Launched this week
Deck
The most capable AI assistant with its own inbox
95 followers
The most capable AI assistant with its own inbox
95 followers
Deck is an always-on AI assistant with its own email inbox. CC it on any thread or hand it a task, and it takes over. It learns your preferences and priorities over time, proactively flags what needs your attention, and can run on a schedule to deliver briefings and status updates automatically. It's built to take ambiguous, multi-step tasks head on, so you can move on with your day.
Deck
The "own inbox" framing is smart — most AI assistants just react to what you send them. Does Deck proactively flag things in that inbox, or is it purely on-demand when you check in?
Deck
@ark_y_k yes, you can set automations for things that hit the inbox immediately (we call these "Actions") and on a schedule (we call these "Scheduled Tasks"). for instance, I use it for reviewing our new customer sign-ups from Loops and send me a bit more about each that might be relevant. this works because the assistant has its own email address, so I can just add my assistant to our Loops team. anything in particular you're thinking about on flagging?
the cc-into-inbox approach is genuinely clever, feels way more natural than opening yet another app to delegate. curious how well it picks up on tone and priority cues over time.
Deck
@songlsankxxlh great question! we have a memory layer built in to learn more about the way a user wants to comm. additionally, there are configurations for how you'd like your assistant to respond, key persons, etc. anything that comes to mind when you think about priority cues?
CC-ing an AI assistant on a thread and having it just take over is a genuinely different interaction model than opening a chat window, that part is clever. what happens when it takes an action based on a misread of an ambiguous email though, does it check with you before actually replying or scheduling something on your behalf, or does it just act and let you catch mistakes after the fact?
Deck
@omri_ben_shoham1 Hey! It doesn't send outbound to anyone but the assistant owner. Our customers are a mix of independent operators and businesses who (mostly) prefer we don't send outbound on their behalf. It's something we've thought about, but at this point it's just too risky. instead, what it will do is chime in (to you) with an angle, a data point you might have missed, or some other insight given what context it already has in thread. as you can imagine, I use it quite a bit for outbound in which case it comes in handy for logging convos, coaching on objection handling, allowing me to have all our data on hand for any interaction and drafting follow-ups.
Been waiting for something like this, the idea of just CCing an agent into a thread and walking away is genuinely useful. One thing that would make me trust it faster is a clear rollback trail, so I can see exactly what it sent on my behalf and undo or edit a message within a short window after it goes out.
Deck
@caner396952 Hey! As of today, Deck doesn't send on your behalf. Our customers are a mix of prosumer and business and they didn't want carte blanche on sending. Curious if that would be something you would want in an email assistant product>
Loomal
"most capable"... as in? Are there any benchmarks published? :D
Deck
@krutiparekh16 "most capable" as in end-to-end multi-step work over email, not just chat. benchmarks soon.
The "give it its own inbox instead of blanket access to mine" call is the right instinct. An agent with its own address that only emails me is a far cleaner trust model than yet another tool asking to read my whole mailbox, and leaning on email as the delegation surface is smart since that behavior already exists.
Where my mind goes is the flip side: acting on anything it's CC'd on means inbound email becomes the attack surface. Wondering how Deck tells a real instruction from a spoofed or injected one. If someone emails its address pretending to be me or my boss, does it act with the same authority, or is there a trust boundary around who's actually allowed to make it do things?