Stilla

If you've ever used traditional tools like Zapier or Make, then you know how rigid they can be. You have to know the trigger, the action, the fields, and the exact structure up front. Stilla stands out because you can use NLP to describe what you want the agent to do, set up which apps it can access, then decide whether access is read-only or read-write. After that, it handles the rest; you just describe the outcome you want and let it build more of the workflow from live context.

For our salespeople making CRM updates post-call, that means Stilla can propose updates based on what was discussed, compare that against what is already in the CRM, and then flag issues like the wrong contact name being attached. And that matters because it's a detail that other tools will usually miss.

Now, setting it up is a different beast entirely. This feature is easy to describe, but the interface isn't as simple. If you don't have a clear picture of what you want the agent to actually do, you'll spend a while trying to get something useful out of it.

A big reason why we'd initially compared Stilla to tools like Granola or Littlebird was because of its meeting capabilities. Obviously, we know now that it's so much more robust than that, but the fact still remains that it does have a great no-bot-required transcription and summary meeting feature.

With any new tool I'm testing, I don't always feel comfortable giving complete access to, say, my GitHub repo, for example. With Stilla, you can get specific. Read all of Slack but only write in one channel. Read-only access to some apps, read-write somewhere else. Disable GitHub entirely so no one on the team can even connect to it. That's the kind of thing that everyone's been waiting for!

I even had it suggest a code change, but didn't give it actual repo access, so it went in the totally wrong direction. And I can't fully blame it for that. That's on me not wanting to hand over the keys. It's really a catch-22 where, without access, it can't do a good job, but without trust, you won't give it access.

My suggestion is to start with lower-risk stuff and just be extra mindful when you're setting up Stilla's access. The more you have to lose, the more careful you need to be.

One of the strongest use cases from the founder was an internal support bot. A company created a Slack channel where employees could ask questions, and Stilla would search Notion docs and Drive policies for the answer. If it found conflicting information, it could ping the right stakeholder and help clean up the outdated doc.

The CRM workflow is probably one of the most practical use cases. After a sales or customer call, Stilla can propose CRM updates based on what was discussed and leave it to the human to approve. That is the right balance: let AI do the legwork, but keep a person in the loop.

I would trust this workflow earlier than code generation. Updating a CRM draft or flagging a correction feels like the right place to build confidence. It is practical, recurring, and easy to review.

That being said, we've recently switched to an AI CRM called Lightfield that automatically does all this.

The weekly PR changelog agent is a more technical example, but it shows what Stilla can do when it has enough access. It can scan merged PRs, separate customer-facing changes from internal infrastructure work, post a Slack summary, and help update the "What's New" section. That is exactly the kind of recurring task that sounds important and then quietly dies because no one owns it.

Intentionality is the key here. There are a ton of integrations and controls to set up. Slack, Linear, GitHub, Notion, email... You can connect pretty much anything, which is exactly why you shouldn't. If you start plugging things in just because they're there, it'll get noisy. The value comes from deciding what you actually want Stilla to touch.

A great way to test and "dip your toes in the water". Test Stilla on something simple, like scheduling a meeting, it can cross-check calendars, adjust time, and draft invites.

No matter where you start, it's your upfront intentionality that determines if everything else you layer on will work. The more you set clear boundaries, the more reliable it is, and the more workflows you can layer in.

Something small but honestly impressive is what happens as you're typing. While you're in your notes or in a chat, Stilla starts surfacing suggestions on the right-hand side in real time. It's more of a collaborative thing versus it being some sort of autocorrect that finishes your sentences like Gmail does.  

As you're typing, you'll see it just starts recommending things to you, so it's almost like you have this little AI agent that will also do kind of an autocorrect suggestion to the right-hand side.

The suggestions feel more contextual than they do mechanical, and because it's happening inline while you're already in the flow of work, it doesn't break anything.