In late 2023, we were deep in customer discovery — 60+ interviews trying to find a hair-on-fire problem. One thing was undeniable: every company wanted to use LLMs to improve productivity. Obvious, right?

The harder part was what came next.

The gap

Companies wanted custom AI — trained on their internal knowledge, embedded in their actual workflows. But getting there required deep integration work that most teams couldn’t justify. The ROI wasn’t proven, the technology was still moving fast, and the teams we were targeting (technical support, SRE, internal ops) sat inside Series A+ orgs with long procurement cycles.

We had interested practitioners but couldn’t get past the chicken-and-egg: nobody wanted to bet on an unproven solution, and we couldn’t prove it without someone betting on us.

The turning point

OpenAI DevDay in late 2023 changed everything. Custom GPTs and the Assistants API made it dramatically easier to build custom AI. But there was still a massive gap between “I built a custom GPT” and “my team actually uses it in Slack every day.”

We decided to close that gap. One week later, we shipped PlugBear — a no/low-code connector that let you plug any custom LLM into Slack, MS Teams, Discord, or HubSpot.

What happened next

We launched on Product Hunt, shared it on a few communities, and got immediate traction — something none of our previous experiments had achieved. Users started paying organically within weeks. No sales team, no outbound, no paid channels.

Over the following months, we learned what customers actually needed: not just a connector, but AI agents grounded in their knowledge bases that could handle real work — customer support, internal Q&A, operational tasks.

The product evolved. The name changed too — our Techstars batchmates unanimously told us “Runbear” sounded better than “PlugBear” as a brand. They were right.

Where we are now

Runbear connects custom AI agents to the tools teams already use. We’re not building another chatbot — we’re building agents that actually reduce the workload for support teams, ops teams, and engineering teams.

139 paying customers later, we’re still early. But the signal is clear: teams want AI that works where they work, not in another tab they’ll forget about.

If you’re building with LLMs and want to ship them into real team workflows, check out Runbear.