Ray Fitzgerald has a theory about why your company's AI-generated social posts look schizophrenic.
It's not the prompts. It's not the model. The problem, he argues, is that brand—the visual vocabulary, the tonal guardrails, the ineffable sense of what makes something feel like yours—exists nowhere an AI agent can actually reach it. Your guidelines live in a PDF somewhere. Your color palette might be in Figma. Your voice? That's institutional memory, the kind that sits in a senior copywriter's head.
So when an AI tool spits out 50 Instagram captions in 90 seconds, half of them sound like they came from a different company entirely.
Fitzgerald's answer is Bloom, a San Francisco startup barely six months into public existence that's trying to turn brand itself into callable infrastructure. Not a tool. Not a platform. Infrastructure—something any AI agent, anywhere, can query the way it might query a database.
The pitch is straightforward, almost engineering-minded in its clarity: your brand becomes an API. Connect Bloom to your existing materials—websites, decks, Figma files, past campaigns—and it builds what Fitzgerald calls a "canonical brand system." Once that system exists, any agent generating email subject lines or video thumbnails can hit Bloom's endpoints and get back assets that actually look like they belong to you.
"Your brand becomes a working system that any agent, app, or platform can call," Bloom's launch page declared when the company emerged from Y Combinator's Spring 2026 batch. The product went live in December 2025. Since then, the company claims it's generated north of 175,000 assets—a self-reported figure as of June 2026.
Whether that's traction or just velocity is harder to say.
Infrastructure for Agents That Don't Know What 'On Brand' Means
The mechanics aren't exotic. Teams connect Bloom via REST API or through an MCP (Model Context Protocol) connector—the latter built for environments like Claude Desktop, Cursor, and VS Code. Bloom ingests whatever you feed it, constructs an internal model of your brand identity (palette, typography, voice, visual cadence), and exposes that as callable infrastructure.
Developers hit something like POST /images/generations with a brandSessionId parameter. Back comes an image that matches the visual DNA Bloom extracted earlier. Rate limits cap out at 120 requests per minute per API key. The credit system meters usage: a 2K image runs one credit, 4K costs two, variants scale from there.
Pricing is still in flux—early-stage startups tend to experiment—but as of mid-June, the "Scale" plan sat at $90 per month for 500 credits, with API and MCP access plus unlimited seats. Documentation references other tiers (Plus, Pro, Max), though details remain fluid.
The MCP connector is where things get interesting, or at least more ambitious. Anthropic's open Model Context Protocol lets agents call external systems directly from their conversational context. Bloom's developer guides walk through building "an AI marketing agent with on-brand creative" by plugging the MCP server into Claude or community tools like n8n. Live templates show integrations with Resend for on-brand email blasts and n8n for cross-platform social variants.
The roadmap teases what's next: Slack, Shopify, Notion, HubSpot, Webflow, Meta Ads. All listed as "coming soon."
A Founder Who's Built Agentic Systems Before
Fitzgerald's background offers some clues about why Bloom is designed the way it is. Before this, he was a founding engineer at Inscribe, a YC Summer 2018 company focused on fraud detection. There, he built the AI Agents platform—agentic infrastructure in a domain where autonomy has to coexist with tight compliance boundaries.
Brand consistency has a similar shape, if you squint. Creative freedom, yes. But within guardrails.
He also founded Lucid, an AI-native word processor, earlier in 2025 before pivoting to Bloom later that year. The timing aligns with a broader wave: YC's Winter and Spring 2026 batches tilted heavily toward agent tooling, and major infrastructure players—Nvidia, IBM, DigitalOcean—all launched agentic stacks in roughly the same window.
According to Bloom's careers page, the company raised $1 million in a pre-seed round that included Guillermo Rauch, Vercel's founder. That figure is self-reported and hasn't been independently verified. But if accurate, it's a modest war chest for infrastructure plays, which tend to burn through capital faster than application-layer startups.
The team remains small: three people according to the YC directory as of mid-June, though that count may already be outdated if hiring has picked up.
The Pain Point Is Real. The Market? Still Forming.
