Would you let AI handle your inbox? I’m doing it for science
Abstract created by Sensible Solutions AI
In abstract:
- PCWorld explores the dangers and advantages of utilizing AI brokers like Claude for electronic mail administration, following Notion Mail’s latest shutdown that left customers depending on AI sorting.
- AI electronic mail automation gives interesting advantages together with decreased inbox muddle and higher group, however poses operational dangers like misfiling or unintended deletion.
- Privateness issues stay vital as AI brokers entry delicate private and monetary knowledge, requiring customers to fastidiously weigh comfort towards potential safety dangers.
After I realized that Notion, the favored on-line workspace service, was shutting down its Notion Mail product, it wasn’t the shutdown itself that acquired my consideration.
No, it was this: as a result of so many Notion customers had handed over their electronic mail sorting duties to AI brokers, they’d stopped bothering to open their inboxes.
Letting AI brokers type via all of your electronic mail has lengthy been thought-about a killer app for AI, though the comfort doesn’t come with out some critical dangers.
First, there’s the privateness facet. Do you really need AI brokers poking via all of your electronic mail messages? What in the event that they see one thing private or spot an account quantity? May an agent be tricked (via immediate injection or in any other case) to leak your non-public knowledge to an attacker or one other third social gathering?
Second, there’s the danger of one thing going haywire. What if the brokers mis-file a message or (gulp) deletes your whole inbox, all due to a foul immediate? Or what if the AI sends an electronic mail to the unsuitable particular person or distribution group?
These are all good questions, and the solutions for many of them are: It relies upon. Totally different AI suppliers provide completely different guardrails for his or her mail-handling brokers and integrations, and in relation to do-it-yourself agentic instruments like OpenClaw, something goes.
So sure, there are many causes to really feel cautious if downright scared of the prospect of letting AI brokers unfastened in your inbox — and to be clear, this entails taking issues fairly a bit additional than a easy, read-only AI triage.
However there are some tempting upsides too, reminiscent of getting slightly assist plowing via the hundreds of unread “Different” messages in my Outlook inbox. It might be a significant aid to get a crew of brokers to type via all these emails, plucking out the handful of “actual” messages and nixing the remaining.
It was the considered my ballooning Outlook inbox that persuaded me to present AI brokers a crack at my electronic mail — and whereas I am sorely tempted to enlist their assist with my work Outlook, I’m making an attempt it first utilizing a separate Gmail account. (Chill out, boss.)
As I discussed earlier than, there are many AI suppliers that supply electronic mail integrations, together with Gmail. Google’s Gemini, after all, can faucet into your Gmail, as can ChatGPT and Notion. I selected to present it a attempt with Claude, by way of the Cowork tab within the Claude desktop app.
Claude’s Gmail integration comes with a wide range of safeguards, together with the truth that it might solely draft electronic mail messages however received’t ship them with out your OK. Likewise, Claude can put electronic mail messages within the trash (the place they’ll sit for 30 days), however can’t delete them completely. It’s additionally straightforward to dam Claude from coaching in your knowledge, together with the Gmail it reads.
I labored with Claude to find out what it could — and wouldn’t — do with my electronic mail every morning.
Ben Patterson/Foundry
Utilizing Claude Opus 4.8, the present most-powerful Claude mannequin for on a regular basis customers (or at the least till Fable lastly returns), I created a morning Gmail automation: Test all my messages from the previous 24 hours, classify every thread as “Essential” or “Archiveable,” label the “Archiveable” messages and take away them from my inbox, label something that appears like a receipt, give me a triaged abstract, after which draft replies in my very own voice to enterprise or college correspondents, whereas leaving replies to mates and family members as much as me.
It’s solely been a day since I began my AI-in-my-Gmail experiment, however I’m already seeing outcomes. My inbox is reduce right down to measurement, the “unread” depend is far smaller, and some stray receipts have been safely tucked into my “Receipts” folder. Claude hasn’t drafted any electronic mail messages but, however I think about I’ll see some within the coming week.
Am I involved about Claude sifting via my private messages? Certain, however not any greater than I’m about Google doing the identical factor.
And whereas the danger of Claude pulling a HAL 9000 with my messages can’t be discounted, the promise of getting my inbox beneath management is making me prepared to roll the cube.
I’ll preserve you posted.
