Alex Pareto

Building.

Previously Head of Spend Management Engineering @brexhq, founder Demeanor (@ycombinator S18), NTWRK, @facebook

[Alex Pareto @alexpareto]

React's downloads per day are sort of an interesting measure of how many people are writing code during a season (winter > summer)

[Alex Pareto @alexpareto]

Amazing how LLMs are changing UI patterns.

Notion blocks were simple enough that people started shipping websites with them.

Now there is a race to unwind that… to make room for a new type of UI easily configurable agents. But still as human reviewable as a Google Doc.

[Notion @NotionHQ]

New block in Notion: HTML.

Build interactive HTML right on your Notion page. Ask AI to turn your content into interactive explainers, prototypes, or diagrams.

Share with your team to use and tinker together.

[Alex Pareto @alexpareto]

What's fascinating about recent model developments is how cost of frontier intelligence seems to have stabilized around Opus level ($25/MTok). If anything it might actually be going up (see: Fable).

The price decline historically was aggressive. Remember o1 costing $60/MTok and o3 costing $8/MTok a couple months later?

~$25/MTok feels like a price that OpenAI + Anthropic have found that people are consistently willing to pay at scale.

I think it's why we've seen many Opus + GPT 5.5 tier updates - but less so for Sonnet/Haiku or Mini models.

We'll see if the pressure from open source and the phase out of Max plans brings a refocus

[Alex Pareto @alexpareto]

For a little bit I was anti-post training models. The bitter lesson was that these would be outclassed by frontier models - and we saw this repeatedly over the last few years.

This was especially true as frontier model prices seemed to be cut in half as performance doubled.

But over the last 6 months, (American) model costs have stalled while performance has continued to increase. It seems that both OpenAI + Anthropic are focused on their Opus tier models in a battle to win complex agentic use cases.

What this means is that post training models still is not the best way to get the best overall performance - Fable 6 will still beat Harvey/Decagon's in house models when it's ready. But, Harvey/Decagon will be able to serve their in house model at a fraction of the cost for only marginal performance loss.

Post training isn't a quest to beat frontier models on performance - it's a quest to beat them on intelligence per dollar for a specific use case.

The question will be if frontier labs start to focus on efficiency + costs again - and we start seeing models like Haiku 5 or GPT Mini bring the same performance as Opus 4.5 and GPT 5.4.

That's when the bitter lesson will play out.

[Alex Pareto @alexpareto]

A corollary of this is that Codex/Claude Code could cause the third party harnesses a lot of grief if they were to offer the Max plans to enterprises as well.

But of course, the cost may be too much to stomach at that scale.

[Alex Pareto @alexpareto]

Took me a while to understand how Factory/Devin compete with the Anthropic/OpenAI $200/month Max plans. Regardless of product quality - per token pricing is too steep for any cost minded person to justify against the Max plans. On token pricing our team would regularly spend $5000+ per person per month.

The reality is - they don't. They compete with enterprise plans once Anthropic/OpenAI move customers to per token pricing plans (at ~150 users). Then the token savings from the third party harnesses become extremely compelling.

[Alex Pareto @alexpareto]

Took me a while to understand how Factory/Devin compete with the Anthropic/OpenAI $200/month Max plans. Regardless of product quality - per token pricing is too steep for any cost minded person to justify against the Max plans. On token pricing our team would regularly spend $5000+ per person per month.

The reality is - they don't. They compete with enterprise plans once Anthropic/OpenAI move customers to per token pricing plans (at ~150 users). Then the token savings from the third party harnesses become extremely compelling.

[Marty Kausas @marty_kausas]

Our Anthropic bill is about to jump from $400K → $1.4M/yr.

Not because usage exploded, but because we're about to cross 150 seats.

Past 150 seats you're forced into Enterprise tier. Seats stop including any usage, every token bills at standard API rates. At our current run rate that's 3.5x overnight.

Unfiltered thoughts on AI spend:

1. We should spend tokens to grow as aggressively as possible. But most people (me included) aren't conscious of what they're spending.

2. Visibility comes first. People see their personal number and they're shocked. I accidentally spent $4,000 in 3 days in Claude Code.

3. For engineering the spend is clearly worth it. Pay for the best model, it saves more than it costs.

4. For a lot of other roles it's questionable. Apps nobody uses, skills someone already built. No ROI.

5. Spend limits are coming. We already require approval for more tokens on our support team.

The era of token-maxxing is coming to an end.

[Alex Pareto @alexpareto]

Couldn’t think of a better campaign to highlight Brex’s unique culture. I joined as an engineer and over the years got to launch multiple zero to one products from scratch, work hand in hand with Pedro, and, best of all to learn and be surrounded by many talented builders - that community goes far beyond the office.

Incredible school for founders!

[Pedro Franceschi @pedroh96]

Brex is the only company proud of how many people leave.

Over 100 alumni became founders (4x the industry average), raising more than $800M along the way, with $1B+ in exits, and even 1 IPO.

@brexHQ is the best founder school on the planet. Quitters Welcome.

[Alex Pareto @alexpareto]

@derekstavis @linear Why?

[Alex Pareto @alexpareto]

@vcapretz @brexHQ Welcome to the team!

[Alex Pareto @alexpareto]

@pedroh96 @DOGE @elonmusk Ready to book my flight 😎

[Alex Pareto @alexpareto]

@4eo @shaig Hey Leo - I work on our receipts products here at Brex, happy to help get you set up on all of our auto-receipt features. Just shot you a dm!

[Alex Pareto @alexpareto]

@jamesreggio @hy3na_xyz @brexHQ @hy3na_xyz thanks for the feedback! Got this resolved for you moving forward - sent you a DM with the details

[Alex Pareto @alexpareto]

@j_longfur flow state 😎

[Alex Pareto @alexpareto]

Awesome event hosted by the @WonderVentures team - especially if you’re interested in entrepreneurship and entertainment!

[Wonder Ventures @WonderVentures]

Our own @gormandotcom will be conversing via Zoom w/ some of the most prominent young voices in entrepreneurship & entertainment @BlairImani @AmandaSCGorman @jaylenjbarron & #ZoleeGriggs ! Join us Wed (9/23) 2-3:30pm PST. Free & open to all! Register at https://t.co/Vzdfrruq12.

[Alex Pareto @alexpareto]

@ryancenicola That’s a good one

[Alex Pareto @alexpareto]

Thinking of ways to apply the KonMari Tidying Method of “discarding” to work. Ideas...

- Delete calendar events instead of responding “maybe”
- Archive/inbox zero emails instead of marking as read
- Leave slack channels instead of muting them

Others?

[Alex Pareto @alexpareto]

@tlbtlbtlb I wonder if there could be an “ingredients list” for feed algorithms.

For example, Coca-Cola and Pepsi’s formulas are secret - but we still get the ingredients list and the sugar content.

[Alex Pareto @alexpareto]

@drakerehfeld @GetSpectrum how is there no alternative that’s used to phone call support in 2020

[Alex Pareto @alexpareto]

Super excited to launch what we’ve been building at Brex recently! This just the beginning...

[Henrique Dubugras @hdubugras]

“Inbox Zero” your expenses with Brex! More robust expense management is something we’ve been working on for a while. There is a lot more from where this came from!

[Alex Pareto @alexpareto]

@drakerehfeld @gefencodes @CoupletCoffee @bryce @abhanath I do indeed love coffee

[Alex Pareto @alexpareto]

@drakerehfeld https://t.co/TrdByuXmxl