# Companies can move workflows into an AI improvement loop

Markdown copy of https://realalpha.app/s/companies-can-move-workflows-into-an-ai-improvement-loop/ on Real Alpha, the best minutes from long interviews with the people building AI. Cite and link the page, not this copy.

- Speaker: [Salim Ismail](https://realalpha.app/people/salim-ismail/) (Host) · Founder, Open ExO · Co-wrote Exponential Organizations; was Singularity University's founding executive director; sold Ångströ to Google.
- Host: [Peter Diamandis](https://realalpha.app/people/peter-diamandis/) · Founder, XPRIZE · Co-founded Singularity University and founded Zero-G; co-wrote the books Abundance and Bold.
- Host: [Dave Blundin](https://realalpha.app/people/dave-blundin/) · Founder & GP, Link Ventures · Has co-founded 23 companies; founding CEO of Vestmark and chairman of insurance marketplace EverQuote.
- Host: [Alexander Wissner-Gross](https://realalpha.app/people/alexander-wissner-gross/) · Computer Scientist & Founder, Reified · Has backed and advised 40+ tech companies, directly and through 021T Capital and Reified; writes The Innermost Loop.
- Show: Moonshots
- Episode: [Big Tech’s $3T AI Bet, Claude Writes 90% of Anthropic’s Code, AIs Editing AI Brains | EP #308](https://realalpha.app/v/x09f_BuJsoY/), Nov 4, 2026
- Moment: 49:09 to 56:39 of the episode (8 min), playing on https://realalpha.app/s/companies-can-move-workflows-into-an-ai-improvement-loop/; the episode from there: https://realalpha.app/v/x09f_BuJsoY/?t=2949

## Quote

> “if you're over 50 people, set up this intelligent stack on the side and start **moving workflows over one by one** and start rerunning them.”
> — [Salim Ismail](https://realalpha.app/people/salim-ismail/) at [54:15](https://realalpha.app/v/x09f_BuJsoY/?t=3255) · Moonshots

## Takeaways

1. Anthropic’s internal Claude use grew more than 250-fold.
2. Salim recommends moving workflows over one at a time.
3. Track how quickly a system improves its own capabilities.

## About this segment

Anthropic says Claude wrote over 90% of its new code, while Salim Ismail describes moving company workflows into an AI stack. The shift offers a practical path to continuous process improvement.

From “Big Tech’s $3T AI Bet, Claude Writes 90% of Anthropic’s Code, AIs Editing AI Brains | EP #308”, Nov 4, 2026. [Watch the full episode](https://realalpha.app/v/x09f_BuJsoY/)

## Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

[49:17](https://realalpha.app/v/x09f_BuJsoY/?t=2957) Okay, our next story is a big one. One of our favorite topics on this pod during the singularity, recursive self-improvement, AI able to build the next better AI. So Anthropics IPO filing just made it a line item. So since April, the value of Anthropics internal Claude usage priced at the rates charged to consumers, so I'm sure they're getting their AI at a much lower price because they mark it up, what, 80%, has exceeded monthly cash payroll.

[49:49](https://realalpha.app/v/x09f_BuJsoY/?t=2989) So internal usage grew by more than 250-fold between January and September. And Claude now writes more than 90% of the code. Let me share that image here. I mean, it's crazy. Their payroll is less than their AI usage rates. 250x internal Claude usage over the last, you know, nine months. 90% of September's new code written by Claude. And more money being spent on AI internally than employee salaries.

[50:27](https://realalpha.app/v/x09f_BuJsoY/?t=3027) Your thoughts, gentlemen? Well, so one of the engineers outside my door here, Cameron Crowe at Lightspeed or Lightwave, he spent half a billion tokens on Friday, which kind of triggered my – yeah. And I was like, wow, that's really cool. That's about $5,000 of tokens. That's impressive. Yeah. And so I'll find out tomorrow what he created with his half billion tokens. But I was telling everybody, like, this is great, assuming it was productive, and I'll find out.

[50:53](https://realalpha.app/v/x09f_BuJsoY/?t=3053) But this is absolutely fantastic. So I really wanted to benchmark that against Anthropic, but I can't tell what their payroll is or how many of the 5,300 are engineers. So I'm not quite sure how he is comparing to them. But I suspect it's similar where you ain't seen nothing yet on this chart. 250X in that, what, six months is crazy. But it's going to go up more and more and more because it's just that productive.

[51:19](https://realalpha.app/v/x09f_BuJsoY/?t=3079) And, you know, until they run out of ideas of things to build, which, you know, maybe never with AI helping you with ideas, this should just keep skyrocketing. And then, Sam, in our last pod, we talked about, you know, him wanting to do recursive self-improvement of his data centers using robots. Robots building robots, building data centers, building models, building better robots. It's crazy. Alex, you know, are we at full RSI yet?

