I still use email. I simply no longer use my inbox as the place where I decide what deserves my attention.
That distinction matters.
Email was designed to help people communicate. Today, it has become the default destination for almost everything: real conversations, automated notifications, system alerts, newsletters, sales sequences, invoices, calendar updates, contact-form submissions, security warnings and endless internal threads.
The important messages are still there. The problem is that they are surrounded by so much noise that finding them has become a task of its own.
So I stopped trying to manage my inbox better. Instead, I created an AI routine that decides which emails deserve my attention and delivers them to me somewhere I am more likely to see them.
Why I stopped reading my inbox
The first reason is volume. Most inboxes contain far more messages than a person can reasonably evaluate. Even if you scan only the sender and subject line, you are still spending time making hundreds of tiny decisions:
Is this important?
Does this require an answer?
Is this a notification?
Can this wait?
Is this person already connected to one of our projects?
The second problem is long email threads. An important conversation may contain 15 or 20 messages, several recipients and multiple decisions. Reading the latest email is often not enough. You need to understand the entire history before you can respond properly.
What I need in that situation is not another notification saying that someone replied. I need a summary of what has happened, what has been agreed and what is expected from me.
Spam is the obvious third problem, but traditional spam is only part of it. Modern inbox noise also includes messages that are technically legitimate but practically irrelevant. Cold outreach, automated follow-ups, backlink requests, partnership proposals generated at scale and generic pitches can pass every spam filter while still wasting your time.
There are also notifications that exist primarily as safeguards or backups. A platform may send an email because a report was generated, a payment was processed, a user signed in, a task was assigned or an automation was completed. These messages are useful if something goes wrong, but they do not all need to interrupt my day.
This is the real issue. Email treats every delivered message as something that deserves space in the same interface. It does not understand the difference between a trusted contact asking an important question and a system informing me that a scheduled process completed successfully.
If you want to reach my inbox, become a contact
The rule I created is simple:
If you want to make it to my inbox, make sure that you somehow get into our contact list.
This does not necessarily mean that I need to manually add you to an address book. It means that there must be a meaningful relationship between you and our organisation.
You may be a client, a partner, a supplier, a community member, a prospect who has entered a legitimate conversation or someone connected to one of our active projects. If GrowthOS knows who you are and why the relationship exists, your message deserves a different level of attention.
An inbox filter usually asks, “Does this message look like spam?”
My routine asks a more valuable question:
“Do we know this person, and what is our relationship with them?”
That is a much better signal.
The Claude routine that replaced my inbox
I created a simple routine in Claude:
“Get my emails. Check the senders against our contact list in GrowthOS. Send me a Slack message for emails coming from known contacts, including the time, subject line and part of the email body.
Exclude internal senders from growthrocks.com and growthgirls.com.”
Each sentence represents a separate action and requires access to a different system.
Claude needs to connect to my email account through a tool that can retrieve and read messages.
This does not mean giving an AI unrestricted control over the mailbox. The routine only needs the permissions required to read the relevant messages and their basic metadata, such as:
- Sender
- Recipients
- Date and time
- Subject line
- Message body
- Thread history, when context is required
The agent can check only unread emails, emails received within a specific period or messages that have not already been processed. This avoids repeatedly reviewing the same messages.
“Check the senders against our contact list in GrowthOS”
The important comparison is between the email sender and the contacts stored in GrowthOS.
GrowthOS acts as the source of context. It can tell the routine whether the sender is a known client, prospect, partner, community member or another recognised contact. It may also provide the company, project, relationship owner, segment and previous activity associated with that person.
This turns a basic email address into business context.
An email from [email protected] means very little on its own. An email from an active client involved in a specific project immediately deserves more attention.
The cross-check can use the exact email address, the company domain or another reliable identifier. Exact matches should normally be prioritised because matching only by domain can produce false positives, especially for large companies.
“Send me a Slack message for emails coming from known contacts”
After the sender has been verified, the routine sends the relevant email to Slack.
Slack becomes my attention layer. Instead of opening my inbox and searching for important messages, I receive a curated stream containing only the emails that match the rules.
This could be delivered to a private channel, a direct message or a dedicated channel such as #priority-emails. The channel should be specific enough to remain useful. If the routine posts everything, Slack will eventually become as noisy as the inbox it replaced.
“Include the time, subject line and part of the body”
A notification that says “You received an important email” is not enough. It simply moves the problem from one application to another.
The Slack message needs enough information for me to understand what happened without immediately opening the original email. At minimum, it should contain:
- The sender’s name and company
- The time the email arrived
- The subject line
- A useful excerpt or short AI-generated summary
- The relevant GrowthOS relationship or segment
- A link to the original email
- Whether the email appears to require a response
For long threads, I would rather receive a summary than the beginning of the latest message. The useful output might be:
Maria from Client X replied at 14:35 regarding the September campaign. She confirmed the proposed budget but asked whether video production is included. A response is required before Friday.
That is much more useful than seeing the first 200 characters of the email.
“Exclude internal senders from growthrocks.com and growthgirls.com”
Internal emails are excluded because they already belong to other communication and operational flows.
The domains growthrocks.com and growthgirls.com identify internal senders. Without this exclusion, every message from a colleague could be treated as a known-contact email and forwarded to Slack, creating unnecessary duplication.
The routine is intended to surface relevant external communication. Internal messages can follow their own rules, priorities and channels.
What the routine requires
The prompt is simple because the underlying connections do the heavy lifting. For this routine to work, Claude needs access to four capabilities.
First, it needs an email connection that can safely retrieve messages and threads.
