Ex-Ramp Engineers Burn Six Months of Code to Build Melius: Inside the $25 Million Bet on AI Creative Production

Why Three FinTech Engineers Trashed Their Core Codebase to Build an Agentic Studio

Most founders treat their initial codebase like a family heirloom. Throwing away six months of software engineering feels like setting cash on fire. Yet that is exactly what Joowon Kim, Young Kim, and Arnav Ramu did when they realized their first product was solving the wrong problem.

The three founders cut their teeth at Ramp, the hyper-growth corporate card and financial automation company. Young Kim had served as Ramp’s director of engineering after the company acquired his previous startup, Venue, in 2023. Joowon Kim was a senior software engineer who spent his spare time making video content. Ramu brought investment banking and private equity rigor from Guggenheim and software investor Hg. When they teamed up, their natural instinct was to build financial and operational software for marketing: an automated performance tool designed to audit, track, and tweak digital ad budgets across paid channels.

Six months into development, customer feedback revealed a harsh truth. Marketing teams did not need another dashboard calculating cost-per-click or shifting budget between Google and Meta. The actual pain point was content starvation. Digital platforms burn through visual assets at an unsustainable pace. High-converting creative assets suffer audience fatigue within days. Managing ad spend was pointless if brands lacked fresh, high-quality images and video variants to feed the advertising machines.

Recognizing that the product lacked commercial viability, the team scrapped the entire codebase. Instead of optimizing media budgets, they set out to generate the creative assets themselves.

In July, the company emerged from stealth under the name Melius, pitching itself as an “agents lab for creative work”. Rather than forcing designers to juggle prompt engineering across half a dozen single-purpose generators, Melius coordinates autonomous AI agents across a shared digital canvas. The software ingests brand guidelines, orchestrates image and video generation, edits assets, and prepares final creative packages ready for distribution.

The market response was immediate. Within two months of its public debut, Melius passed $1 million in annualized recurring revenue (ARR). Backing that momentum, the company announced $25 million in total funding. The capital comprises a $5 million seed round led by General Catalyst and a $20 million Series A led by CRV, with participation from Genius Ventures, Vine Ventures, and Anti Fund.

This move underscores a major shift in enterprise software: companies are moving away from passive analytics dashboards and embracing generative tools that actually perform the labor.

---

The Economics of Automated Creative Production: Melius, Higgsfield, and Traditional Agencies

To understand why enterprise software investors poured $25 million into a newly launched startup, one must look at the economics of commercial advertising. For decades, producing a high-end commercial ad meant hiring an outside creative agency, contracting production crews, booking studio space, and spending weeks in post-production editing.

Generative AI shattered the early cost floors of production, but it introduced a new operational mess. Creative teams spent hours typing complex prompts into standalone image apps, upscaling faces in separate plugins, generating clips in standalone video platforms, and stitching files together in desktop video editors. The process remained fragmented, manual, and slow.

Melius entered the arena alongside explosive heavyweights like Higgsfield, which reached a $5.4 billion valuation in August after generating more than $700 million in annualized revenue in roughly three years. While Higgsfield focuses on high-speed consumer and direct-to-creator social video pipelines, Melius targets enterprise marketing departments, direct-to-consumer (DTC) brands, and ad agencies that require end-to-end agentic collaboration.

The table below contrasts the financial and operational mechanics across traditional creative agency retainers, fragmented prompt-by-prompt generation, and unified agent platforms.

Operational Metric	Legacy Creative Agency	Fragmented AI Tools	Melius Agent Lab
Typical Turnaround Time	3–6 weeks per campaign pack	2–4 days (manual stitching)	2–4 hours (end-to-end agents)
Average Cost per Finished Asset	$1,200 – $4,500	$45 – $120 (labor + API fees)	$6 – $18 (fully automated)
Monthly Creative Output Capacity	10–25 finished video/image variants	80–150 variants	500–2,000+ variants
Human Labor Focus	Storyboarding, shoot logistics, manual cutting	Prompt tweaking, upscaling, file transfers	Strategic art direction and final asset approval
Brand Consistency Mechanism	Strict manual brand book reviews	Hit-or-miss random seed drift	Persistent brand agent guidelines
Target Operating Model	Billed hours / Monthly retainer	Multi-seat tool sprawl	Centralized agentic creative OS

Note: Data points for legacy agencies and fragmented workflows reflect cross-industry research averages for mid-market DTC and consumer brands; Melius metrics reflect platform operating benchmarks and performance targets.

