Every few months I update my AI workshop slides. I have to… this stuff changes fast.
One of the slides I always revisit tracks a simple thing: the credits I burn across every agent I run in a month. Which ones are actually getting used?
And here’s what I’ve found, every single time: my email inbox manager is always in the top three. By a lot.
It processes emails for about 4 cents each. And it handles more of my inbox than I do.
Why Email Is the First Agent to Build
I hear a lot of people talk about the AI agents they want to build. Analysis tools. Research assistants. Competitive intelligence systems. Things that sound impressive.
But when I ask how often they’d actually use those tools, the answer is usually… once a week? Maybe once a month?
Here’s my filter for what to automate first: how often does this happen?
I call this 80-20 agent building. The idea is simple. Automate the tasks that recur every single day before you chase the flashy stuff. Daily frequency creates compounding ROI. A one-off automation saves you an hour once. A daily agent saves you 20 minutes every day for the next three years.
Email qualifies. By a lot.
Most people check their inbox multiple times a day. They get dozens or hundreds of emails. And a huge portion of those emails require roughly the same response every time. Confirmations. Scheduling threads. Quick questions. Acknowledgments.
That’s not your highest-value work. That’s just volume.
What the Agent Actually Does
My email agent connects to my inbox and monitors incoming messages. For any email where it has a reasonably high confidence about the right response, it drafts the reply and sends it on my behalf.
Not saves it to drafts. Sends it.
That part surprises people. When I mention this in workshops, there’s always a nervous reaction. “What if it says something wrong?”
Fair question. The answer is good configuration. You set clear rules for what the agent handles and what it escalates to you. Time-sensitive client messages? Escalate. Scheduling confirmations? Handle. Questions you get every week? Handle.
The goal isn’t 100% automation. It’s handling the easy 80% so you only deal with the 20% that actually needs your judgment.
The Client Who Went from an Hour to 10 Minutes
Last year I set up a similar system for a client. He was running four different ventures and drowning in email. We built an agent that pre-drafted responses for his most common email types.
His feedback after a few weeks: “When I open my inbox now, everything is already drafted. I just review and hit send.”
He went from spending about an hour on email every day to maybe 10 minutes. That’s 50 minutes back every single day. Without hiring anyone. Without changing any of his other systems.
Same thing happened with a consulting client named Amanda. She’s a tax professional and hit peak season with an inbox that was completely out of control. We set up an agent with the right filtering rules and routing logic. She told me afterward the volume reduction was “dramatic.” Said she couldn’t believe how much noise the agent was catching before it even reached her.
This is what a good email agent does. It doesn’t try to handle everything. It gets in front of the routine stuff and gives you back the cognitive space for the work that matters.
The 4 Cents Reality
People always react to the cost. Four cents per email sounds wild. But that’s actually what it costs when you run this through an AI system like Lindy.
And when you’re processing 30 or 40 emails a day, that’s a dollar or two. For getting your inbox under control. For reclaiming an hour.
The ROI on that trade is kind of absurd when you think about it.
How to Start
If you want to build something like this, here’s where I’d begin:
- Start with one inbox category. Don’t try to automate everything at once. Pick one type of email you get all the time. Scheduling requests work well. So do FAQ-style questions.
- Write out how you’d respond. Before you configure anything, write out 3-5 example responses. This becomes the reference material for your agent’s instructions.
- Run it in draft mode first. Most platforms let you have the agent draft without sending. Do this for a week. Review the drafts. Tune the instructions based on what it gets wrong.
- Expand from there. Once it’s handling one category well, add another.
The honest version: building this took me a few hours the first time. Now I’d do it in under an hour. And it’s the agent I’d give up last.
Your inbox is full of emails that don’t require your judgment. They just require someone to respond.
That someone doesn’t have to be you.
