Jean-Denis Greze 💡

I generally agree.

The best platforms will allow you to just give intent, or the result you want in the world, and the AI will increasingly be able to take it from there.

Today, though, most people see empty text boxes and don't know what to put in there because their mental model of what can/cannot be done gets stale fast as the technology improves.

I think knowing that you can now ask it to do more complex things, or even the model's self-awareness about what it can do, can lag.

(A great tip is to ask AI "how would you do that and can you do it for me?")

x.com/thdxr/status/2…

Forbes put Town on its first Agentic 20 list, alongside OpenAI, Meta, Stripe and a bunch of other great teams building AI agents.

There’s so much work in everyone’s day that isn’t particularly hard. You just have to remember to do it. Following up with someone, scheduling the meeting, finding something buried in your email, keeping track of all the little things that otherwise fall through the cracks.

I want Townies (your AI assistants) to increasingly take care of that stuff for you. The less time you spend telling an assistant what to do, the better.

We have a lot left to build, but it’s very cool to see Town included here.

Thanks to @annatonger and @Forbes for the write-up, and thanks to everyone using Town.

If you've ever met the folks who work at Town, you know that the heart and magic of the product comes from my co-founder @tonydevincenzi and his incredible team.

This @arenamag piece by @freeshreeda is special to me because she captures Tony's constant striving for creating a product with a real soul. Take a read.

new in Arena by me

It's precisely because @TownAI is so good that I started to ask questions like "what is the ultimate end-goal/purpose of software wrt the human condition?"

are we approaching the end of software? is software finding its final form?!!!!

Big September of #shipping, and also Townies that do a lot more for the same price!

Fun to talk with @jgreze (@TownAI) and Deeni Fathia (@GammaApp) about building a consumer AI business

These two teams are best-in-class in developing delightful, retentive products that individuals adopt...and then bring into work.

My takeaways:

- Consumers are unlikely to pay for AI directly (via subscription). But, using consumer as a low cost enterprise funnel can be very effective, especially if you have a product where the output is naturally shareable and is used to do meaningful work.

- You can't serve everyone. Some highly technical users are going to want a level of granularity / dial tuning that doesn't make sense to provide. You can lean in to this in select areas of the product where the average user does care - for Gamma, this is selecting image models.

- For products that are adopted across a team, pooling tokens is one way to approach pricing. Employees will naturally have varying levels of adoption of AI tools, and allowing sharing across a team can feel more fair / give customers higher ROI (they'll be more likely to stay!)

- The ecosystem resets every 2-3 months, so you need to be always reinventing yourself as a company. Gamma (which has more than 100M users) just rebuilt and relaunched its product in ~3 months to better serve how people are looking to tell stories with AI now.

- Trying to beat the labs at their own game is often a mistake. Leaning into a specific angle that you feel unlocks the market (for Town, this is multiplayer) or a specific type of customer is often better than trying to take the same approach with fewer resources.

Our @a16z team is hosting a private after hours for founders & operators next week in SF @immad @gaganbiyani @jgreze more with lightning talks plus drinks snacks & good convos few spots left partiful.com/e/RMn4q33co7VK

Join tomorrow: this should be a blast!

our @a16z team is hosting a private "after hours" event for founders & operators next week in SF!

@immad (founder of mercury), @gaganbiyani (founder of maven, udemy, and horowitz andreessen academy), @jgreze (founder of town) and more will be there + give lightning talks

rest of the night will be drinks, snacks, and good convos

we have a few spots left! apply here to join: partiful.com/e/RMn4q33co7VK…

I love it when people I actually know can deliver on the @assanalysis.

@TownAI and their team put in the reps and it's clearly delivered.

If you've got an agent you want to submit to AA, dm me, or just submit on assistantanalysis.com

The @AnchorBrowser manages to get these evals in between solving some pretty gnarly agent problems too.

We have a new #1 on the Assistant Analysis verified leaderboard and it's @TownAI

94.8% verified task completion on Benchmark v0.2.0 — the highest score we've published. Congrats to @jgreze and team!

It turns out Stripe is… Stripe for agents!

We’ve had this for a bit and excited to make it official! Buy all of the things (safely) with Stripe Link and Town!

I think a lot of people are about to start doing work outside their traditional role because AI allows them to, and if that's right, a lot of vertical AI products aren't going to work.

If you look at people's jobs across the economy, more people will do a greater variety of things in their role. They won't just do the one thing anymore.

Say someone builds a vertical AI product that helps a dentist with all the dentist stuff.

Then the dentist needs to do some marketing for their practice. Or something else outside the core set of things that product was built for.

Small business owners have to do a lot of different things. If someone else built a more general product that lets them do an okay job at all of it, they're going to use the general product.

That's what I see.

Agents have started buying things online over the last month or so, and we still don't know who should be responsible after a bad purchase.

With Stripe's Link, you get a notification and approve every purchase. I think Stripe set it up that way because mistakes can happen. It's a good product.

But there are lots of cases where people won't want to approve every time, like repeat purchases, or ordering from a wholesaler. Some people draw the line at money moving around, and I get that.

Here are the questions I think are still in play.

- How much can the agent spend?
- When should it ask a human for approval?
- Which categories are off limits?
- Who should be responsible for a bad purchase - the merchant, the processor, or the user?
- How should recurring purchases work?
- Who should catch purchases that don't match your usual pattern?

There is a space to build businesses here! Or maybe we all trust our agents to get it right?

There's no alpha in being lazy with AI.

Lots of really smart people out there are trying to do things alongside you. If everyone's being lazy and one person isn't, that person will likely find something everyone else is ignoring.

The fact that it has become easy to create doesn't mean it's good. AI can only take you so far, and then you still have to ask yourself if you're bringing more value.

I recently bought a baseball glove for my 7-year-old son, and I didn't particularly enjoy doing the research, which got me thinking about the two very different kinds of shopping I do.

Imagine I'm walking around SoHo, going into cool storefronts, seeing well-dressed people on the street and asking myself if I want to look like them. That's fun. I don’t want an assistant doing that for me.

For a lot of my purchases, though, I just need a thing.

So for the glove, I asked Town to research gloves for kids his age, explain the different sizes, read some reviews, and give me a few options. When it did, I looked at the options, picked one I liked and asked Town to buy it.

The research was important but I wasn't looking to enjoy it.

A lot of shopping for me is: I need a thing. I don’t want to make an obviously bad choice. I care a little about the color and how it looks and feels.

Beyond that, please just get it done. I get the argument that people enjoy shopping and browsing. I do too.

Just not for everything I buy.

Town top-of-funnel team be like

Instead of shipping "me too" products months after everyone else, we're really focused on pushing the boundary of what's possible with AI work assistants. It's what our customers pay us for!

If you've been following along, multiplayer has been front-and-center for us:
- Agent to Agent
- Townies in Group Texts (x.com/TownAI/status/…)
- CC Your Townie (x.com/TownAI/status/…)

And now, a multi-person and multi-Townie experience that allows entire organizations to work at the speed of AI. I'll be talking a lot more about this new world we're enabling in the weeks aheads.

Let's go!

Per @CristinCulver: "The OG AI mascot 🤩"

I think the moat in AI assistants is going to be multiplayer. Here's why:

One of the features people who really get into Town love is agents working together.

You ask your assistant (we call it townie) something it doesn't know. It figures out that someone else's townie might have the answer, asks it, and brings the answer back to you.

Once your whole team is using it, it's hard to imagine going back to a product where every assistant is on its own.

I don't think anyone has really figured out multi-user AI yet. That's where the network effects are going to come from.

Talked about it with @HarryStebbings on 20VC: