# Longer AI work horizons could shift companies toward outcome delegation

Markdown copy of https://realalpha.app/s/longer-ai-work-horizons-could-shift-companies-toward-outcome/ 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/) · Chairman, OpenExO · 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/) · Founder, Reified · Has backed and advised 40+ tech companies, directly and through 021T Capital and Reified; writes The Innermost Loop.
- Show: Moonshots
- Episode: [Jensen’s Spy Probe, Anthropic’s Debut Week, and Elon’s 30M Robot Workforce | #310](https://realalpha.app/v/ILQwMvCA2_o/), Nov 11, 2026
- Moment: 43:18 to 50:58 of the episode (8 min), playing on https://realalpha.app/s/longer-ai-work-horizons-could-shift-companies-toward-outcome/; the episode from there: https://realalpha.app/v/ILQwMvCA2_o/?t=2598

## Quote

> “The thing I got from this was this is – we're getting to the point where we can **delegate outcomes rather than tasks**, right?”
> — [Salim Ismail](https://realalpha.app/people/salim-ismail/) at [47:01](https://realalpha.app/v/ILQwMvCA2_o/?t=2821) · Moonshots

## Takeaways

1. AI autonomy time horizons are presented as growing rapidly.
2. Companies may delegate objectives instead of individual tasks.
3. Scarce access to top models could become a pricing strategy.

## About this segment

The hosts discuss models taking on longer tasks and the prospect of delegating outcomes, not just instructions. They also consider how frontier labs might ration their most capable models.

From “Jensen’s Spy Probe, Anthropic’s Debut Week, and Elon’s 30M Robot Workforce | #310”, Nov 11, 2026. [Watch the full episode](https://realalpha.app/v/ILQwMvCA2_o/)

## Transcript

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

[43:23](https://realalpha.app/v/ILQwMvCA2_o/?t=2603) So an analyst at Nikola J2030 looked at Meter's new report on Anthropic and estimated that AI's, quote, time horizon, how long a model can reliably work on finishing a task, is doubling every two months. He put Anthropic's internal Model 4 at about 240 hours. That's a 10-day project by a human expert. Meter itself, which has deep access to Anthropic, found that from October 2025 through October 2026, capabilities there ran about twice as fast as the year before.

[43:59](https://realalpha.app/v/ILQwMvCA2_o/?t=2639) So I'm showing the chart here, Alex. Can you give us an understanding of what this means and how important it is? Sure. So first, some definitions. This is a chart of AECI, which is the Anthropic, Anthropic's version of the EPIC AI Capabilities Index. So this is a broad basket of lots of other benchmarks that attempts to model autonomy time horizons, how long models can perform to some percentage success rate a given task that would be in the human equivalent.

[44:30](https://realalpha.app/v/ILQwMvCA2_o/?t=2670) So 100 hours could mean 50-plus percent chance or 80-plus percent chance at performing a task that would require a human equivalent 100 hours. And what we're seeing is multiple eras of autonomy growth. growth first. And by the way, the vertical axis for those not looking is already logarithmic. So this, to the extent this is bending upwards, this is super or hyper exponential growth. We see a regime from GPT-2 to GPT-3, we see a regime from GPT-2 to GPT-3, we see a regime ETE call it 4.5 at one slope.

[45:04](https://realalpha.app/v/ILQwMvCA2_o/?t=2704) And then we see the next stage going up to opus 4.5 at a higher. So this is super exponential growth already. And then we get to the present regime that we're in, where there are these mysterious internal anthropic models that anthropic keeps talking about, like model four, that don't seem to correspond to generally available models. The brumant is that anthropic has a bunch of internal models, like model four, that they're keeping internally purely for the benefit of training models that get generally released.

