Posts

Every portfolio manager knows the efficient frontier - the set of optimal portfolios offering maximum returns for given risk levels. What if AI prompts had their own efficient frontier?

As we all start to use AI, prompt optimization will be a consistent challenge. GEPA, GEnerative PAreto, is a technique to discover the equivalent efficient frontier for AI.

Reading the paper, I noticed the initial results were promising, with a 10-point improvement on certain benchmarks & a 9.2 times shorter prompt length. Shorter prompt length, & we all know that input prompts are the biggest driver of cost (see The Hungry, Hungry AI Model). So, I implemented GEPA in EvoBlog.

How long & how quickly can a business compound ?

This is a question every investor asks of every business, public or private.

In the 2010s, Slack & Atlassian became titans. On the day Salesforce announced its intent to acquire Slack, it was equally valuable to Atlassian at ~$27b.

The revenue curves look similar in the out years, similar growth rates. Atlassian continues to compound at massive scale.

OpenAI hit $12 billion ARR within five years of ChatGPT’s launch ^[1] . Anthropic reached $200 million in revenue in January 2024 ^[2] . Meanwhile, Salesforce took ten years to reach $1 billion ARR ^[3] .

We build teams in pyramids today. One leader, several managers, many individual contributors.

In the AI world, what team configuration makes the most sense? Here are some alternatives :

First, the short pyramid. Managers become agent managers. The work performed by individual contributors of yore becomes the workloads of agents. Everyone moves up a level of abstraction in work.

One trillion tokens per day. Is that a lot?

“And when we look narrowly at just the number of tokens served by Foundry APIs, we processed over 100t tokens this quarter, up 5x year over year, including a record 50t tokens last month alone.”

In April, Microsoft shared a statistic, revealing their Foundry product is processing about 1.7t tokens per month.

Yesterday, Vipul shared Together.ai is processing 2t of open-source inference daily.

Now that we’ve compressed nearly all human knowledge into large language models, the next frontier is tool calling. Chaining together different AI tools enables automation. The shift from thinking to doing represents the real breakthrough in AI utility.

I’ve built more than 100 tools for myself, & they work most of the time, but not all the time. I’m not alone. Anthropic’s Economic Index report reveals that 77% of business use of Claude centers on full-task automation, not co-piloting.

Like breakfast at a diner, software is cheap & fast to make. Ask for a new task management tool & you’ll have the first version in less time & for less money than an omelette.

AI-built tools may not last long. Some survive only a few minutes, long enough to answer “What’s our turnaround time this week?”

Understanding technology is at the core of how we research new theories. As part of that effort, Theory is starting a developer relations team.

Developer relations for venture capital? What does that mean?

Internally, we’ve built hundreds of agents, we’ve struggled to debug tool calls, tested large action models, & work through AI every day. We’ve hosted events on how to build with AI.

MCP enables an AI to understand which tools rest at its disposal : web search, file editing, & email drafting for example.

Ten months later, we analyzed 200 MCP tools to understand which categories developers actually use.

$4b in revenue run-rate. There are two data giants both now at this mark after Databricks announced it surpassed the threshold.

This is an opportunity to compare the two leading data companies at a revenue intersection.

Head-to-Head Comparison

Both are at $4b in revenue. Each claims over 650 customers paying $1M+ annually. Each boasts strong net dollar retention (140% vs 125%).