Inside Superhuman’s 2025 International Hackathon

Yearly, the Superhuman crew—encompassing crew members making merchandise Go, Grammarly, Coda, and Mail—comes collectively to attach and construct throughout our in-person hackathons. In 2025, we took issues to the subsequent stage by flying 491 crew members from our international hubs throughout engineering, product, design, and advertising groups to our San Francisco hub for the primary time since changing into Superhuman.

The power was unimaginable: 278 groups hacked for over two days to carry their concepts for brand new Superhuman product experiences to life. We additionally put AI entrance and middle, offering workers with assets on how you can finest use in style instruments like Claude Code, Cursor, and Databricks. On the third and last day, 113 groups offered their demos to a panel of judges and their teammates, with 5 initiatives finally taking residence prime honors.

We’ll spotlight a few of these hackathon initiatives, the challenges groups overcame to create a demo in two days, and what’s subsequent for them. Needless to say whereas these concepts are thrilling, they’re experimental ideas for inner inspiration, not accessible product choices (at the least, not but!).

Superhuman Command In all places

Considered one of Superhuman Mail’s core strengths is its Command Middle, which lets customers rapidly discover, manage, and act on emails through the use of keyboard shortcuts. The six-person crew behind Superhuman Command In all places (SCE) expanded this by letting customers use the identical Command Middle shortcut to entry all of Superhuman Mail and Grammarly’s options of their browser. Customers would be capable to entry all kinds of instruments, together with core performance comparable to setting reminders or snoozing Grammarly, in addition to activating AI brokers like Reader Reactions.

A snapshot of SCE in motion

Past particular person actions, SCE permits customers to finish complete workflows with out reaching for his or her mouse. As an example, customers can use SCE to launch Grammarly’s Proofreader agent and work by means of suggestions fully from the keyboard by typing “J” and “Ok” to maneuver between playing cards, “E” to just accept options, and “D” to dismiss them.

“Since customers primarily interact with Grammarly whereas typing, it’s usually extra intuitive for them to depend on the keyboard to assessment and proper their writing,” defined Mangat Rai Modi, a software program engineer who labored on this hackathon mission.

The crew constructed the mixing in just some hours, leaning closely on AI instruments to rise up to hurry. “Nobody on the crew was used to the code base, and I’m notably not well-versed in frontend growth,” Mangat shared. “We used AI to rapidly perceive the codebase and vibe-coded our method to the MVP. In any other case, it will have been not possible to ship a working characteristic in two working days.”

Trying forward, the crew sees their work pushing for a brand new design sample for interacting with AI brokers. “A keyboard-based person expertise has the potential to reshape the mindset of each designers and customers,” Mangat defined. “Trying forward, we are able to use this paradigm to design our AI brokers to be equally keyboard-friendly, which may also help customers be extra productive.”

Whiteboarding in Coda

Coda has helped 1000’s of groups collaborate successfully, however Lead Buyer Success Supervisor Thomas Baucom saved listening to requests for a simple method to whiteboard in Coda for brainstorming and diagramming. When the hackathon got here round, Thomas determined to deal with this hole through the use of Claude Code and Cursor to construct and launch a whiteboarding prototype.

With this characteristic, any Coda person might create a brand new whiteboard inside a Coda doc. From there, they might draw freely, add shapes, import photos, or take away and redo components as wanted. One standout characteristic is AI diagramming: customers describe the diagram they need in plain textual content, and the whiteboard generates an editable model that may be additional tweaked.

Screenshot of the Coda Whiteboard expertise

As a solo crew, Thomas leaned closely on Claude Code, which allowed him to ship rapidly. However having highly effective coding help created a brand new problem: understanding when to cease constructing and begin cleansing.

“With fashionable AI instruments, it’s extremely straightforward so as to add yet another interplay or polish go,” Thomas defined. “However whiteboarding is an area with robust person expectations, which suggests the product wanted to feel and look proper, not simply technically work.” That’s why Thomas prioritized UX refinement over characteristic growth: tightening interactions, dealing with edge circumstances, and making certain the core expertise felt easy and intuitive.

Thomas’s journey to constructing has been years within the making. “I began with no coding background and discovered largely by sitting subsequent to extremely gifted engineers, product managers, and designers, absorbing how they suppose and measure high quality,” Thomas shared. This osmotic studying, mixed with hands-on expertise by means of Coda Packs, step by step raised his technical abilities and confidence—main him to embrace his maker identification. “I’m genuinely proud to work at an organization that values rituals like hackathons and doesn’t gatekeep constructing behind job titles.”

