Ray Fitzgerald has a habit of building systems that replace judgment calls with code. At Inscribe, a Y Combinator alum from the 2018 batch, he spent two years designing fraud-prevention agents that sift through financial documents—the kind of pattern-matching work once reserved for trained specialists. Now he's back with a different proposition: What if brand consistency, that eternally subjective battleground of design reviews and heated Slack threads, could be automated too?
His new company, Bloom, emerged from Y Combinator's Summer 2026 cohort with a straightforward bet. Marketing teams are already handing off more creative work to AI agents—copy generation, image creation, campaign assembly. But those agents don't know what your brand looks like. They don't understand voice guidelines, preferred color palettes, or the subtle distinction between "playful" and "unserious." Bloom's pitch: make brand identity queryable infrastructure that any agent can access programmatically.
"We think brand is becoming infrastructure: one canonical source of truth that any system can query," Fitzgerald wrote in the company's Y Combinator launch post this past June. It's a clinical way to describe what amounts to a fundamental shift in how companies might manage their public face—less art direction, more API calls.
Since going live in December 2025, Bloom claims to have generated north of 175,000 on-brand assets for customers—a company-stated figure without external validation. For a three-person startup, that's either a sign of serious traction or the kind of volume you get when automation is doing the heavy lifting. Probably both.
How It Actually Works
The mechanics are straightforward, if not particularly glamorous. Bloom ingests whatever brand materials a company already has—websites, Instagram feeds, Figma files, old pitch decks—and runs what the team calls "visual DNA analysis." The system extracts fonts, color schemes, visual patterns, tone markers. All of that gets consolidated into a single brand repository: aesthetic parameters, voice guidelines, reference imagery, product specs.
From there, customers can spin out images, video, copy, entire campaign suites. The platform handles multi-format generation (2K and 4K images at different credit costs), editing workflows, AI-powered resizing, semantic search across brand libraries. Standard creative tooling, in other words, but designed to be called by machines instead of manipulated by humans.
What distinguishes Bloom from template-driven alternatives—think Canva's Brand Kit or AdCreative.ai—is the infrastructure layer underneath. The company exposes its brand system through a REST API and supports the Model Context Protocol, a specification that lets agents query external tools. That means any AI system—ChatGPT, Claude, custom enterprise agents—can programmatically request brand context and generate assets without human intervention. The MCP endpoint connects to Claude Desktop, Cursor, VS Code, and a handful of other development environments.
It's worth pausing on that architecture. Bloom isn't really a design tool in the traditional sense. It's middleware for a workflow that barely exists yet but that a growing number of companies seem to believe is inevitable.
Agents First, Humans Second
Fitzgerald doesn't hide the product philosophy. "Designed to be used by agents, rather than humans," the documentation states plainly. "The primitives (API, MCP, and soon CLI) have excellent agent ergonomics."
The documentation itself reflects that priority. Bloom publishes an llms.txt file—a structured index optimized for large language models—alongside standard API references. Rate limits sit at 120 requests per minute per API key. Credits get consumed per asset: one for a 2K image, two for 4K.
The company's walkthrough guides read like instructions for autonomous marketing loops. One example scenario: an AI agent monitors ad performance, detects creative fatigue in real time, calls Bloom to generate replacement assets, then pushes them directly to advertising platforms. No human approval step. Another guide demonstrates generating an entire product launch suite—ads, social posts, email headers, hero images—in a single agent session.
Whether marketing teams actually want to operate this way remains an open question. But the infrastructure is assembling around them regardless.
In the first half of this year, enterprise software companies moved aggressively to build agent-native layers. Adobe announced a brand visibility solution in April. Skai, a marketing performance platform, shipped MCP support that same month. Greenhouse enabled MCP connections to its recruiting system in May. Bloom slots into that emerging stack as the creative and brand context service—assuming the stack itself becomes standard operating procedure.
Early Signals, Uncertain Scale
The December launch generated 15,000 signups and "millions of impressions," according to the company's Y Combinator Jobs page. The 175,000 assets figure suggests usage has continued beyond initial curiosity, though without knowing customer distribution or repeat rates, it's hard to gauge whether that represents a handful of power users or broad adoption.
As of June 2026, Bloom had raised approximately $1 million, according to its Y Combinator Jobs listing. Investors include Guillermo Rauch, founder of the deployment platform Vercel, and Nebular VC. The San Francisco team consists of Fitzgerald and two others, including a founding engineer—lean even by YC standards.
Pricing follows a straightforward credit model. The publicly listed Scale plan runs $90 monthly for 500 credits, with no per-seat fees. Higher-volume tiers exist, though the pricing structure has evolved since the company's initial launch. (A common growing pain for startups iterating toward product-market fit.)
The Infrastructure Gamble
Bloom's entire thesis assumes a near-future where brand consistency gets enforced not through design systems and human gatekeepers, but through API calls and programmatic generation. Brand becomes less a set of static guidelines locked in a PDF and more a live, queryable layer that any system—internal tools, external agents, third-party applications—can reference on demand.
There are signs a market is forming around this vision. Context Deck positions itself as "brand infrastructure and context engineering for AI." A separate company called ProjectBloom offers multi-brand kits with agent workflows. Deeplumen launched what it calls an "Agentic Page" in March to distribute brand content for AI discovery. Adobe's brand visibility solution, unveiled at Adobe Summit in April, aims to keep brand assets "visible, accurate and trusted across AI discovery surfaces."
Fitzgerald's edge, perhaps more than the founders expected, is that he's coming from the agent side. Before Bloom, he built Lucid, an AI-native word processor. Before that, two years at Inscribe building agents that parse financial documents to flag fraud patterns. He knows what agents need to consume brand as infrastructure because he's built the agents that would consume it—a rare symmetry in a space where most founders are either design people or infrastructure people, rarely both.
Still, the bet depends on adoption curves outside any single startup's control. How quickly do companies shift creative workflows to autonomous agents? How willing are marketing teams to cede brand governance to programmatic systems? Does the Model Context Protocol actually become the standard connective layer for agent tooling, or does it fragment into competing specifications?
Early ecosystem estimates—treated here with appropriate caution—suggest 97 million monthly MCP downloads and over 10,000 active servers, according to researchers tracking the space. Those figures come from secondary sources and should be read as directional rather than definitive. But the direction, at least, seems clear.
For now, Bloom has a product, some early traction, and a thesis that aligns with where parts of the software industry appear to be headed. The rest is execution in a market that's still being invented, which is about as much certainty as any early-stage founder gets.
