We spent about a year and a half building Omnim, an AI marketing system for small and mid-sized B2B companies. Today the same team builds ArtifactBridge. This post explains why we made that switch.

What we set out to build

Our thesis was very simple. AI should be able to do all of the marketing. We didn’t mean helping a marketer write a LinkedIn post a bit faster. We meant running the actual work: preparing a product launch, watching competitors and responding to them, and turning customer calls into case studies and proof that sales can use.

So we built a harness around the models. We had our own agent runtime, an orchestrator that split the work between specialised agents (a product analyst, a narrative writer, channel writers, a sales enablement writer), and workflows that chained those agents into repeatable multistep processes. A company would connect its context, and Omnim would run its marketing every week.

What we kept running into

The multistep workflows were brittle. Every step had to hand clean output to the next one, and when a single step missed, the whole run fell apart. We spent more of our time keeping the pipeline standing than improving the marketing it produced.

At the same time, the models kept getting better underneath us. When we gave a plain frontier model good company files, meaning the positioning, the ideal customer, the voice, past launches and what had worked before, it could do most of what we were building a specialised layer to do. The intelligence was not the hard part. The setup around it was.

What we realised

The frontier labs will win the AI game. They put more into their models than any startup can put into a specialised agent, and every new release moves the line of what a general model can do on its own.

That makes it hard to justify building a specialised agent for work whose underlying use cases are relatively simple, and marketing inherently is. Research, positioning, writing, distribution and measurement are not problems a model struggles to understand. What a model lacks is your company: what you sell, who you sell it to, how you talk and what you already decided.

If you have the right connections, the right system, the right prompts, the right files and everything set up, you can run marketing very well on any model you want. The AI is just that good. It doesn’t need a specialised agent, it needs a good system to run on.

I still see value in specialist tools that have real expertise, data and years of maintenance behind them. But for most of what a small company does in marketing, I don’t think the answer is another agent. It is the setup the agent works from.

So we decided to build the system

The question became: what does that system look like, and could we build it in a way that runs anything, not just marketing?

When we tried to run our own work this way, we found the problem we hadn’t planned for. A system like this works well for one person and their agent. It breaks as soon as a second person joins. Our marketing documents broke exactly like that: with two people’s agents working on the same strategy file, we had three versions by lunch and nobody was sure which one was real. Useful prompts stayed in one person’s chat, decisions lived in DMs, and every new colleague had to be briefed from scratch.

That is what ArtifactBridge is for. It is a shared workspace where people and their AI agents work from the same current knowledge. The company’s documents and decisions are available to every agent the team already uses. Workflows that work are saved as skills, so someone else’s agent can run them without rebuilding the process. Work happens in rooms where a colleague and their agent can join and see the goal, the context and what is already decided. Agents propose changes, and a person approves what becomes company knowledge.

Omnim started as a multi-agent workspace, and in a way we built exactly that in the end. The difference is that the agents are not ours, they are yours, whether that is Claude, ChatGPT, Codex or whatever you already use, and that is how it should be.

What happens to Omnim

Omnim as a product is retired, and omnim.ai now points here. I don’t see the time we spent on it as wasted. Building the agents ourselves is what made the collaboration problem concrete enough for us to choose it, and marketing is still one of the first things we set up with ArtifactBridge, including our own.

If you were using Omnim or talked to us about it, I would like to hear from you at support@omnim.ai.

If you are trying to make AI work across your team rather than only for yourself, that is what we are building now.

Kubo, cofounder