Arnav Kothari

building and deploying products for enterprise storytelling | co-founder @thenarrativecom

[Arnav Kothari @arnavkot]

We spent the last few months building a list of the 10,000 most influential people in tech on X.

How we defined "influence" by making a web:

1. We started with about 100 accounts nobody would argue with, pulled who they follow, then pulled who those people follow. Over 500,000 accounts surfaced.

2. Then we scored every account by who inside the web follows it. Not a raw count. A follow is worth more when it comes from someone influential. Karpathy following you moves your score a lot. An unknown account following you barely moves it at all.

3. Once we had the final 10,000, we pulled each person's last 5 posts, skipped the pinned post, and took their median reach and likes. The final ranking blends that with the follow score.

One of the most influential people on the list has 3,742 followers. An AI researcher whose work gets cited by the biggest names in the field. Most of the major AI CEOs follow this person. But if you sorted by follower count you'd never find them.

Meanwhile some accounts with 100,000+ followers didn't make the list at all. Big audience, but almost none of the people actually building AI follow them.

Spent about $20,000 (the X API is expensive to say the least).

I think most people confuse reach with influence. A post that gets 50,000 views from the right 50,000 people is worth more than a post that gets 500,000 views from the wrong ones.

We can finally partially measure that, and we're now deploying it across our clients to figure out who they need to reach.

With everything going headless, databases are easier to derive value from than ever.

Comment "List" and I'll send it to you.

[Arnav Kothari @arnavkot]

Fable 5.1 has come closer to human writing.

I've always been surprised that writing is a harder problem than code. Especially since both have countless valid answers.

The difference: code's answers are checkable. You can write tests, and the output passes or it doesn't.

Writing has no test suite. The only checker is a specific human's reaction, and it changes with every reader.

You can't optimize toward what you can't score. And when you score by averaging thousands of readers, you get exactly one thing: models that write for the average reader. And nobody is the average reader.

[Arnav Kothari @arnavkot]

@thsottiaux Yup same but I still browse X on Dia - functionally Codex does it all but certain UI just is better on a larger screen

[Arnav Kothari @arnavkot]

@paulg likes and follows from probable spam are also being hidden. ai makes it easy to build these filtration layers over large datasets. expecting instagram to do something similar soon

[Arnav Kothari @arnavkot]

A week ago X removed re-posts from showing up in your profile feed

Finding it in a sub-tab in your profile is a strategic move that results in more quote-posting

Looks like their gunning for a community-building ripple effect

[Arnav Kothari @arnavkot]

X before Elon was pre-AI

Post AI

- distribution has never been more important
- there's more noise than ever

That's the opportunity @aaditsh and I are capturing with The Narrative Company.

[Elon Musk @elonmusk]

𝕏 is reaching new all-time highs in usage almost every week

[Arnav Kothari @arnavkot]

Personal branding 101

No one understands reach on X better than the team behind it

[Ben Lang @benln]

Grok Bot internal best practices from 6 teams at SpaceXAI:

[Aadit Sheth @aaditsh]

Looking for a top engineer for some amazing contract work.

If you're an engineer at a company like @tryramp @stripe, @AnthropicAI, @cursor_ai we'd love to work with you.

We're building the "For You" for enterprise social teams. The best way for an enterprise to know what to engage with on social from their corporate and executive accounts.

If you're interested in building narrative intelligence on top of X (it's super fun), DM me.

[Arnav Kothari @arnavkot]

we built Nia - a "For You" for enterprises

our tool identifies conversations an enterprise would benefit from engaging with

i hit a wall on how to improve our accuracy and then i came across this post by @XOpenSource

since we're building our own recommendation system over X, i got inspired to throw their github repo into Codex and pull ideas from it to improve our algorithm

sometimes it's just a little move like this, which makes being the 'idea guy' worth it

[X Open Source @XOpenSource]

https://t.co/soxNY8YdP7

[Arnav Kothari @arnavkot]

most investors: "if you can't describe your company in 30 seconds, you don't have a scalable business yet."

enterprise storytelling requires applying this mindset to your business regularly. as you expand, can you continue to break yourself down into a handful of narratives?

[Arnav Kothari @arnavkot]

X is where I go when I've got more sessions running than my brain has context for

[Aadit Sheth @aaditsh]

Nia (Narrative Intelligence Agent) is coming together very nicely.

At this point, it's scouring tens of thousands of posts a day and delivering only 3-4 on X that are actually worth our customers engaging with.

It filters out 99.9% of the BS you have to read when you're on the X timeline.

And it doesn't just match on your company name. It also catches relevant industry conversations around your product, company, or leadership. Then it surfaces what's worth engaging with and advises whether to reply, repost, or quote post.

The magic is in the filtration. There will be at least tens of posts daily that SEEM like they're worth engaging with. But they won't pass Nia's filters. Engagement isn't strong enough, or not enough influential profiles follow the original author or there might be several other reasons. The algorithm is pretty complex at this point. But what's true is simple: if Nia surfaces a post, it's likely worth engaging with. And it'll guide you on what to do.

Nia has already helped a few replies and posts hit a couple million impressions (pretty huge, when brands normally get thousands of views). I'm particularly excited to get Nia into more corporate social and exec social hands. Not everyone has the time to use X themselves. Nor do they want to. But when it comes to promoting, they still want to talk about their launch on X. That's not how it works.

X rewards you for being a PART of the conversation. Replying, quote posting, adding thoughts, extending the thread. You cannot just show up, launch your product, and get millions of views unless you spend a bomb on ads. And then you're not growing the brand organically anyway.

Nia helps you get into the right conversations on X that are trending up, at the right time, for you and your company. Excited to see how it evolves. If you'd like to check it out, DM me.

[Arnav Kothari @arnavkot]

Make a default skill or workflow that allows you to pull other harnesses in.

When I want Grok to do a search for me inside Claude Code, I just wanna say "use Grok to search XYZ" or "use sol 5.6 on high against PR #."

With harnesses being able to do multi-agent work, this is v imp. Users may end up spending less tokens on Claude Code but at least they won't churn.

[Arnav Kothari @arnavkot]

@aaditsh Legend, thank you

[Arnav Kothari @arnavkot]

@Billovskii Sent!

[Arnav Kothari @arnavkot]

We spent the last few months building a list of the 10,000 most influential people in tech on X.

How we defined "influence" by making a web:

1. We started with about 100 accounts nobody would argue with, pulled who they follow, then pulled who those people follow. Over 500,000 accounts surfaced.

2. Then we scored every account by who inside the web follows it. Not a raw count. A follow is worth more when it comes from someone influential. Karpathy following you moves your score a lot. An unknown account following you barely moves it at all.

3. Once we had the final 10,000, we pulled each person's last 5 posts, skipped the pinned post, and took their median reach and likes. The final ranking blends that with the follow score.

One of the most influential people on the list has 3,742 followers. An AI researcher whose work gets cited by the biggest names in the field. Most of the major AI CEOs follow this person. But if you sorted by follower count you'd never find them.

Meanwhile some accounts with 100,000+ followers didn't make the list at all. Big audience, but almost none of the people actually building AI follow them.

Spent about $20,000 (the X API is expensive to say the least).

I think most people confuse reach with influence. A post that gets 50,000 views from the right 50,000 people is worth more than a post that gets 500,000 views from the wrong ones.

We can finally partially measure that, and we're now deploying it across our clients to figure out who they need to reach.

With everything going headless, databases are easier to derive value from than ever.

Comment "List" and I'll send it to you.