Jacob Bank’s Post

Video! We're pretty sure it's going to be important for at least two reasons: 1) Video is increasingly becoming the default medium for people to learn how to do things. Watching a video is almost always a better way to learn than reading a blog post. I certainly use video tutorials for everything from changing wiper blades to searing a steak. 2) For Relay.app in particular, video is a really natural way to showcase dynamic workflows in action that are really hard to describe in the abstract. And while that all makes sense, as of a month ago we really had no idea where to start. While we're still total novices, we've made a ton of progress, so I thought I'd share what we've done and what the impact has been. First, we've created three types of videos on our own: 1) (~5) 3-5 minute Product tutorials from Thijs van As to explain how to use key product features like Iterators, Paths, Webhooks, and our AI steps. 2) (~5) 3-10 minute use case demos from Thijs and myself to showcase how you can use Relay.app for common automations like automatically extracting information from invoices or sending a mail merge from Notion. 3) (~30) 1 minute quick tutorials for specific use cases like sending notifications for upcoming calendar events. Beyond those, we've worked with a handful of "guest creators" that have made some awesome videos. Some of these are hosted on our channel from folks like Alexander Noles, Demetri Panici, Spencer Tahil, and Kaitlin Pierce (if you want to make a guest video on our channel, please let me know!). And others are on the channels of automation experts like Matthias Frank, Zack Stevenson, and Dan Leeman (check those folks out if you haven't yet!) What's been the impact of all of this? In the last week 38% of our signups came from Youtube! And they're really high intent, with a meaningfully higher activation rate than signups from other channels. So lots more to do, but it's definitely been a successful month of experimentation.

If you're a heavy Notion user, I bet you have a database of people somewhere. They could be customers, clients, employees, or party guests. And I bet at some point, you'll need to email all of them. Here's a quick demo video of how you can use Relay.app to do a simple mail merge with a Notion database: https://lnkd.in/g-hG372Y Would love feedback from all of the Notion experts out there! Matthias Frank Eva Zivanovic Beatriz Estrella Alex Sherwood Madison Payton Roxana Rodriguez C. Demetri Panici

I've been trying to figure out a reliable way to determine which use cases can be fully automated and which need a human in the loop, and I might have something useful. In short, there are 3 key dimensions to consider for a given use case. 1. How high are the stakes? 2. How good at this are today's AI tools? 3. How much of a personal touch is needed? Here are a couple of examples: - Summarize the notes of a meeting and send them to the team over Slack. It's low stakes, AI is really good at it, and a personal touch doesn't add much --> fully automate it. - Send an onboarding email to a critical customer. It's high stakes, AI probably isn't great at it yet (likely lacking context and won't nail the tone), and a personal touch is very important --> Needs a human in the loop! We've already built a pretty good system in the product that lets users dial up or down how much human oversight, and the next step will be to help them figure out how best to approach it for their particular use case.

Agents! Everyone's talking about them, but... it's still totally unclear exactly how they'll be useful to most of us. Here's our quick take: 🤖 An AI Agent is a "bot" or entity that is equipped with certain capabilities (actions) that it may perform semi-autonomously so that it can achieve the goals that are set by its creator. It does not have to exist within a chatbot experience, and in fact, we believe many agents will be more practically useful when they're invoked as part of repeated workflows (e.g. "automatically file the relevant issues in Linear based on a new message in Front"). One of the major challenges in making agents actually useful is to strike the right balance between their flexibility/autonomy, and the predictability/control you need to trust the output. We've been working hard to achieve that with a combination of techniques: 1. A very clear articulation of what actions the agent can take and what data/context is required to take those actions. 2. A structured workflow that makes it very predictable when the agent will be invoked and what's going to happen after it's done. 3. A set of human-in-the-loop oversight features so that a person can check the agent's work as it goes and override if necessary. We're still in the very early days, but it seems to be working so far! If you're interested in learning more, we just published a comprehensive tutorial here: https://lnkd.in/g6B5CXwQ

Don't know how I missed this when Zack Stevenson first posted this, but this is a great demo of using AI to create tasks automatically using extraction from email. I remember we tried so hard to build this in Gmail from ~2015-2019 and now you can just build your own and it actually works!

Cool post! Another reason why human-in-the-loop steps are so important. Keep automations simpler and more robust.

Vertical solution or horizontal lego bricks? This is probably the toughest and most frequent strategic conversation that we have at Relay.app. Many, many smart people have told us that we should pick a more specific vertical or function and entirely focus on building a solution for that (e.g. focus on customer success and nail onboarding, QBRs, etc). And I can see why: - Easier to identify your ideal customer - Easier to create use-case specific messaging - Easier to know what features and integrations to prioritize There's a good reason why "niche down and focus" is some of the most often repeated startup advice, and there are lots of successful companies that build for a specific function or vertical from Workday to Zendesk. And we've tried at various times to do this with HR, CS, and other functions! But so far, we've found that we're much more successful when we're building a flexible toolbox that can be used for many different things. Part of the reason for this is our team's experience (e.g. working on Gmail) and we like these types of problems, but there are a few other reasons I like horizontal products: - Can be harder to grow initially, but the ceiling is really high! - More robust to function-specific downturns. e.g. we were working on employee onboarding when everyone stopped hiring. that was hard. If HR is having a tough year, we still have plenty of other users to serve. - Lower competition for underserved long-tail use cases that don't have a vertical solution. - Early users are tech savvy advocates that do cool things with the product and spread the word. On those last two points, I want to highlight a recent example. One of our users Spencer Tahil is a freelance marketing and growth ops expert, and he's built a couple of really awesome use cases in Relay.app that combine automation, AI, and a human-in-the-loop. 1. Automatically finding opportunities on Upwork (https://lnkd.in/gVmvP5s9) 2. Automatically onboarding customers (https://lnkd.in/gwsU3vXJ) There are many, many people that need use cases like this, and no vertical tool will ever be able to serve them. Plus, when you have an early user like Spencer generating ideas and content, it paves the way for all future users. So for now at least, flexible toolbox it is!