Nobody owns support at a five-person company

Open any help desk's marketing site and count how fast you hit the word "agents". Then "queue", "SLA", "workload distribution", "shift".

Those words describe a real job at a real company. Somebody's whole week is support. There's a rota. Someone reports on first response time in a meeting.

Now count the people at your company. If the answer is under ten, none of that describes you.

What it actually looks like

Support is a thing that happens between other things. A founder answers a question at 11pm because it appeared. An engineer gets pulled off a branch because the message mentioned an error code. Whoever notices the notification first deals with it.

There's no queue. There's an inbox somebody happens to have open.

And it's spread everywhere. Some in a shared Gmail. Some in whoever's personal inbox got the email. Some in a Slack channel where a customer's question was pasted for a second opinion and then quietly died. Nobody knows what the response time is. Nobody knows what people ask most. Not because it isn't worth knowing — because measuring it would be a whole other job.

The tools built for this either assume a support team and drown you in configuration, or they're a shared inbox with no memory that makes you the routing layer.

The specific thing that wears people down

It isn't volume. Five people can handle the volume.

It's that most of it is the same question. Password reset. Where's my invoice. Does this integrate with X. How do I export. You've written the answer eleven times. You'll write it again, slightly worse, because you're annoyed and it's Thursday.

Buried in that repetition is the one that's actually urgent — the customer about to churn, the bug only they've hit — and it looks identical in the list until you open it.

That's the real cost. Not the hours. The attention tax of triaging a stream of mostly-identical messages to find the two that matter.

Which is the argument for an agent

Not because AI is having a moment. Because the repetitive part is genuinely, structurally automatable, and the important part genuinely isn't.

An agent grounded in your own docs handles "how do I export my data" perfectly, forever, at 3am, in a decent tone. That's not a hard problem and it doesn't need to be a person.

The angry customer with the edge case needs a person. It should get to one fast, with everything already attached.

The trick is the split, and being honest about which side a conversation belongs on. Which is why our agent escalates readily and tells you why — and why it starts in a mode where it drafts but never sends, so you can watch it be right for a couple of weeks before you trust it.

What we deliberately didn't build

No shift scheduling. No SLA policy engine with escalation matrices. No workload balancing across agent pools. No skills-based routing.

Those aren't missing. They're for companies that have a support department, and if you have one, there are excellent tools already — genuinely, go use them.

We built for the company where support is everyone's fourth priority and still somehow the thing that decides whether people stay.