Howard here, founder of Alknoma.

We didn't start with a product. We started with an ANNOYANCE. Agents were impressive while you were talking to them, and then, as soon as you stopped typing, they effectively disappeared.

You'd put one in Slack and end up with another bot sitting in a channel waiting for someone to tell it what to do.

You don't prompt your teammates.

So the real problem was never, "How do we make the model smarter?" It was: how do you give a language model a spine? Something that lets it chain call after call, across days and months, reaching out to different people, all without a human babysitting it.

Sounds impossible if you know from experience just how complex chaining LLM calls can be. Three things turned out to matter, and we learned each one the painful way.

1. What a piece of work actually is.

Small enough that it can be validated, iterated, executed and closed 99.9% of the time.

Too big and the agent starts hallucinating. Too small and it loses all semantic meaning.

2. Resolving time should be first-class.

A lot of work in a team is essentially waiting.

Waiting isn't failure. It's the norm. And that happens to be one of the things LLM agents are worst at. Ayven reasons about when it should wake up next, EVERY SINGLE TIME, across thousands of runs a week.

It is about recognising a deadline, annual leave, or when a teammate has gone quiet. The async gap is hard to nail down. Ayven does that.

3. What is the base unit of a fact.

In a team, things change constantly. Someone changes their mind, a deadline moves, a customer replies, a meeting gets cancelled. An agent would treat information that was true two days ago as gospel. Change is the critical signal, and that was difficult to implement when teamwork is so unpredictable. Ayven makes sure the right things change with it.

The magic moment that wowed us, and made us decide to launch Ayven: we gave Ayven the agenda for September. Nothing else. The next thing we knew, the board was full, and accurate to our plans. It put together the pieces of what execution would look like.

Later, I asked: "What's stopping us from launching on Product Hunt?" Ayven then told me about every dependencies that I did not previously know. Nobody wrote that answer. It was just the work being captured in its truest, current state.

There is a lot more to Ayven that we haven't covered here.

Talk to us if you want this deployed in your own environment. Tell us what you want it connected to.

Ask me anything :)