We hired our own AI before we asked you to

Most products start with a pitch deck. Ours started with a staffing problem.

Last year, our forward-deployed engineers were generating more technical questions than our developers could answer without losing their afternoons. So we built someone to answer them. We called him Jeff. He read the docs, knew the codebase, and replied in Slack like any other engineer would.

That was supposed to be the whole job.

Jeff got promoted

The surprising part wasn't that Jeff could answer questions — it was how quickly the team stopped thinking of him as a tool. People thanked him. People argued with him. People @-mentioned him in threads he hadn't been invited to, and he kept up.

So we gave him more rope: a GitHub handle, a seat in Linear. He started triaging bugs, writing fixes and opening pull requests — reviewed by the team like anyone else's. One Tuesday, a customer reported a bug in a Slack channel at 09:14. By 09:23 Jeff had found it in the codebase, fixed it and opened a PR. By 09:55 the fix had shipped and he'd messaged the customer himself. Nobody asked him to do any of that. (I wrote up that morning in detail here.)

Then marketing wanted one

Separately, one of our marketers had a Claude Code project quietly doing real work. The problem was that it lived on one laptop, in one window, with one person. We wanted it to be multiplayer — to live in Slack where the team already talked, to connect securely to the tools she used, and to act on its own instead of waiting to be asked.

That became Jill — and today her successor Llena runs our marketing ops: competitor research, the content calendar, the comparison pages on this very website. The post you're reading sits in a calendar she maintains.

A tool became a team member

Here's the thing we couldn't unsee: Jeff and Jill weren't tools we opened and closed. They were colleagues — with handles, memory, and the run of the right systems. Once you frame it that way, the design decisions write themselves:

• A hire should arrive knowing the job. Nobody hires a senior engineer and then teaches them what a pull request is. So Llev ships pre-loaded with skills that already work — the onboarding is about your tools and your edge cases.
• A colleague doesn't wait to be asked. Half of Jeff's value was that he spotted the work himself. Proactive isn't a feature flag; it's the job description.
• Memory has to compound. Every conversation and every file makes the next answer better. A team member who forgets last Tuesday is just a chatbot with a name.
• Access demands accountability. Jeff has the access of a teammate, so he's held to the standard of one — approval gates on the risky steps, and every action on the record.

Why we're telling you this

Because the question we get asked most isn't "which model does he run on?" It's "does this actually work anywhere real?"

It works here. It has for a while. We hired the AI first, gave him the worst of our backlog, and only started selling him once we couldn't imagine the team without him.

If you'd like to see whether that holds up on your stack, meet Llev before you commit — or skip the call and start the 14-day free trial. He can be in your standup this afternoon.