The problem with consumer AI in a nutshell is if you hand an uninitiated user an agent, it's nothing but a blank box to them. Coders get it instantly, but for everyone else it’s just not intuitive. Solve that, and you reach the next few hundred million real adopters.

AI’s disruption to all technical work is entirely a function of the speed that non-coders begin adopting. Post Astra/Opus, it is already sufficiently capable; it’s now only a user experience and learning curve barrier.

Yup. Since almost a year ago it flipped from a model problem (finally good enough!) to an experience problem. OpenClaw was the first mini aha moment - still not for the faint of heart. But now we are just starting to see consumer experiences that can be. Why I am back in VC :)

@nikitabier The answer is: apps. While builders want super apps, or blank boxes - humans want guided experiences and namespaces. * Travel agent/app * Shopping agent/app * Health agent/app * Stocks agent/app * Money agent/app So much to ship. One use case at a time (and then spiral out).

@nikitabier Yes

It’s also more work, not less. Netflix is less work. shopping is less work. Consumers don’t really want to design productivity or delegation in to their lives.

Any consumer agent builders who want an interesting solution for their users? DM!

Thanks Josh, great consumer lens as always. Prediction: Muse gets muted by Apple Messages. Messages + DoorDash shows how Apple cracks the "black box" you describe: they win w/developer relationships, privacy, E2EE, Passwords app, ApplePay all inside iOS's most-used app. Lowest friction wins.

Thank you — and I know you get it. Man didn’t realize you were at Nvidia - you have always served the world’s most important community - developers with the biggest platforms

Working on exactly this and would love to chat about it.

The most entertaining outcome is the most likely. @elonmusk

100%. People need to get out of the various tech bubbles and invent new user experiences driven by AI capabilities for regular users

If Muse leans more heavily on this tab, time in app will skyrocket.

just like taking a person to a hardware store who has never done things with their hands… we (not engineers , not builders) tend to keep saying and thinking “we’ve never seen anything like this before” not true as humans, we’ve been through many technologically rooted, massive productivity gains… it’s just that most of us are not builders so their tends to be lots of hype and FUD while the builders quietly harness the productivity benefit of technology waves, almost always delivering many many purpose built uses of the tech where industry/domain knowledge unlocks the value of Applying the tech in daily business and consumer activities… behaviorally, Google has made one of the most useful “on-ramps” to purpose built AI Chatbot here they’ve leveraged the behavior of “Googling something” to introduce anyone who searches to a conversational search; lots and lots of people are stumbling into “purpose specific AI” in this way because their Google search is purpose specific to begin with

Hi @grok — Isn’t the blank box problem worse for coders? Prompt engineering isn’t building; it’s generating code.

The uninitiated user is 99% of households. The winning agent shows up already knowing the family.

Context informed initiation is one solution and creative discovery would is another - send me a dm and I’ll show you some stuff I’ve been working on that I think may be interesting conceptually, plus would love to hear your take!

completely agree.

Just get a mom to download everything in her head to someone she trusts. I suspect the solution is a bit more human than most realize.

coders know to type into the box. everyone else needs the agent to start the conversation and ask what they're trying to do.

absolutely, the biggest challenge is how to have a great experience with freemium while not having a large tokens bill

Exactly Josh and I am guilty of getting into that trap and so is everyone else when competing with instinct ai or muse or dots. Coders get agents instantly because they arrive with a backlog of well-defined tasks. Everyone else has to think of something to delegate before they can type anything, and delegating is a skill most people have never practised. So better onboarding or suggested prompts won't fix the blank box. The product has to supply the task itself. We are now looking into the mvp for a couple therapist with similar concept to dots! Do you think a self aware agent like that has a place in the ecosystem?

hear hear

building incredible onboarding is key, coz onboardin would be disguise giving the the system enogh context for it to never start off as a blank box so onboarding ideally becomes the new frontier here.

@illscience What are your thoughts on general purpose consumer agents vs branded, packaged, use case specific agents

@illscience I think the answer is both ! And sometimes your general purpose agents will transfer you to a specific / expert one

Grok Bot now proactively scans your email and identifies ways it can help which is very helpful.

Exactly

The ai capability improves every week, but the ux is years behind.

Start in the share sheet: send a PDF, screenshot or email to the agent, then offer two things it can do with it. The user already has the object; making them describe it again in an empty chat box throws away the context.

One thing Instinct nailed was asking users to connect their email early. Once it has that context, it can immediately show what a personal agent can actually do. And I do think the next wave is gonna make these agents friendly and polished enough that people outside tech start using them every day. Like how we came from OpenClaw to what we have now.

Its a very chicken and the egg problem. Even with Muse you need to approve connectors. Do you see a middle data aggregation layer showing up?

Yeah

Exactly. Nobody wants a one click bot, they want a one click plumper, electrician, mechanic, etc.

@illscience IMO the UI is actually a very powerful lever here, one takeaway from Muse is that it still matters enormously even with models with exceptional general capabilities

@illscience AI native hardware will take that lift. 1 button + voice-only + standalone design without phone dependency

yeah empty chat box is rough. starting from stuff already in their inbox helps a lot

@illscience I love watching talk about things they are passionate about