Superhuman Listening

Product groups draw from a variety of helpful buyer suggestions and knowledge, however the knowledge is inherently messy. Insights aren’t structured and are scattered throughout totally different instruments, making it tougher for product groups to grasp and construct what customers need.

A pattern output of a abstract of how Superhuman Listening can summarize knowledge throughout varied sources to tell the product roadmap

Superhuman Listening would resolve this drawback by centralizing fragmented buyer suggestions knowledge right into a single supply of fact that might be simply queried. For instance, if a person had their product roadmap in Coda, they might import key insights from messy, unstructured knowledge throughout varied buyer suggestions instruments (comparable to Gong or Salesforce), demonstrating the impression of particular person options in a extra environment friendly, streamlined method. The Superhuman Listening hackathon crew used this basis to construct an agent that helps cross-functional companions (comparable to gross sales engineers) rapidly decide whether or not buyer suggestions is mirrored within the product roadmap, making it simpler to maintain prospects knowledgeable.

The mission was constructed by a cross-functional crew of 13 people from gross sales, knowledge science, engineering, product administration, and person analysis.

“Initially, we have been two groups, however we merged mid-hackathon once we realized that we might merge our two concepts into one and construct one thing extra helpful collectively,” knowledge scientist Stephen Kaiser defined. “It turned out nice. We labored in Slack and met just a few occasions to remain aligned and focus on a few of the gnarlier challenges. It actually helped that your entire crew stepped up and was very proactive about contributing.”

That collaboration was important, particularly given the info’s complexity and messiness. “We spent a variety of time within the hackathon simply untangling the info,” Stephen mirrored. “Whereas we made actual progress, we nonetheless haven’t totally reduce by means of the noise. As an example, we used LLM API calls to extract urgency and sentiment from Zendesk buyer help tickets, however realized too late that the majority tickets have been simply noise, not precise product suggestions.”

This studying helped the crew chart their subsequent transfer. “Going ahead, we wish to concentrate on really validating this modeling and making certain we are able to seize and report the appropriate indicators persistently sufficient.”

Inclusive Language agent

Writing shapes how we join at work, each with teammates and prospects. However even with the perfect intentions, bias or non-inclusive phrasing can slip by means of, affecting interpersonal relationships, broader buyer belief, worker engagement, and, finally, enterprise outcomes. “We see how a lot small phrase selections form tradition, belief, and alternative inside massive organizations, which might unintentionally and quietly reinforce bias,” stated linguist Jade Razzaghi. That perception led her and fellow linguist Katia Starenka to construct the Inclusive Language agent. “We needed to create a simple method for folks to be taught extra inclusive language patterns with out slowing them down or making them really feel judged,” she added.

The agent can analyze a bit of textual content, establish points with dangerous or biased language, clarify why it’s problematic by linking to authoritative on-line assets, and supply a transparent, inclusive rewrite that preserves the unique intent. Every flagged merchandise can also be assigned a severity stage, so customers can simply establish and deal with essentially the most crucial suggestions first.

A screenshot of what the Inclusive Language agent interface seems to be like

One of many largest challenges was delivering complete outcomes with out overwhelming customers. The crew tackled this on two fronts: refining the outcomes and bettering the interface. For the outcomes, they iterated on the immediate and added a few-shot instance to enhance accuracy. On the design facet, they added a facet panel that summarizes complete points, which customers can filter by class and severity. Customers can then drill all the way down to a particular problem to see detailed suggestions and prompt rewrites.

Trying forward, the crew is happy by the device’s potential to drive higher inclusivity throughout enterprise organizations. “Many firms have DEI tips, however there’s an actual hole between these rules on paper and what folks really kind once they’re shifting rapidly to get work performed,” Jade defined. “We see actual potential for the Inclusive Language agent to assist organizations resolve this drawback with out including friction.”

Construct with us

Though these initiatives are hypothetical, our annual hackathon by no means fails to depart us feeling extra impressed and enthusiastic about working at Superhuman. Within the phrases of CEO Shishir Mehrotra, “Hackathons are probably the greatest methods to carry groups nearer and provides folks area to get artistic.”

In the event you’re enthusiastic about constructing modern new issues for Superhuman, or when you have a cool thought for the subsequent Superhuman Hackathon, apply to hitch our crew! View our open roles right here.