Not another AI bot, a real colleague to work alongside your team. Give your team superpowers or even run your company on autopilot.
Replies
Best
Really interesting direction. How do you think about control vs. autonomy - especially when decisions have real business impact? Congrats on the launch!
@jared_salois Thanks! We treat autonomy as a dial, not a default. Scarlett proposes and drafts freely, but anything with real consequences (sending, spending, publishing, external actions) goes through an approval step unless you explicitly raise the trust level for that lane. So you get speed on low-risk work and a human gate where it matters.
Report
one thing that caught my attention was the ability to use your managed keys for external services. that's a great onboarding experience. how easy is it for users to switch to their own API keys if they decide to later?
@olivia_bennett7 Appreciate that! The managed keys are just there so you can start in seconds. Swapping to your own is a settings change, not a migration: drop your key in and Scarlett uses it going forward. No rebuild, no lost history.
Report
congratulations! the she Just works philosophy really stood out. what were the biggest engineering challenges you faced over the past year to make Scarlett reliable enough for daily use?
@barnaby_lloyd Thank you! The hardest part was memory and context: keeping the right information in view without drowning her in noise. Close behind was making tool use and approvals dependable enough to trust every day, which meant a lot of work on how actions are gated, retried, and confirmed.
Report
Congrats! the autopilot feature is fascinating. can users customize how much control Scarlett has over marketing, support or operations based on their comfort level?
@joshua_martinez7 Yes, exactly. Control is set per area and per action, so you can let her run fully in one lane, require approval in another, and keep her read-only somewhere else. You raise the leash as your comfort grows.
Report
Played around with Scarlett for a bit and was surprised how naturally it slotted into our slack workflow without feeling like another chatbot. The autopilot features still need some tuning but the teammate framing actually clicked for me.
@can4ojy Really glad the teammate framing clicked, that is exactly what we were going for. And fair on autopilot needing tuning, we are actively refining it, so any specifics you hit would genuinely help.
Report
congratulations! after a year of testing internally, what was the biggest wow moment that convinced your team she was ready for a public launch?
@easton_carter Thanks! The turning point was watching her run recurring work on her own, notice something off in the data, and come back with a flagged issue and a draft fix before anyone asked. That shift from tool to teammate was the moment.
Report
Congrats on the launch! The "real colleague, not a bot" positioning is the hard part to earn. Curious how Scarlett handles uncertainty — does she escalate to a human in-thread when she's not sure, or attempt and flag afterwards? That handoff boundary is usually where trust with an AI coworker is won or lost.
Report
Congrats on the launch! I'm curious, after a year of using Scarlett internally, what's one workflow your team can no longer imagine doing manually? Those real-world use cases are always the most interesting.
Report
the "she just works" framing is refreshing after a year of watching demos that require a PhD to trigger.
real q: when scarlett does 90% of the work, does the client on the other end of the slack/imessage know they're talking to an agent, or does it happen invisibly? asking because agent-mediated work is going to be a huge trust question in the next 12 months and being on the right side of the disclosure norm feels important.
The iMessage-native choice is the detail that jumps out at me. Meeting people in the channel they already live in beats asking them to adopt a new app. We build voice AI that phones aging parents every day, and we landed on plain phone calls for the same reason: the interface people already trust tends to win over the "better" one they have to learn. Question for you, Dan: over a long-running relationship, how do you keep Scarlett feeling like the same colleague from session to session? Is the compiled, entity-scoped memory enough to hold a consistent persona, or did you end up tuning tone separately? That continuity has been the hardest part for us.
Replies
Really interesting direction. How do you think about control vs. autonomy - especially when decisions have real business impact? Congrats on the launch!
Scarlett.
@jared_salois Thanks! We treat autonomy as a dial, not a default. Scarlett proposes and drafts freely, but anything with real consequences (sending, spending, publishing, external actions) goes through an approval step unless you explicitly raise the trust level for that lane. So you get speed on low-risk work and a human gate where it matters.
one thing that caught my attention was the ability to use your managed keys for external services. that's a great onboarding experience. how easy is it for users to switch to their own API keys if they decide to later?
Scarlett.
@olivia_bennett7 Appreciate that! The managed keys are just there so you can start in seconds. Swapping to your own is a settings change, not a migration: drop your key in and Scarlett uses it going forward. No rebuild, no lost history.
congratulations! the she Just works philosophy really stood out. what were the biggest engineering challenges you faced over the past year to make Scarlett reliable enough for daily use?
Scarlett.
@barnaby_lloyd Thank you! The hardest part was memory and context: keeping the right information in view without drowning her in noise. Close behind was making tool use and approvals dependable enough to trust every day, which meant a lot of work on how actions are gated, retried, and confirmed.
Congrats! the autopilot feature is fascinating. can users customize how much control Scarlett has over marketing, support or operations based on their comfort level?
Scarlett.
@joshua_martinez7 Yes, exactly. Control is set per area and per action, so you can let her run fully in one lane, require approval in another, and keep her read-only somewhere else. You raise the leash as your comfort grows.
Played around with Scarlett for a bit and was surprised how naturally it slotted into our slack workflow without feeling like another chatbot. The autopilot features still need some tuning but the teammate framing actually clicked for me.
Scarlett.
@can4ojy Really glad the teammate framing clicked, that is exactly what we were going for. And fair on autopilot needing tuning, we are actively refining it, so any specifics you hit would genuinely help.
congratulations! after a year of testing internally, what was the biggest wow moment that convinced your team she was ready for a public launch?
Scarlett.
@easton_carter Thanks! The turning point was watching her run recurring work on her own, notice something off in the data, and come back with a flagged issue and a draft fix before anyone asked. That shift from tool to teammate was the moment.
Congrats on the launch! The "real colleague, not a bot" positioning is the hard part to earn. Curious how Scarlett handles uncertainty — does she escalate to a human in-thread when she's not sure, or attempt and flag afterwards? That handoff boundary is usually where trust with an AI coworker is won or lost.
Congrats on the launch! I'm curious, after a year of using Scarlett internally, what's one workflow your team can no longer imagine doing manually? Those real-world use cases are always the most interesting.
the "she just works" framing is refreshing after a year of watching demos that require a PhD to trigger.
real q: when scarlett does 90% of the work, does the client on the other end of the slack/imessage know they're talking to an agent, or does it happen invisibly? asking because agent-mediated work is going to be a huge trust question in the next 12 months and being on the right side of the disclosure norm feels important.
Refocus
The iMessage-native choice is the detail that jumps out at me. Meeting people in the channel they already live in beats asking them to adopt a new app. We build voice AI that phones aging parents every day, and we landed on plain phone calls for the same reason: the interface people already trust tends to win over the "better" one they have to learn. Question for you, Dan: over a long-running relationship, how do you keep Scarlett feeling like the same colleague from session to session? Is the compiled, entity-scoped memory enough to hold a consistent persona, or did you end up tuning tone separately? That continuity has been the hardest part for us.