The data explains why Melius scaled past a $1 million run rate in less than 90 days. In enterprise marketing, performance drops the moment ad creative runs stale. When consumer brands run the same video hook for more than ten days, conversion rates routinely plunge by 30% to 50%.

Legacy production budgets cannot keep up with this demand. If a company needs 200 distinct creative variations every month to test different demographics, hooks, and aspect ratios, paying $2,000 per video costs $400,000 every month. An autonomous agent workspace that brings marginal asset production costs down to low double digits fundamentally rewrites that financial equation.

---

Operational Friction: How Melius Impacts Marketing Budgets, Lead Times, and Brand Stability

The deployment of autonomous creative agents causes immediate downstream effects for corporate marketing teams, media buyers, and design studios. Shifting creative production from human handoffs to software agents directly changes operating expenses, production lead times, and digital asset control.

1. Slashing Operating Expenses and Eliminating Tool Sprawl

The modern marketing technology stack suffers from severe licensing bloat. A standard digital marketing team often pays for:

• Stock photography libraries ($250–$500/month).
• Standalone AI image generation accounts across multiple creative seats ($30–$60/user/month).
• Video generation and avatar platforms ($80–$250/month).
• Voiceover and sound effects libraries ($40–$100/month).
• Specialized visual editing software ($35–$60/user/month).

Beyond the raw subscription cost, the largest expense is human coordination. Graphic designers and video editors spend up to 40% of their billable hours downloading files from one application, converting file formats, uploading them into another engine, and matching aspect ratios for different social channels.

Melius consolidates these tasks into an integrated agent pipeline. An art director provides a master campaign brief in plain English. Specialized AI agents split the work: one agent writes script variants, another generates brand-compliant visuals, a third lays out motion layers, and a fourth sequences audio tracks.

By removing manual file handoffs between separate tools, companies cut creative production overhead by an estimated 65% to 80% while reducing software licensing sprawl down to a single platform.

2. Collapsing Asset Delivery Lead Time from Weeks to Hours

Speed dictates digital marketing performance. When a cultural trend erupts on TikTok or a rival brand launches an aggressive promotional push, modern marketing departments must respond the same afternoon.

Under legacy workflows, launching a response campaign takes weeks:

• Days 1–3: Briefing meetings, creative reviews, and concept approvals.
• Days 4–10: Scripting, design production, and initial rendering.
• Days 11–15: Revisions, audio mixing, resizing for vertical and horizontal displays, and executive sign-off.

By the time the campaign launches, the market trend has passed.

Agent-driven platforms compress this timeline into hours. Because Melius operates on a real-time collaborative canvas, agents generate multiple visual concepts simultaneously. Creative directors review working iterations on the fly, instruct the agent to swap color palettes, alter pacing, or adjust typography, and export dozens of ready-to-publish variations in minutes. Marketing campaigns move from static monthly roadmaps to real-time, same-day deployments.

3. Preserving Brand Identity and Eliminating Asset Drift

The primary objection enterprise brands raise against generative AI is output inconsistency. Raw generative models are notoriously unpredictable:

• Facial features change between video clips.
• Hex color codes drift off-brand.
• Typography hallucinated by diffusion models creates unusable visual garbage.
• Logo placement and proportions fail legal compliance standards.

Melius solves this problem by enforcing a unified operating system where agents follow persistent brand rules. Rather than treating every prompt as a blank slate, the platform’s agents are pinned to locked design systems:

• Master font kits and typographic hierarchy rules.
• Exact corporate color libraries and vector assets.
• Pre-approved character references that retain visual continuity across scenes.
• Structured output templates that ensure correct safe zones for platform interfaces like Instagram Reels and YouTube Shorts.

This guardrail system prevents the random drift that makes unguided consumer AI models unsuitable for enterprise advertising.

