How I Automated My Entire Email Workflow with Agent-S

Email is the tax you pay for having clients. I used to spend the first two hours of every morning just triaging my inbox — reading, sorting, replying, flagging things for follow-up. Two hours. Every single day.

Today that takes me about 15 minutes. Here’s exactly how I set it up.

The problem (in numbers)

Before automation, my typical morning looked like this:

~60 emails overnight (mix of client threads, newsletters, notifications, cold outreach)
25-30 need some kind of action (reply, forward, schedule something)
10-12 need a real, thoughtful reply from me
The rest are noise that needs archiving or a quick template response

That breakdown meant I was writing the same kinds of responses over and over. “Got it, I’ll have this by Friday.” “Here’s my calendar link.” “Let me look into that and get back to you.” Rote stuff that feels productive but isn’t.

The setup: Agent-S as my email brain

I use Agent-S because it can actually interact with my email client like a human would — opening messages, reading context, composing replies. It’s not just an API integration; it has a real browser session where it does the work. I talked more about why the “own computer” approach matters in a separate post, but the short version is: it can do anything I can do in a browser.

Here’s the system I built, step by step:

Step 1: Define the sorting rules

I gave my agent a set of plain-English rules for categorizing incoming email:

Priority 1 — Needs my brain:

Client emails with questions, decisions, or escalations
Anything with urgency signals (deadlines, “ASAP”, “blocker”)
New inbound leads / referrals
Anything involving money, contracts, or legal language

Priority 2 — Agent can draft:

Meeting scheduling requests
Status update requests
Simple confirmations or acknowledgments
Follow-up reminders

Priority 3 — Auto-handle:

Newsletter/marketing (archive)
Notifications from tools (archive unless error/alert)
Cold outreach (archive or template decline)

These aren’t programmatic filters. They’re instructions the agent interprets with judgment. If someone sends a “quick question” that’s actually complex, it routes to Priority 1. It reads the content, not just the subject line.

One thing I learned early: be specific about edge cases. I had a client who’d send casual emails that were actually urgent decisions buried in friendly language. I added a rule: “Anything from [client name] with a question mark goes to Priority 1 regardless of tone.” Small adjustments like that made a huge difference.

Step 2: Set up the morning routine

Every day at 7am (before I’m even awake), the agent runs through my inbox:

Reads every unread message
Categorizes into Priority 1, 2, or 3
For Priority 3: archives or applies template response (sent to drafts, not sent automatically)
For Priority 2: drafts a contextual reply based on my writing style and the thread history
For Priority 1: flags it, adds a one-line summary so I can scan quickly

The whole process takes 8-12 minutes depending on volume. By the time I sit down with coffee at 8:30, my inbox looks like this:

Flagged items: 8-12 messages that need my actual attention, each with a summary
Drafts folder: 10-15 pre-written replies waiting for my review
Everything else: already handled

The summaries alone are worth the setup. Instead of opening a 12-message thread to figure out the latest ask, I see: “Client asking for revised timeline on Phase 2 — wants to push deliverable by one week. Needs your OK.” I can respond in 30 seconds instead of 5 minutes.

Step 3: Teach it my voice

This was the part I was most skeptical about. I pointed the agent at my last 500 sent emails and told it to learn my patterns.

Things it picked up on its own:

I start emails with the person’s first name, never “Hi” or “Hey” for clients
I use em-dashes way too much (guilty)
I keep paragraphs short — rarely more than 2-3 sentences
I sign off with just my name, no “Best” or “Cheers”
For scheduling, I always offer 3 time slots

The first week, I edited about 40% of its drafts. By week two, that dropped to 15%. Now I’m approving 90%+ without changes. The ones I do edit are usually because I want to add a personal touch the agent couldn’t know about (like referencing something from a call that wasn’t in email).

Here’s a thing that surprised me: the agent caught a tone inconsistency I didn’t know I had. I was more formal with newer clients and casual with long-term ones. It replicated that automatically — picked up the pattern from my sent history without me ever stating the rule.

Step 4: Follow-up automation

This is the part that actually made me money. I set up a rule: if I send an email and don’t get a response within 3 business days, the agent drafts a follow-up.

Not an annoying “just bumping this” follow-up. A contextual one that adds value — referencing something specific from the original email, keeping it under 3 sentences, and adding a new piece of value when possible (a relevant link, a simplified next step, a concrete deadline).

Before this, I was terrible at follow-ups. Things would just die in my outbox because I forgot. Since setting this up, my response rate on proposals has gone up about 35%. On one deal alone, a follow-up the agent drafted led to a $4,200 project that I absolutely would have let slip.

I also set up a secondary rule: if there’s still no response after 5 more business days, draft a final “closing the loop” email. About 15% of my eventually-successful proposals come back on that second follow-up.

