I have a friend who spent most of last year trying to reach inbox zero.
She'd clear everything down to a dozen emails and feel genuinely good about herself. Two days later, she'd be back at 300 and the stress would return. She'd go on another clearing sprint. Repeat.
After watching this cycle a few times, I told her she was measuring the wrong thing.
Inbox zero made sense in a world where email was just communication. Messages came in, you responded, you cleared the queue. The inbox was a task list, and empty meant done.
But that model is outdated. And if you're building AI agents, inbox zero can actually work against you.
Email as a Knowledge Store
Here's what changed my thinking.
I was at Arena Hall in Austin, working with Evan Baehr to design his super agent. We kept routing AI outputs to email — daily memos, meeting summaries, agent-generated reports — and at some point Evan asked why.
The reason is straightforward: when his super agent preps him for a meeting, it searches his inbox. It looks for every prior mention of the person he's meeting, every relevant summary, every note that's accumulated. It doesn't query a separate database or a specially formatted document. It searches email. The inbox becomes the context layer.
The more you route to email on purpose, the richer that context gets. Every memo that lands there is one more piece the agent can find later.
That's the shift. Email isn't just a place to receive messages. When you design it that way, it becomes a searchable knowledge store.
What to Route There
Not everything. The goal isn't to fill your inbox with noise. The goal is to store AI outputs intentionally so they're queryable later.
Things worth routing to email:
- Daily briefings and memos your AI generates
- Meeting summaries and key decisions from calls
- Weekly synthesis reports (what patterns emerged, what action items surfaced)
- Agent reports on topics you track (news, research, competitor updates)
These aren't emails you need to respond to. They're documents that accumulate over time. When your agent queries the inbox for context, these are exactly what it needs.
The Framework Behind It
What Evan and I were building that day is part of a broader design principle I use with AI: context files as assets.
The idea is that context — your identity, your preferences, your history — is one of the most valuable things you can give an AI agent. Most people think about this in terms of prompt files or uploaded documents. But email is already a natural container for accumulated context. You just have to start treating it that way.
The inbox zero crowd sees the inbox as a task manager and measures it by how empty it is. The knowledge-store approach sees it as a living archive and measures it by how useful the agent finds it.
Both are valid mental models. But if you're building agents, only one of them makes your system smarter over time.
The Mindset Shift
I'm not saying you should let spam pile up or never archive emails. Triage still matters. But the goal you're optimizing for changes.
Old goal: process everything, stay at zero, feel in control.
New goal: route AI outputs intentionally, let them accumulate, give your agent more to work with.
There's a kind of patience required here that goes against every productivity impulse. We've been trained to clear the inbox. Anything sitting unread feels like a failure.
But imagine your email as a memory layer for your AI stack. Every summary that passes through it is stored. Searchable. Available the next time your agent needs it for context.
You're not managing email. You're building a brain.
Try This
If you're running AI agents and you haven't done this yet, start with one output.
Pick one thing your AI generates regularly — a daily briefing, a meeting summary, whatever it is — and route it to your inbox. Don't archive it. Don't file it away somewhere. Leave it in email.
Do that for 30 days and then ask your agent to pull context from your inbox for something. See what it finds.
You'll understand the value pretty quickly.
Thanh Pham is the founder of Asian Efficiency. If you want to start building your own AI agent stack, check out the 4-Day AI Sprint or the 25X Productivity System.
