Superhuman charges $30/month for email when Gmail is free. That's always forced them to maintain a different quality bar from most products, and it shapes everything about how they build AI features too.

We invited Loïc Houssier, CTO at Superhuman Mail, to join us on Deployed to talk about what they're building and how they're approaching AI feature development. Full disclosure: I (Ian) have been a paying Superhuman customer for about five years and recommend it constantly. It's one of those rare software products I'd fight to keep using every day.

Loïc and I go back a while, so it was great to catch up on what he's building. He's one of the most fun and energized engineering leaders I've gotten to work with, and that energy comes through in the conversation.

We covered how Superhuman approaches the high-dimensional challenge of "AI for email," from auto-drafts to semantic search to personalized triage. It's an especially hard evals problem! We also got into how they're adopting coding agents internally, including their "quality week" practice that's changed how their engineering team works.

A few things stood out about his mindset and the way the Superhuman Mail team approaches building:

• "Make it work is not enough. You need to make it great." Things other SaaS companies might treat as non-urgent feedback, Superhuman treats like a critical bug. This is part of how they maintain the quality bar.

• Focusing on the hardest examples for their AI evals. They build evals starting from crazy internal queries - like their CEO Rahul trying to find what type of wood he discussed with a contractor three years ago, buried in thousands of other emails.

• Engineering quality week. The first week of every quarter, the whole engineering team focuses on bugs and personal workflow improvement. It's dedicated time to experiment with new AI tools and improve their setups.

• Removing all the blockers for AI tools. A 24-hour security approval SLA. Unlimited budgets for coding agent subscriptions. They didn't want to waste time on procurement when engineers should be experimenting.

Check out the full episode below or on Spotify, Apple Podcasts, and YouTube. Below we dig into each of these themes.

When Your Competition Is Free, Quality Means Something Different

Superhuman charges for email when Gmail and Outlook are free. Loïc frames what they're doing as building a luxury product, and luxury comes with high expectations. The perceived quality matters as much as the actual quality.

"Make it work, not enough. You need to make it great. What is considered a bug here - for most B2B SaaS companies, that would be just feedback. For us, it's a bug. Whether it's one pixel off - one pixel off! - it's a bug."

This mindset extends to their AI features. They have a hard rule: users should never look foolish because of something the AI did. If the AI causes you to send a bad email or miss something important, that's a "look foolish moment" - a P0 bug.

It's a useful forcing function for any teams thinking about building great products: what would it mean if your quality bar was set by luxury expectations rather than "good enough for SaaS"?

Focusing on the Hardest Examples for Evals

Eval tactics look very different in different use cases. For instance, doing evals on a known categorization task is relatively easy: was the answer right or not? Email is one of those domains where the dimensionality is massive, and evals get much harder. Every user has different needs, different volumes, different contexts. Loïc walked through some examples of how they approach building evals for such a high-dimensional space.

Their method: start with the hardest queries from power users, then abstract the dimensions that make those queries successful.

"One query that is pretty famous internally was Rahul trying to ask: 'I remember I was chatting with a contractor three years ago. What was the type of wood that I was looking for for my coffee table?' Rahul receives thousands of emails every day."

That single query surfaces multiple hard problems: a needle-in-a-haystack retrieval problem across a massive volume of emails, then understanding context about contractors, finding the specific email about wood types. From internal queries like this, they identify the dimensions that matter, then build synthetic queries to better cover the problem space.

They're careful about data too: eval seed data only comes from employee inboxes who give permission (with Rahul's founder inbox being particularly useful given its scale), and then they use synthetic data across the dimensions they've identified as important. No customer data unless explicitly given permission.

Engineering Quality Week

One practice that stood out: Superhuman runs a "quality week" at the start of every quarter. The first week after quarterly planning, when designs often aren't finalized and teams aren't fully ramped on new projects, they dedicate to bugs and personal AI workflow improvement.

"That's the week. We don't focus on projects. No timeline, no GA dates. Of course, if there's a fire we'll take care of it. But the number one priority is bugs. Let's focus on bugs."

But it's not just about fixing bugs. It's also dedicated time for engineers to step back and improve how they work - setting up their coding agents, trying new MCPs, improving their repos. When you're mid-project with a deadline as an engineer, you often don't feel like you have time to experiment with your workflow. Quality week creates that space for people.

"Most importantly, this is the perfect time for engineers to take a step back and think about how they work. When you're working on a project, you have a timeline, you have a customer waiting. You're in your flow, you don't want to break something that's working. This week is made for that."

The before and after has been "incredible" in Loïc's words - not just in bug counts, but in team mindset. Starting the quarter by fixing bugs sets the bar for the rest of the quarter.

Removing All the Blockers for AI Tools

Loïc shared how Superhuman has approached the recent wave of AI coding tools. First they focused on removing all the blockers.

"We removed all the blockers. From a procurement standpoint, security approval - 24 hour SLA. Budget? Don't care. You want a paid subscription? Get a paid subscription. We won't waste time there."

From there, they've evolved to a "guild model" - some engineers are always at the frontier trying new tools, then condensing what works into a golden path for everyone else. Not everyone wants to constantly try new things, and that's fine in Loïc's mind. The guild surfaces what works for specific languages, contexts, and use cases, and then others can borrow from that learning.

Also, each repo now has dedicated agent rules with what to check when thinking about design, common pitfalls to avoid, etc. They share AI wins in weekly meetings. It's not enforced, but it's a clear path for people who want to get things done without spending cognitive load on tooling decisions.

Game Design, Not Gamification

Superhuman's founder Rahul has a background in game design, and it shows in the product. But Loïc is careful to distinguish this from gamification.

"Gamification is very negative, almost toxic. Game design is borrowing principles from games - leveling, how to design a good level so that people feel good and have stuff along the way before the end."

The classic example in Superhuman: inbox zero feels like you beat a level. When you clear your inbox, you get a beautiful HD image that changes every day. It's a moment of completion and reward.

They also borrow the principle that games should never lag. When you turn right, you turn right. When you jump, you jump. Superhuman has a rule that everything in the app should be under 100 milliseconds, sometimes even 50 milliseconds. That's a strong architectural constraint, but it makes everything feel instantaneous.

"The worst thing that can happen to a player is the game lags. Because it's frustrating. Bring that to the product - we cannot be slow."

Building In AI Right Now Is Fun - Enjoy It

Loïc's parting advice captures his energy and the way he shows up as a leader:

"This is a freaking awesome moment to be alive as an engineer and tech leader. Tech is fun again. Lean in, support your team, and enjoy the path - we're living through something huge."

PS: Superhuman is hiring. Reach out to Loïc on LinkedIn if you're interested.

Check out the full episode on Spotify, Apple Podcasts, and YouTube and subscribe for future episodes.