The problem Bloom is addressing exists, though it's newer than most enterprise pain points. Salesforce's State of Marketing 2026 report, published in February, found that 75% of marketers have adopted AI in some form. But 84% are still running what the report diplomatically calls "generic campaigns."
Translation: the AI works, but it doesn't sound like you.
McKinsey published a flurry of pieces this spring on agentic marketing workflows, noting that as agents mediate more customer touchpoints—email nurture sequences, social responses, ad creative variations—brand consistency becomes a coordination problem across surfaces the brand team doesn't directly control anymore.
Bloom's argument is that brand shouldn't be a compliance layer you bolt on after the fact. It should be infrastructure from the start: versioned, queryable, callable.
The competitive landscape splits cleanly. On one side, you have platforms like Canva (with its Brand Kit), Adobe Firefly (Generative Match and Custom Models), Typeface (an enterprise brand content platform), and Jasper (Brand Voice for copy). All enforce brand consistency, but they do it inside their own walled gardens. You use Canva's AI because Canva knows your brand. Bloom's bet is that the future is multi-agent, and agents need a neutral brand layer they can all call—not one locked inside Canva or Adobe.
On the other side are generic image generators like Midjourney or DALL·E, which have no brand memory at all. Ad platforms like AdCreative.ai or Omneky generate at scale, but they're purpose-built for performance ads.
Bloom sits in a gap: agent-native creative generation with persistent brand context, accessible via open protocols. Whether that gap is wide enough to support a standalone business remains an open question.
Traction, or Just Noise?
Bloom's December 2025 launch, by the company's self-reported account, hit 1.4 million views, pulled in 14,000 users, and generated over 50,000 images in the first week. An April LinkedIn post—cited in third-party coverage but not independently confirmed—claimed 40,000 users and month-over-month revenue growth.
Those numbers are unverified outside company statements. No customer logos are public. No enterprise deployments disclosed. For a three-person team six months in, that's not necessarily a red flag—plenty of early-stage companies keep customer names close. But it does mean the traction story is, for now, opaque.
Bloom's YC partner is Pete Koomen, a detail that likely matters more inside the Valley than outside it.
Protocol Risks and the Fragile Foundations of Agent Infrastructure
Bloom's reliance on MCP offers both leverage and exposure. Anthropic's Model Context Protocol is an open standard, and its ecosystem expanded quickly through early 2026. But security researchers flagged vulnerabilities in April and May—remote code execution vectors at the protocol level—that exposed MCP servers to potential takeover.
Those vulnerabilities don't implicate Bloom specifically. But they underscore a broader fragility: infrastructure that agents call at scale—whether for brand, data retrieval, or workflow orchestration—needs authentication, sandboxing, and rate limiting that's bulletproof. Bloom's API supports both API keys and OAuth, and the 120-request-per-minute ceiling suggests some thought went into guardrails. Still, as agent usage proliferates, those details shift from "nice to have" to "table stakes."
The agent economy is young enough that basic questions—how do you version a brand? how do you audit what an agent generated and why?—remain unanswered. Bloom is building in that uncertainty.
A Long Bet on Infrastructure Over Applications
Fitzgerald's thesis is clean: brand becomes a capability agents consume, not a set of rules humans enforce. If AI agents generate the bulk of marketing creative two years from now—and that's an "if," not a certainty—someone has to own the brand layer those agents call.
Bloom is betting that layer is infrastructure, not an application. And that it's callable by anything, not locked inside a single platform.
Whether the bet lands depends on how fast the agent economy matures and whether "brand as API" becomes a category companies are willing to pay for separately. At $90 per month for 500 credits, Bloom is pricing for velocity and volume, not enterprise budgets. The roadmap—integrations with HubSpot, Webflow, Meta Ads—looks like a classic land-and-expand strategy, where Bloom stitches itself into workflows teams already run.
For now, it's a small team with callable infrastructure and a strong conviction that the shift from "AI tool" to "AI agent" rewrites the rules of brand management.
The rest? Execution. And in a market this early, execution is everything.
