Dec 11, 2025 · 1h 11m · latent-space
The Future of Email: Superhuman CTO on Your Inbox As the Real AI Agent (Not ChatGPT) — Loïc Houssier
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Superhuman CTO Loïc Houssier discusses the engineering architecture behind AI-driven email, the paradigm shift toward multimodal voice interfaces, and how cross-application context powers the next generation of autonomous executive assistants.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 23.9% of the talking time here. How this is scored →
The hosts as informed peer 5.2 Guest teaching 4.7 Guest disagreement 1.6 The hosts pushing back 2.4
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 1:06:13 Pushback on developer compensation framing
Loïc firmly rejects the mainstream narrative that increased PR output from AI should result in direct compensation disputes, arguing instead from the Maslow hierarchy of shipping satisfaction.
Hardest push from the hosts ▶ 52:32 Swyx challenges assumption on knowledge graphsSwyx directly calls out Loïc's claim that knowledge graph databases aren't ready, asking if he has genuinely benchmarked them or is merely repeating conventional wisdom.
Biggest teaching moment ▶ 8:20 DocuSign's hidden hardware and compliance complexityLoïc breaks down why DocuSign requires thousands of engineers, detailing specialized physical security appliances with anti-theft key wipes and international regulatory standards.
The host holds their own ▶ 34:41 Price per trillion tokens insightSwyx demonstrates deep infrastructure economics expertise from his work at Cognition, distinguishing amateur per-million token pricing from enterprise-scale per-trillion capacity buying.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| The Evolution of Communication: Vocal Media Versus Writing | 3 | 3 | 1 | 1 | The episode opens with an informal intro covering Loïc's unique background in cryptography, offensive security, and nuclear submarines. The hosts engage with playful curiosity about his non-linear career path. | |
| DocuSign Acquisition, European Ecosystem, and Scaling Complexities | 4 | 7 | 1 | 2 | Swyx poses the common industry meme asking why DocuSign needs thousands of engineers. Loïc thoroughly educates the hosts on international compliance, on-prem hardware appliances with physical tamper-wiping, and localized workflows like Japanese Hanko stamps. | |
| Superhuman AI Architecture and Productivity-Driven Inbox Features | 4 | 5 | 1 | 1 | Loïc details Superhuman's AI features, emphasizing productivity gains and solving agent laziness rather than adding decorative AI features. The hosts relate their personal workflow experiences. | |
| Agentic Frameworks, Model Selection, and Deep Quality Evaluations | 5 | 5 | 1 | 1 | Alessio and Swyx explore model selection, routing, and evaluation frameworks. Loïc discusses multi-model routing across Sonnet, Gemini, and OpenAI, alongside canonical benchmark queries. | |
| Dynamic Code Execution, Local Architecture, and Offline AI | 6 | 4 | 1 | 2 | The conversation moves into local caching, offline semantic search, and dynamic code generation with Anthropic for aggregations like Waymo receipts. Swyx probes the server-side versus local architecture tradeoffs. | |
| Inference Infrastructure, Cost Economics, and Conversational Interfaces | 7 | 4 | 1 | 3 | Swyx shares high-volume infrastructure pricing perspectives from Cognition, contrasting token-based pricing with fixed-box models like BaseTen. Loïc explores the future transition from written to spoken communication. | |
| Executive Assistant Automation and Grammarly-Superhuman Context Convergence | 6 | 5 | 1 | 2 | The discussion covers automating executive assistant tasks and leveraging multi-app context from Grammarly and Coda integrations to surpass standalone chat assistants. | |
| Future of Desktop Browsers, Knowledge Graphs, and Competition | 7 | 4 | 3 | 6 | Swyx actively challenges Loïc on whether browsers will disappear and directly presses him on whether Superhuman has actually tested dedicated knowledge graph databases before dismissing them. | |
| Engineering Velocity, AI Coding Tools, and Developer Future | 5 | 5 | 4 | 4 | Loïc details Superhuman's internal adoption of AI coding tools and measurable PR velocity improvements. When Alessio presents the common pushback regarding developer compensation versus output, Loïc directly pushes back, arguing for intrinsic developer satisfaction. |
Statements from this episode (22)
DocuSign's French acquisition required a six-month defense contract carve-out
“And we were working for the Department of Defense in France. So we had the Ministry of Finance in France basically saying, no way, no go, you cannot sell. So we had to do a carve out which is like the funniest acquisition type you can do. So you have your team…”
Houssier: European digital signature compliance requires completely separate software stacks
“It sounds crazy, but like you want to nail Europe. You need a different product. You need a different team to your local data centers because of the compliance. You cannot just run your data centers from the U.S. So you need a local team there. Oh, and by the …”
DocuSign built physical on-prem appliances that wipe cryptographic keys if shaken
“There's a DocuSign box. Acquisition made in Tel Aviv at the time. Wonderful people building like security appliance and where like you shake the box. The keys disappear. Like if someone is like stealing your box, no one can sign in your name.”