Step 5: End-of-day summary

At 6pm, the agent sends me a quick digest:

Emails that came in and were auto-handled
Drafts it wrote that I haven’t reviewed yet
Follow-ups that are due tomorrow
Any threads where it’s unsure what to do (rare, but happens)

This takes 2 minutes to scan and means nothing falls through the cracks. I later added a weekly Friday summary — email volume, response times, and open threads that need attention before the weekend.

The results

I’ve been running this system for about three months now. Here’s what changed:

Metric	Before	After
Morning email time	~2 hours	~15 minutes
Follow-up consistency	Maybe 50%	~95%
Response time to clients	4-8 hours	1-2 hours
Emails I personally write	~30/day	~10/day
Missed threads	2-3/week	Nearly zero

The time savings are nice, but the real win is consistency. I never drop a thread anymore. Every client gets a response within a reasonable window. Every proposal gets followed up on.

Three months later: what’s changed

The system has evolved quite a bit since I first set it up:

I let it auto-send Priority 3 responses. Template replies for cold outreach declines, newsletter auto-archives — I stopped reviewing those after week 3. That shaved another 5 minutes off my morning.

I added a “VIP” tier. My top 5 clients by revenue get special handling. Their emails always go to Priority 1, summaries are more detailed, and if they email after 6pm, I get a phone notification. Simple addition, big impact on client satisfaction.

The error rate has basically flatlined. In month one, I’d catch 3-4 miscategorized emails per week. By month three, it’s maybe 1 per week, and it’s always an edge case I haven’t written a rule for yet. When I catch one, I add the rule, and it doesn’t happen again.

Follow-up drafts got noticeably better. The agent seems to improve over time as it sees which drafts I approve without edits and which ones I rewrite. Month three follow-ups are significantly more natural than month one.

I stopped dreading Monday mornings. Monday used to mean 80+ emails and a 2.5-hour triage session. Now it’s the same 15 minutes as any other day. The agent handled everything over the weekend.

Things I’d do differently

Start with Priority 3 only. I tried to set up everything at once and got overwhelmed. Start by having it handle noise — archives, template responses. Get comfortable. Then add drafting. Then follow-ups. Give each stage a full week.

Don’t let it send without approval. At least not at first. Keep everything in drafts until you’re consistently approving without edits.

Write down your “never” rules. Things the agent should never respond to, never archive, never draft for. For me: anything from my accountant, anything legal, anything from my mom. Also: never auto-respond to someone who’s clearly upset or frustrated.

Why Agent-S specifically

I tried a few email automation tools before this. The problem with most of them is they’re either:

Simple rule-based filters (not smart enough)
AI-powered but read-only (can summarize but can’t take action)
Plugins that need you to stay in the tool to work

Agent-S works because it sits at the computer level. It opens my email the way I would, reads threads the way I would, and composes replies in the actual compose window. It doesn’t need API access or OAuth scopes for every individual service — it just uses the browser.

That means I can change email providers, switch tools, add new services, and nothing breaks. This email workflow was actually the first thing I transitioned from my human VA to AI.

Frequently Asked Questions

Can AI agents read my email?

Yes — but it depends on the type of agent. Most AI chatbots can’t access your email unless you copy-paste content into them. An AI agent like Agent-S that has its own computer and browser can open your email client in a real browser session and read messages the same way you would. It’s not accessing email through a hidden API — it’s literally opening Gmail or Outlook in Chrome. You control what it has access to, and you can see exactly what it’s doing.

How accurate are AI-generated email drafts?

In my experience, accuracy improved significantly over the first few weeks. Week one, I was editing about 40% of drafts. By week three, over 90% approval without changes. The key is giving the agent access to your sent email history so it can learn your patterns, and being specific about rules for different email types. The drafts that still need editing are usually ones requiring context the agent doesn’t have — like something discussed on a phone call.

Is it safe to let AI manage your inbox?

Start with everything going to drafts, not auto-sending. Review every draft for at least 2-3 weeks. Set up strict “never” rules for sensitive categories (legal, financial, personal). The agent operates on its own computer — not on your personal machine — so there’s a layer of separation. I can see everything it does and revoke access anytime. After three months, the only things I let auto-send are template declines and archive actions. Everything else still goes through me.

How long does it take to set up an AI email workflow?

The initial setup took about 2 hours — writing sorting rules, connecting the email account, letting the agent analyze my sent folder. But the real setup happens over 2-3 weeks as you refine rules based on what it gets wrong. Plan for 15-20 minutes of tweaking per day during ramp-up. By week 3, the system was basically self-maintaining.

If email is eating your mornings, try this. Start small, build trust, and scale up. Your inbox doesn’t deserve two hours of your best thinking time every day.