The Human Interface Layer
*Your agents are only as fast as you are*
The Human Interface Layer
Your agents are only as fast as you are
Agents are fast. You’re the bottleneck. Fix the interface.
I spend most of my day working with a team of AI agents. The models are not what slow me down. I am. Every time I look up a link, type when I could talk, or process email one at a time, I’m the constraint in my own system. Here are five things that make me faster so my agents aren’t waiting on me.
Superhuman
Dirty secret: it’s March 2026 and I still use an email client. If you do too, go get Superhuman.
It’s annoying for a couple of days because it doesn’t work like you’d expect. I still fall back to Gmail for searching old emails when I don’t have an agent nearby. But within a couple of weeks the keyboard shortcuts become muscle memory and you’re processing email at roughly twice the speed. The snooze and postpone features are the real win: they let you batch-process by type instead of context-switching on every message. That frees up both time and working memory for the things that actually matter.
TextExpander
I share my company URL, this Substack, my book link, and my LinkedIn dozens of times a week. I share booking links and common snippets for scheduling. I’ve used TextExpander for years and every time I catch myself looking something up more than twice in a week, I create a snippet and drill the abbreviation.
It sounds so small, but I find that it has a material impact on my pace. Micro-frictions compound.
Reclaim / Calendly / Cal.com
Multiple booking links for different classes of meetings. Plan your ideal week, schedule the meeting blocks, share availability. Managing ad hoc meetings becomes less friction than having an EA. The key is having distinct links: one for quick chats, one for deep dives, one for sales calls. Each link encodes context so you don’t have to negotiate it.
WisprFlow
Talk, don’t type. I’ve dictated 362,000 words in 33 weeks at an average of 115 words per minute. Once you’re working interactively with Opus-level agents, there is no way to type fast enough. Add an instruction to your agents to watch for homophones and you don’t even need to clean up the output. It’s good enough. Works on iPhone too.
If you took one thing from this list today, this is it. The bandwidth difference between typing and talking is the single biggest productivity unlock some people haven’t made yet.
MyApp
The one thing I don’t have yet. A personal iPhone app, built just for me. Trivial to build in the agentic era. Here’s the spec:
Context access. My mom’s doctor, my kids’ baseball schedules, the info I need three times a month and can never find. All in one place, structured for me.
Agent input. Voice messages to my agents from my phone. An article idea, a feature request, a GTM thought. Speak it, route it, forget about it until the next session.
Offline reading. A Chrome extension that pushes articles to my phone via a database, like Goodlinks but with agentic superpowers. I can highlight a paragraph and send it to Elise as a research brief or to Morgan as a content idea. The reading surface and the agent system are the same thing.
This is the missing piece. I have agents. I have tools. I don’t yet have the single surface that ties them together in my pocket. Building it is a weekend project. I just haven’t had the weekend yet.
What’s in your human interface layer?