[51:47](https://realalpha.app/v/x09f_BuJsoY/?t=3107) I think in some sense, yes and no. I think we're pretty deep in the singularity at this point. Perhaps that goes without saying in this context. But in some sense, I think I maintain space time locally is perfectly smooth. And what we call recursive self-improvement now will look like nothing a year from now. Yes, it's remarkable by historic standards to have a company that's on the verge of an IPO just filing its amended S1, as is the case with Anthropic, basically laying out the case that it's spending more capital and more compute on its AI researchers than it is on its human researchers.

[52:27](https://realalpha.app/v/x09f_BuJsoY/?t=3147) we arrived like if if you're looking for a naive benchmark of have we passed some threshold in terms of RSI as according to capital or operational expenditure? The answer is yes, we're there. We've arrived at this part of the singularity. On the other hand, to the earlier point, there are so many innermost loops that are about to take off inside the present innermost loop, including physical recursion. So we don't yet have the nanobots. We're going to get the nanobots. We're going to get the nanobots building the data centers or the nanobots building copies of each other. And when we have that happening, we'll look back and say, what on earth were we talking about? Like 50% recursive self-improvement just from anthropic diverting compute to AI researchers versus human employees. That was child's play. It's only real RSI when we have the robots or the nanobots

[53:25](https://realalpha.app/v/x09f_BuJsoY/?t=3205) building better copies of themselves. That's the real, you know, it's sort of a no-true Scotsman fallacy. That's the real RSI when you have physical recursion. So yes and no. And of course, it's not just Anthropic. It's Google. It's OpenAI all moving rapidly in this direction. Yeah. Look, we're seeing this already, right? It used to be like 18 months ago, what, 10% of Claude wrote the next version of Claude, or Claude wrote 10% of the next version, and then It was 40%, 60%.

[53:55](https://realalpha.app/v/x09f_BuJsoY/?t=3235) That's why we're seeing such a ridiculous frequency of model release just because of the recursive self-improvement there. We've been tracking it of can you get to recursive self-improvement at the workflow level inside a company? And the minute you hit that, you can basically let the AI run it for you from that point on. What are you telling the companies you coach, Sulim? What are you telling the companies?

[54:16](https://realalpha.app/v/x09f_BuJsoY/?t=3256) We're saying keep your existing company the way it is if it's big enough. If you're below 50 people, brute force it and do it. But if you're over 50 people, set up this intelligent stack on the side and start moving workflows over one by one and start rerunning them. One company has taken 22 steps and shrunk it to four in this new model, right? And then once you have recursive self-improvements whirring away over here, you can essentially let that be for the rest of time because it's just going to continuously improve.

[54:44](https://realalpha.app/v/x09f_BuJsoY/?t=3284) And just start moving workflows over one by one and little by little grow this new thing. So at some point in the next few weeks, as we finish up this pilot, I'll come with hard numbers of what we've seen. And it's staggering what we're seeing companies keep using this. If you're not feeling what Salim is saying right now. Internalize these numbers. The highest and most expensive model will cost you about a buck a day of equivalent human labor.

[55:08](https://realalpha.app/v/x09f_BuJsoY/?t=3308) So the equivalent human being doing that work would cost you, it's $1 instead of what would normally be, you know, a thousand times that roughly. So when Cam has half a billion tokens consumed in a day, he used up about 5,000 people worth of work in a single day. What can you get done with 5,000 people? Now, that's all very exciting, but the trend line, without any improvements, the trend line is coming down 10x year over year, but it'll accelerate from there with other unanticipated improvements, but 10x year over year.

[55:39](https://realalpha.app/v/x09f_BuJsoY/?t=3339) So Cam, a year from today, can have 50,000 people working for him last Friday. What could you do with 50,000 super high IQ level, totally focused people working for him? That becomes a challenge, right? Actually, one of the biggest challenges we're going to have is what do people point their SI at? Because a lot of people, if you ask them that question, they're going to freeze. They're going to say, I have no idea.

[56:03](https://realalpha.app/v/x09f_BuJsoY/?t=3363) Here's two things. Here's the question we're asking to the CEOs we're working with. How quickly does your system increase its own capability? And that forces them into that. And then the benchmark, not to overuse that word, is how quickly and how can you measure rate of improvement of your processes? Because that becomes a benchmark. Not the actual process, but what's the rate of improvement of the benchmark?

[56:28](https://realalpha.app/v/x09f_BuJsoY/?t=3388) This episode is brought to you by Blitzy, autonomous software development with infinite code context.