you interpreting noise as signal because the alternative is uncomfortable?

Second, distinguish between execution problems and thesis problems. Execution problems — slow shipping, poor UX, weak sales — can be fixed within the existing product. Thesis problems — wrong customer, wrong problem, wrong approach — require fundamental change that incremental iteration cannot deliver.

Third, if you need to rebuild, rebuild fast. The Melius team scrapped everything and shipped a new product within a year. Extended limbo — maintaining a failing product while slowly building a replacement — drains resources and confuses the team.

Fourth, communicate with investors early and often. The investors who funded your pivot are the ones you told the truth to before the pivot was necessary.

Fifth, trust your co-founder relationships. The hardest conversation in a startup is "this is not working." Teams that can have that conversation without blame survive. Teams that cannot usually do not.

Melius's $20 million Series A is a bet on the second product, not a reward for the first one's failure. But the failure — and the courage to burn the codebase and start over — is what makes the second bet credible. Investors funded founders who demonstrated they could recognize reality and respond to it, not founders who would have spent another year optimizing a product nobody wanted.

That is the messy part of being a founder. And it is the part that matters most.