Senior Product Marketing Manager

The making of Superhuman Docs: how Superhuman Inc. designed AI views, a vibe coding tool for teamwork. by Jeremy Olson, Product Management Director I've been building productivity software for over 20 years, and creating AI views for the new Superhuman Docs has been the most fun I've ever had. The best shorthand I have for it is "a vibe coding tool for teamwork." It transforms the flat, boring data your team already has into whatever you can imagine. A table of dinner guests becomes an interactive drag-and-drop seating chart. A list of priorities becomes a multiplayer voting game in which everyone gets $100 to spend, and the bills fly across the screen as your team votes. Superhuman Inc. is currently testing AI views in private beta. Join the waitlist if you'd like to be one of the first to try it. The idea for this feature came about when Superhuman Inc. tried building AI tools for its team. Superhuman Inc. wanted something that, on the surface, seems straightforward: robust, interactive apps that could easily connect to its team's data. Superhuman Inc. quickly discovered that it was very hard to get both. Powerful vibe coding platforms can build full apps, but they require technical chops to connect to live team data. Similarly, the tools where its team's work already lives - AI chat, documents, and SaaS apps - can generate static artifacts and HTML blocks, but they're mostly point-in-time snapshots rather than collaborative apps. AI views bring the best of both worlds. It lets you create powerful, full-fledged apps that connect to your team's work, right in the collaborative doc you already use. Getting here wasn't easy. To build a vibe coding tool specifically for teamwork, Superhuman Inc. had to solve the four biggest problems with existing vibe coding tools. 1. Connected to your live data and work. Most vibe-coded internal apps never go multiplayer because connecting to your team's live, shared data across docs and SaaS tools is hard. Most knowledge workers won't even attempt it. AI views solve this by staying directly connected to a Superhuman Docs table. The sync is two-way: Drag a guest to a new seat, and the table updates. Or add a new guest to the table, and they are instantly waiting in the unassigned queue in the AI view. That connection also means one table can power many views. A bugs table can be a triage board, a swipe-to-prioritize tool, and a leaderboard, all reading from and writing back to the same source. And because AI views are built on Docs' integration layer, the data doesn't have to start in Docs. Sync data from Jira, Asana, or Salesforce and build your team a custom UX for all their SaaS tools. 2. AI native by default. One of my favorite parts of AI views is that you don't use AI only to build an interface - you can use AI to do work for you inside the interface once it's been created. First, there's no setup required. AI is natively built in. You don't have to sign up for API keys, set up .env files, or worry about terminal configs. It just works out of the box. Second, you can use AI within your app. Ask AI to automatically triage your bugs as they roll in. Build an AI chatbot for your team's meeting notes. Create a button that assigns seating for dinner guests based on free-form notes. Finally, AI design principles are built in. For example, the built-in AI knows when to give you an explicit button to trigger something versus when to just quietly enrich things in the background. It uses tables to cache results, so you don't need to regenerate them from scratch every time you load the page. It creates great AI-native software almost automatically. 3. Built for collaboration, and lives right where your team already works. An AI view isn't a solo tool. When a teammate changes the data, everyone else sees it instantly. A view also knows who's logged in, so the same table can power a "My tasks" view for each person and an "All tasks" view for the team. You can assign work, vote, and collaborate inside a view just like anywhere else in Superhuman Docs. Because AI views are built on tables, if someone wants a slightly different view than what you built, they can just ask for it and get their own separate view of the same underlying data. No more stale copies. And it all lives inside a doc. An AI view sits on the same surface as everything your team writes and reads, so interactive tools show up next to the thinking that explains them. Scroll through a single planning proposal and pass a draggable timeline, a filterable wall of user feedback, a roadmap rendered as mountain terrain, and a working budget calculator - each in line with the words that give it context. The doc stays in the container your team already knows how to share, comment on, and find. 4. As easy to share as a doc and just as secure. While it is getting easier and easier to vibe code a cool demo, deploying it and giving it access to enterprise data is another beast entirely. Most vibe coding platforms are completely open-ended - the agent pulls whatever frameworks and packages it needs from the internet to build your feature and allow apps to connect to any API out there. And because it's quite difficult to securely deploy and share your vibe-coded apps with just your teammates, Superhuman Inc. has already seen thousands of vibe-coded internal apps unwittingly leak highly sensitive corporate data onto the internet. Rather than hoping the AI picks safe frameworks and libraries from the internet, Superhuman Inc. has vetted and handpicked reliable ones. Rather than blindly hoping the AI uses the internet and APIs safely, AI views can access corporate data through its secure Packs platform, which has built-in observability and controls. And rather than having to vibe code secure authentication from scratch, you can give access to specific teammates simply by sharing the doc with them, automatically leveraging Superhuman's secure authentication infrastructure. The result: Sharing an AI view is as easy and secure as sharing a doc. Describe it. Build it. Have way too much fun doing it. That's AI views: connected to your team's work, AI-native, collaborative right inside where your team already works, and as easy and secure as sharing a doc. Powerful and safe-to-share have usually felt like opposite ends of a spectrum. Now you don't have to choose. I think about it this way: The long tail of things AI is actually useful for is going to get filled in by regular people building clever, specific tools nobody at a company would ever think to build - bottom-up, not top-down. AI views are designed for exactly that. You don't need a technical background to make it happen. You simply describe what you need. The phrase Superhuman Inc. has heard most from early testers is, "This is so much fun!" When it's so easy to build, iterate, and share them with your team, you can't help but feel the delight of creating whatever your imagination can dream up. Superhuman Inc. think you'll have a blast too, and can't wait to see what your team builds! Join the waitlist.