[45:34](https://realalpha.app/v/ILQwMvCA2_o/?t=2734) So to the extent that's the case, it suggests several things. One, it suggests that the scenario that we've talked about on the pod in the past, which is the frontier labs are incentivized to take their most capable models and not make them generally available and use them for their highest revenue per token goal, which is recursive self-improvement, training other better models. to the extent that's the case, that would appear to be the case here. Secondly, if you just extrapolate these hyper-exponential curves, 2027, as I was suggesting earlier, again, I don't actually think there is a singularity, a literal mathematical singularity within the metaphorical singularity. I think it's an optical illusion. But if one were to go looking for mathematical singularities based on trends in autonomy time horizons like this, I think 2027 is probably

[46:28](https://realalpha.app/v/ILQwMvCA2_o/?t=2788) where I would look for exhibit A for a mathematical singularity in terms of autonomy time horizons. Translation, the game is afoot over the next 18 to 24 months. Yeah, we're in the depths of the singularity. Takeoff, you're looking for hard takeoff, look no further than 2027. Salim? Two or three things. First, I thought that this was – it's clear where I think when we talk about this, we're clearly in the singularity because there's no model by which you can accurately predict the future right now.

[46:59](https://realalpha.app/v/ILQwMvCA2_o/?t=2819) This is like complete breakdown, so we have to reinvent every predictable model. The thing I got from this was this is – we're getting to the point where we can delegate outcomes rather than tasks, right? And that becomes really, really interesting. Well, you can say to now, you can just say, achieve this objective. And right now we're delegating tasks to AI. We're saying, go process that email list.

[47:27](https://realalpha.app/v/ILQwMvCA2_o/?t=2847) Go figure out my calendar. Go do very, very. specific tasks, right? Now you can see more kind of abstract things like optimize my calendar for 20% rest, 40% podcast, 60%, whatever, right? And so there's some really interesting kind of guidance you can give it. The bigger picture for me as I was thinking about this, AI's time horizon is going to now go faster than the time between board meetings, right? So who's really running the company so you're going to get things done so quickly slim give me a really good concrete example just hammer it into into 5.2 right now and it'll it'll prove your point okay um optimize my calendar for come on that's boring get us to another star or something you know i'm doing that We know Spermi Explorer mission on this pod.

[48:24](https://realalpha.app/v/ILQwMvCA2_o/?t=2904) Designing you a life form for me that is able to get rid of all the mold in my apartment. Here's one. Optimize my health protocol and wellness protocols so that I can do various things, including drink a couple of glasses of wine when I want to, plus keep my cholesterol down. Oh, my goodness. Salim, what kind of massive transformative purpose is this, Dave? I'll give you one. Hold on, hold on, hold on.

[48:56](https://realalpha.app/v/ILQwMvCA2_o/?t=2936) I think I'm going to go back to this idea that as AI and eventually robots do more and more doing, we get to spend so much more time being. And I think we need to get way better at that side of it. I'll give you a better one, Dave. This is going to run on MaxFast, CloudFable 5.2, one million contacts MaxFast. You cannot spend more money. Maximum money. Cure the top 5,000 diseases, period. Cure. Okay, and while it's calculating that, Imad, let's go to your comment here.

[49:27](https://realalpha.app/v/ILQwMvCA2_o/?t=2967) You were going to say. I think once you get to 100 hours, that's everything. Like, what does it mean? Like, you just reset it to go more than 100 hours, right? Like, you reload it inside. So I think basically you've reached the critical mass. But I just thought of the bull case for Anthropic, which is right now, you know, Dave is spending $20,000, $100,000 up and down and left and right, the bull case is that they're going to go to all the major corporates and say, here's Model 4. You can only get it if you do a one-year upfront commitment. Yeah.

[50:00](https://realalpha.app/v/ILQwMvCA2_o/?t=3000) They'll spike revenues. I love that. Didn't see that coming. You're right. What do we do then? Just because you couldn't have access to GPUs, they'll basically say, these latest models, we only have a limited number of tokens. And then what can they charge for those tokens? Anything. I think so. I mean, they already do that. Like, all the top frontier leads, OpenAI and Anthropic, will already strike enterprise agreements with large purchasers of tokens to do pre-purchase commitments.

[50:29](https://realalpha.app/v/ILQwMvCA2_o/?t=3029) Yeah, but this is different. This is cutting off the other path. That's the scary part. You can approximate one with the other. Again, I think the market wants to clear, the flops want to flow, and you can always raise prices or lower capabilities per unit price for everyone who's not prepaying. So here's my question. They haven't lowered it yet.