---

The Broader AI Toolchain: Infrastructure Buffers and the Race for the Agentic Creative Stack

Melius does not operate in a vacuum. The generative media space is expanding at a dizzying pace. To survive against well-funded incumbents, the company relies on a layered technological architecture that isolates it from rapid underlying model obsolescence.

Foundational AI models for images, video, and audio are becoming commoditized commodities. When a new open-source video model drops on Hugging Face, or a proprietary lab updates its diffusion weights, single-feature wrapper startups face sudden obsolescence. If a startup’s entire value proposition is a clean UI around a single model API, any upgrade by larger competitors wipes out their business.

Melius protects itself through a workflow-first architecture:

• Dynamic Model Orchestration: Rather than tying its software to a single neural network, the platform serves as an orchestration layer. It routes image tasks, video rendering, and voice synthesis to whichever underlying engine delivers the lowest cost and highest visual fidelity.
• Dedicated Cloud GPU Execution: Generating high-resolution commercial video requires massive computational muscle. High-volume studios often pair custom agent workflows with scalable infrastructure on platforms like RunPod to run dedicated ComfyUI pipelines and serverless GPU clusters, keeping inference costs low during burst production runs.
• Collaborative Canvas Persistence: Melius functions as a multiplayer working canvas rather than a chat dialogue. Like Figma or Miro, multiple human editors and autonomous agents interact within the same document. This design creates deep organizational lock-in. Once an agency builds its client asset library, visual nodes, and agent behaviors on a canvas, moving to a competitor is difficult and costly.

While competitors like Higgsfield pursue massive consumer virality and scale, Melius focuses on the business workflow layer. As Joowon Kim noted, the digital advertising market is vast enough to sustain multiple billion-dollar winners. With global digital ad spending topping hundreds of billions of dollars annually, capturing even a fraction of production budgets represents a massive commercial prize.

---

Market Shakeout Scenarios: How Autonomous Creative Software Will Reshape the Agency Ecosystem

The rapid rise of Melius, backed by $25 million from Tier-1 venture firms like CRV and General Catalyst, confirms that automated creative production has crossed the threshold from experimental novelty to core enterprise infrastructure. Over the next 12 to 24 months, this technology will spark a major shakeout across advertising agencies, software vendors, and marketing teams.

The Margin Squeeze Facing Legacy Production Houses

Creative agencies that depend on selling billable production hours face an existential margin crisis:

• Client Budget Audits: Enterprise procurement teams are already auditing agency invoices. When brand executives see that an in-house team using agent labs can deliver 50 high-performing video variations in an afternoon, signing off on a $50,000 monthly retainer for a dozen assets becomes impossible to justify.
• The Death of Junior Production Tasks: Traditional agencies rely on junior designers and copywriters to handle manual resizing, formatting, and minor retouching. Because autonomous agents perform these mechanical tasks instantly, the operational foundation of conventional agencies will collapse unless firms fundamentally change how they charge clients.
• Commoditization of Baseline Assets: Generic visual content will drop in economic value toward zero. Agencies that fail to provide high-level brand strategy, original intellectual property, or deep cultural insight will find themselves competing in a brutal price war against software that charges pennies per asset.

Three Non-Negotiable Rules for Creative Winners in the Agent Era

To survive and dominate this transition, marketing leaders and production studios must anchor their operations around three strategic rules:

1. Shift Focus from Manual Execution to Strategic Direction: The competitive advantage in creative work is moving from mechanical skill—knowing how to manipulate complex desktop software—to prompt precision, taste, and brand vision. The best creative directors will not be the fastest photo retouchers; they will be the sharpest conductors of autonomous agent swarms.

2. Build Modular, Machine-Readable Brand Systems: Companies must stop treating brand books as static 100-page PDF documents sitting forgotten in a shared folder. To thrive in an automated production environment, brand rules must be converted into clear, machine-readable constraints: structured color palettes, vector asset repositories, negative-prompt guardrails, and persistent stylistic parameters that agents can ingest and follow reliably.

3. Treat Iteration Speed as the Primary Growth Engine: In algorithmic digital media, volume and testing speed determine performance. The winners of the next advertising cycle will not be brands that polish a single television commercial for three months. The victors will be companies that run hundreds of parallel creative tests every week, using autonomous platforms to instantly turn real-time performance data into new, high-converting creative assets.