Superhuman in Australia: what it really takes to transform higher education with AI. I've had a version of the AI-in-higher-ed conversation in a lot of different rooms. Each conversation tends to follow familiar beats - the anxiety, the cautious optimism, the question underneath every question: are Superhuman doing this right? When I had the opportunity to travel with the Superhuman for Education team to Australia in June for the EDUtech conference, I was surprised to encounter a conversation that felt different. The institutions I spoke with in Australia had already made decisions about AI and were ready to do the work. Instead of asking if they were doing this right, they were asking what taking this seriously actually demands. Moving past the AI productivity conversation. In most regions, the conversation about AI in education - and in the workplace - still centers on efficiency. AI is seen as a tool that saves time, reduces friction, and helps students produce more. In Australia, that framing was fading in real time. What I heard instead - in sessions and, more honestly, in one-on-one conversations with educators - was something more fundamental. Generative AI has changed more than just what students use to get their work done. It's disrupted some of the core assumptions underpinning how Superhuman teach and assess. Now, institutions are asking what learning needs to look like in an AI-saturated world. Australia isn't waiting on assessment reform. Australia's higher education regulator has made it plain: detecting AI use in student work with certainty is no longer possible. Institutions are now facing the challenge of redesigning how they assess learning. When you can't reliably catch AI use after the fact, you have to rethink what you're actually trying to measure. What does authentic student work look like now? What does it mean to demonstrate that you've genuinely learned something? What I saw at EDUtech was that Australian universities are reckoning with those questions in real time. Students want to be able to show their work. I was surprised to encounter a lot of skepticism toward detection tools. Publicly, many speakers were against them, but when I talked with educators one-on-one, a different instinct kept surfacing. Students need a way to demonstrate that what they've produced is genuinely theirs. The instinct behind detection in this case isn't punitive. Students aren't necessarily trying to avoid getting caught. Rather, their credentials only mean something if they can be defended. This points to something I believe pretty deeply: the institutions that figure out how to make student work more visible - as a form of support, not policing - are going to be the ones students and employers trust most. The next phase of AI in higher ed. The conversations I had at EDUtech - and the sessions Superhuman ran with Indian River State College and Texas A&M University System - kept coming back to the same question: what does responsible AI adoption actually look like when you build it from the ground up for your institution and around your students? A few things Superhuman is seeing work: * Start with the partnership, not the product. The institutions that saw real results didn't start with a finished tool and then figure out how to use it. They built alongside Superhuman to ensure the tools deliver what students and faculty need. * Design for the moment before students write. If detection isn't a reliable answer, process visibility is. Understanding how a student engaged with an assignment from the start makes authentic work visible and defensible. * Let outcomes do the talking. When the AI foundation is right, the outcomes follow. At comparable institutions, Superhuman is seeing retention increase, completion rates improve, and academic integrity violations drop. * Find your champions in unexpected places. The people most ready to build something new aren't always the ones with the most senior titles. They're the ones who know standing still isn't an option and feel the urgency to take the next right step. What I saw in Australia was students and educators who care deeply about integrity and authentic work, who are genuinely grappling with what AI means for higher education, and who have decided what they stand for and are building from there. That's exactly where transformation starts.

Superhuman has launched a new auto-draft feature that identifies important emails and generates replies that sound less robotic. The email client uses a mixture of AI models from Anthropic and OpenAI to draft responses based on users' previous conversation tone, offering three variations to choose from. During testing, 40% of auto-generated drafts were sent within one day, with 60% requiring no manual editing. The feature learns from usage patterns to improve future responses. Users can personalise drafts by adding details about themselves and their role in settings. The feature aims to help users manage overflowing inboxes more efficiently. Superhuman was acquired by Grammarly last year and is now developing Superhuman Go, an assistant that works across platforms whilst maintaining context.