Second, it needs access to GrowthOS and a tool that can search or retrieve contacts.
Third, it needs permission to send messages to the appropriate Slack channel.
Finally, it needs the logic to combine those systems. The agent must be able to read an email, extract the sender, search for that person in GrowthOS, verify the relationship, apply the exclusion rules, create a concise summary and post the result to Slack.
The workflow looks simple from the outside, but it is more than an email filter. It is a cross-system decision:
The AI is not valuable merely because it can read the email. It is valuable because it can combine the email with business data and decide what should happen next.
What happens to website submissions?
Filtering known contacts solves only part of the problem.
Companies also receive messages through multiple website engagement points:
- Contact forms
- Request-a-demo forms
- Partnership applications
- Backlink requests
- Newsletter replies
- Community registrations
- Event forms
- Webinar submissions
- Support requests
- Resource downloads
The label on the form does not guarantee that people will use it correctly.
A person may use the contact form to request a backlink even when the page explicitly says not to. Someone may use a partnership form to sell a completely unrelated service. Automated bots may submit promotions, scams or pharmaceutical spam. A legitimate prospect may select the wrong category simply because they do not understand your internal structure.
Traditional automation trusts the form.
If someone submits a “Partnership Request,” it creates a partnership lead. If someone submits a “Contact Us” form, it sends the message to a general inbox.
AI does not need to accept that classification.
AI can classify the intent, not just the form
When a submission arrives, an AI agent reads the actual content and considers additional signals, including:
- The message body
- The sender’s email address and domain
- The page where the submission happened
- The selected form options
- The company and job title
- Previous interactions
- Existing GrowthOS records
- The services or topics mentioned
- Commercial intent
- Spam and abuse indicators
- The action the sender appears to expect
It then reclassifies the submission according to its real intent.
A message sent through a general contact form may actually be a qualified sales opportunity. Another may be a support request from an existing client. A third may be a job application, a backlink request or irrelevant spam.
Each one should enter a different flow.
This means that website submissions can still result in people being added to our contact list, but not as an undifferentiated collection of email addresses. They are added to a specific segment, with a classification that explains who they are, what they want and how the organisation should handle them.
For example:
| Actual intent | GrowthOS segment | Destination |
| Qualified sales enquiry | Prospect | Sales workflow |
| Existing client request | Client | Account team |
| Podcast guest proposal | Media or contributor | Editorial workflow |
| Backlink request | Outreach | SEO review queue |
| Job application | Candidate | Recruitment workflow |
| Unrelated sales pitch | Vendor outreach | Low-priority queue |
| Spam or abusive content | Rejected | No further action |
The form is only the entry point. The message itself determines the route.
The agent can also prepare the response
Classification is only the first stage.
Another agent can attempt to reply based on the training, rules and context we have provided. The response should not be a generic confirmation such as “Thank you for contacting us. We will get back to you shortly.”
It should understand what the person asked, check the available business context and respond appropriately.
For a qualified enquiry, it may ask the missing questions required to evaluate the opportunity. For a backlink request, it may explain our policy. For a podcast proposal, it may request the guest’s profile, topic and previous appearances. For an unrelated sales pitch, it may decline politely without consuming a team member’s time.
The important word is “contextually.”
The agent should know which form was used, what the person wrote, whether they already exist in GrowthOS, which segment they belong to and which response policies apply. It can then draft the answer or send it automatically when the case falls within an approved category.
More sensitive, ambiguous or commercially important messages can still be escalated to a human.
Email becomes infrastructure
I have not eliminated email. I have changed its role.
Email remains a universal transport layer. Almost anyone can send a message, and almost every system can generate one. That makes email extremely useful as infrastructure.
But the inbox is no longer the best interface for deciding what matters.
The new interface is the agent that reads the messages, understands the senders, checks the business context, classifies the intent, filters the noise, routes each message and gives me only what deserves my attention.
That is why I don’t read my emails anymore.
My agents do.
I read the decisions, summaries and conversations that actually require me.
What is AI email management?
AI email management uses AI to understand, classify, summarize, and route emails based on their content and context. Instead of relying only on folders, keywords, or spam filters, an AI agent can determine who sent a message, what they want, whether they are a known contact, and whether the email actually requires attention.
How can AI decide which emails are important?
AI can combine information from the email with business context from other systems. In my routine, the sender is checked against our GrowthOS contact list. Emails from known clients, partners, prospects, community members, and other recognized contacts receive a different level of attention from unknown or irrelevant senders.
How does the AI email routine work?
The routine retrieves new emails, identifies the sender, checks that person against contacts in GrowthOS, excludes internal company domains, and sends relevant external messages to Slack. The notification can include the sender, time, subject, summary, relationship context, and whether a response appears to be required.
Can AI also manage website contact form submissions?
Yes. Instead of trusting the form category alone, AI can analyze what the person actually wrote and classify the submission by intent. A contact form submission could be identified as a sales enquiry, client request, job application, backlink request, media opportunity, vendor pitch, or spam and routed to the appropriate workflow.
Does AI email management replace email?
No. Email remains a useful universal communication and transport layer. What changes is the interface used to decide what deserves attention. Instead of manually reviewing every message in an inbox, an AI agent can filter, classify, summarize, and route messages so humans focus on the conversations and decisions that actually require them.
Was this article useful?
Theodore has 20 years of experience running successful and profitable software products. In his free time, he coaches and consults startups. His career includes managerial posts for companies in the UK and abroad, and he has significant skills in intrapreneurship and entrepreneurship.
