Open any email client launched in the last 18 months. There is a button labeled “Summarize.” You click it. A paragraph appears. It tells you the email is from your accountant about Q1 filings.
This is impressive technology. It is also solving the wrong problem.
The right question is not “what does this email say.” The right question is “did I need to open this email at all.” AI summaries answer the first. They make the second worse.
The summary trap
Here is the dynamic. You open your inbox. There are 200 unread. You start clicking summarize on the ones that look interesting. Each summary is a few seconds. You read 40 of them. You feel productive.
But you spent 30 minutes generating summaries of emails that, half the time, you did not have to read at all. The summary did not save you the open. It just made the open faster.
Triage saves you the open. Summarization, applied to the wrong layer, just optimizes the activity you should not be doing.
When summaries actually help
We are not anti-summary. Summaries are useful in three specific places:
- Long threads you have to catch up on. A 35-message thread you missed for two days, where the relevant question is buried at message 22. Summaries earn their keep here.
- Forwarded chains with quoted history. Untangling who said what is a real chore. Summaries help.
- Compliance review. “What was discussed about X across last quarter's mail.” Summary plus search.
In all three cases, you have already decided the email matters. The summary helps you process it. That is the right layer.
When summaries do not help
Summaries do not help when you have not yet decided whether the email matters. They give you false confidence. You read the summary, feel informed, archive without reply. Two weeks later your colleague says “did you see that email?” and you say “yes” and you have not done what was asked.
Summaries are also unreliable on the threads that matter most. Tone matters in important email. Subtext matters. A summary flattens both. You miss the “this is the third time I am asking” energy. You miss the implicit deadline.
A summary of an angry customer email reads like a support ticket. The email itself reads like a fire alarm.
Why the industry shipped this
We can guess. Summaries are demo-friendly. They look magical. They make a feature on a Product Hunt page. The actual hard work, building a triage layer that knows your history with a sender, takes longer and is less photogenic.
We respect the tradeoff. We just disagree with it. STAMP does not ship a one-click summary. We ship a tag system that tells you why a thread is in front of you. Different bet.
Privacy is the other half of this
Most AI summaries today work by sending the email content to a third-party server, getting back a paragraph, and displaying it. Some do this on-device. Most do not.
If your email client is sending your investor's confidential message to a server you have never audited, that is a security posture you should be aware of. We covered this in why your email client should not read your email.
STAMP's view: classification happens on-device. We do not send your email anywhere to summarize it. If you ever want a summary, we can do it on-device with a smaller model, locally. Slower, less impressive, no privacy cost.
What a triage-first email client looks like
A triage-first client answers, before you open anything: which seven threads need a human reply from you today, and why.
That is the layer we should be optimizing. Once you know the seven, you do not need a summary of the other 193. You need to ignore them.
If you want this paradigm, STAMP is one option. Hey is another, with a different metaphor. Both are bets on triage being the missing layer.
What to try this week
Stop clicking the summary button. For one week, only click into emails you have decided, before reading, are worth a real human reply. If you cannot decide from the subject and sender, you probably did not need to open it.
You will read fewer emails. You will not miss anything important.
Where to go from here
If this argument is interesting, email triage — the missing layer is the longer version. On-device email classification, explained is the technical follow-up.
Triage first. Summaries later. hello@stamp.email