Houssier: AI agents that hand trivial choices back to users waste time
“And I don't need, typically, the agent to say, hey, I found this lot, and this lot, which one do you prefer? I just asked for 15 minutes. Find it. Do it. I have an admin, when I was asking her, like, on Slack, find me 15 minutes. She's not asking me if I need,…”
Superhuman CTO: Ask AI uses agentic paginated search across email batches
“So we had to implement the pagination search. So like semantic search for like the first, I would say, 40 and deep search. Not that one. Okay. Next 40, next 40. So I'm kind of like using this agentic loop, and while you don't have find the answer, continue and…”
Claude Sonnet outperforms OpenAI on agent handoffs and reducing laziness
“Sonnet was really great for, like agent head off. Like the laziness was really great. OpenAI version of it was not that good.”
Houssier: Switching between agentic frameworks will become increasingly difficult
“It's still easy to switch from one hatcheting framework to the other. But at some point I think it will be harder and harder and like the stickiness of them will be tricky.”
Superhuman builds dynamic on-the-fly aggregation lambdas with Anthropic
“We're working right now with Anthropic to basically do kind of like a building on the fly, small, kind of a key component of lambdas that will build the code to do the aggregation.”
Superhuman uses Turbopuffer on the backend to store search vector embeddings
“We use a Turbo Puffer on the back end to store like .”
Superhuman uses Baseten to run LLaMA and BERT classification models
“We use Base-Ten to run some I would say some LAMA, some BERT model for classification.”
Apple's local on-device iOS AI capabilities have been underwhelming so far
“IOS has done some work there, but it was underwhelming so far.”
Houssier: Superhuman starts with top models before optimizing inference cost
“We would always start with the highest and more expensive model. To get the right quality. And when the quality is nailed, then we can spend time trying to optimize.”
Pricing AI inference per million tokens is amateur thinking at enterprise scale
“A lot of people think about cost in terms of dollars per million tokens, right? And I think that that is actually amateur thinking. This only, only the kind of pricing you care about if you're a solo developer, but once you're in a large scale, like you guys a…”
Houssier: Superhuman testing beta auto-drafting meeting availability in emails
“Right now it's in beta. But right now my emails, like internally, when someone is asking, hello, can we meet next week for lunch? Automatically I will have like three slots proposed in the draft and I can just like send the draft that is prepared for me.”
Houssier: Human EAs will always beat LLMs on interpersonal context
“And this is where your EA will always be better than an LLM because she knows the type of people you are okay to have lunch with, or maybe they have the context because, ah.”
Houssier: Grammarly captures cross-app context that ChatGPT cannot access
“Grammarly knows, basically, where you, I say, you're switching from Google Doc to Salesforce. To LinkedIn, and now you're writing an email. So we have this augmented context even more, so like much more precise compared to something like ChatGPT, for example. …”
Web browsers will become thinner and eventually embed directly into the OS
“I think the browser would be like more and more thin. I believe they would be like thinner and thinner, but they will disappear or they would be like just embedded in the OS eventually.”
Superhuman views OpenAI and ChatGPT as direct competitors in productivity
“Oh, the answer is like ChatGPT, or like OpenAI and superhuman are competitors. Like this is what we fight against to some extent.”
Superhuman engineers grew pull request throughput from four to six per week
“In Q one, we were roughly set at four PR per engineer per week. In Q one, in Q two, we were closer to five PR per engineer per week, and in Q three, we're closer to six. So the global throughput, and again, PR per engineer per week, we can debate, but that's a…”
Superhuman operates with 50 engineers and roughly 100,000 paying users
“Superhuman may, like, we have 50. 50 engineers. And your user base is roughly a million? Yeah, like, less than that. Like paying users probably 100,000.”
AI will widen the performance gap between good and lazy engineers
“I do believe that AI will do only one thing. It will Separate faster the good engineers from the bad engineers. If you're a good engineer, and you're using AI well, you will be an amazing engineer. If you're a poor, lazy engineer, and you don't want to underst…”
Pure technical moats are gone; UX is now the only differentiator
“In the world where the technical moat is not that a moat anymore, because Like startups in two weeks, they can build something that is close to what you're building. The difference is like the, how you think about the user or the flow and all of that.”