Should have seen it coming but will still be interesting
Weโre ending our partnership with Cursor following its acquisition by SpaceX. Under our proposal, Cursorโs direct access to our models would end on November 12.
We know that the people most affected by this decision are the developers who rely on OpenAI models in Cursor. We care about their experience in this transition and weโre ready to go above and beyond to support them.
openai.com/index/our-decisioโฆ
1
60
one thing iโve noticed with a growing number of younger gen z founders is the extreme overuse of AI.
every product decision is generated by AI. every opinion comes from AI, rather than the founder.
i wonder if thatโs why so many of the biggest recent wins seem to be coming from 30+ buildersโฆ
people who developed product judgment before AI now have a huge a big advantage right now. people who outsource what seems like all of their thinking to AI will basically never develop that judgment at all.
Stasia Carson ๐พ ๐ฎ retweeted
First time founder: I want to build something nobody's ever built. If it looks like anything that exists, I won't try it.
Second time founder: I want to be 95% like the businesses that already work, and 5% different based on something I understand well.
Only been a month and it's up only. Assistant category booming. 20% weekly growth and more on the way. iykyk
Most AI assistant tools want blanket access to your inbox on day one. We went the other way. @hellodeckai gets its own email address. You CC it, forward it, put it to work. Just launched on Product Hunt.
producthunt.com/products/decโฆ
1
1
39
Stasia Carson ๐พ ๐ฎ retweeted
the models have no moat (OpenAI, Anthropic, XAI)
the IDEs have no moat (Cursor, Windsurf)
the harnesses have no moat (Cognition, Factory, LangChain)
the app builders have no moat (Replit, Lovable, Bolt)
the wrappers have no moat (Harvey, Abridge, OpenEvidence)
the inference providers have no moat (Together, Fireworks, Groq)
the voice layer has no moat (Sierra, Decagon, ElevenLabs)
the data labeling companies have no moat (Scale, Surge, Mercor)
the AI infrastructure has no moat (Baseten, Modal, Railway)
the neoclouds have no moat (CoreWeave, Lambda, Crusoe)
the generative media companies have no moat (Runway, Higgsfield, Suno)
apparently nobody in AI has a moat
except the venture firm โ ๏ธ
We're moving towards an era where AI works in the background and grows more and more multiplayer.
Group tasks on @hellodeckai is a step in that direction.
It functions like a beacon that can perform some [arbitrary task] on a schedule and report back to a group, team, etc.
Some pretty amazing things being done with these.
The most useful update is usually the one everyone needed, but only one person received.
A flight price drops. A client follow-up is due. A recurring report is ready. One person sees it, then has to copy it, forward it, or remember who else needs to know.
Group Tasks on Deck change that.
Deck is an AI assistant that can handle scheduled tasks and share the result with multiple people at once. Set up the task once, choose who should receive it, and let Deck keep everyone in the loop when thereโs something worth knowing.
Itโs useful for founders sharing weekly business updates, sales teams tracking follow-ups, consultants delivering recurring reports, operators coordinating deadlines, or friends waiting for the right moment to book a trip.
And itโs not limited to one workflow. Deck can support email automation, calendar management, recurring reports, follow-ups, research, and other everyday work automation, without adding another complicated productivity tool to your routine.
Some things really are better together.
3
1
72
The future of AI is probably going to feel a lot more like working with a colleague.
You give them context. You tell them what matters. And you trust them to keep making progress after you stop talking.
For @hellodeckai , this is our Vigil World Engine. It takes snapshots throughout a user's day and nudges them with interesting insight using just-in-time modeling.
The interesting part is that the user gets value from it, and we stay top of mind so long as the data is unique and relevant.
Then you get this feedback loop. The more useful the insight is, the more people engage with it. And the more they engage, the better signal you have that you are actually helping.
36
Continuing my diatribe: I think a lot of people underestimate how much effort it takes to use a blank chat box well.
You have to turn a half-formed thought into a clear enough question that the model can do something useful with it.
Take a new AI product that says it can help you with content. First, you have to conceptualize what you are trying to do and put that into words. Then you have to ask a good enough question to get to something genuinely valuable in the first or second hit.
At that point, the natural comparison is just ChatGPT.
45
To be clear: memory in your app is **not** just some nice feature... ๐
I think consensus is pretty clear that memory in your app is just some nice feature you add later (or don't).
If you are asking people to build a relationship with an AI product, it has to remember enough that the next conversation feels like a continuation of the last one. Running into a lot of new AI apps where it just doesn't remember and it's infuriating.
70
One thing I think about frequently is whether chat is going to end up being a temporary interface for AI.
It works. But it asks the user to show up with the problem, the context, and the question every single time. From my xp, asking the right question is 95% of getting the right answer.
38
You have a conversation with somebody in a meeting, send them a few things, and then you break. The expectation is that you're handing off things and whatever the goal/objective is, you both keep working towards that goal/objective. That is pretty close to how I think AI should work as compared to a lot of things which are one shot today.
43
I think consensus is pretty clear that memory in your app is just some nice feature you add later (or don't).
If you are asking people to build a relationship with an AI product, it has to remember enough that the next conversation feels like a continuation of the last one. Running into a lot of new AI apps where it just doesn't remember and it's infuriating.
Email is not glamorous. But it is where a lot of a business and day-to-day runs.
That is why I think it is such an interesting place for agents.
49
Stasia Carson ๐พ ๐ฎ retweeted
we really quickly went from "one-person billion-dollar company"
to "it's over, only big startups survive"
Replying to @etainos
@andrewdsouza this is a fun one, @hellodeckai just made โcreate anythingโ a forward action ๐
Ok was skeptical of the new codex voice. Now, wow. One thing I wish was easier is having parallel convos or being able to give multiple directives. Anyone found a way to do this? As in delegate a task to one session and then continue a convo with another?
1
35










