The Pinhole View of AI Value

I’ve been listening to 20VC since it appeared. When I want to understand how venture capitalists are thinking, Harry Stebbings is my go-to source. He asks the questions I’d want to ask, and he’s not afraid to push back on his guests. I’ve learned a lot.

However.

In recent episodes, Harry made a claim that the path to AI profitability runs through labor replacement. “Stop paying $1M in salaries by paying $100K for this AI-based service.” Simple. Clean. Measurable.

And woefully incomplete.

Labor replacement looks at value creation through a tiny pinhole. Yes, cost reduction is one way AI creates value. But it’s not the only way, and I’m not even sure it’s the most important way. Focusing only on headcount reduction is like saying the only value of a car is that you don’t have to pay for a horse.

The NPV Framework

If we momentarily take the naive view that a company’s value equals its net present value—the sum of all future cash flows discounted to today—then there are four fundamental levers:

1. Smaller costs at the same time

2. Same costs but later

3. More revenue at the same time

4. Same revenue but sooner

Harry’s labor-replacement thesis focuses on the first lever alone. That ignores many sources of value.

And even NPV doesn’t capture the full picture. Company value also encodes optionality—how many ways exist to improve NPV in the future. A company with more options is worth more than one with the same cash flows but fewer options. (If you haven’t read my material on options and software, the short version is: flexibility has value, especially in uncertain environments. And what environment is more uncertain than one being transformed by AI?)

Let me give you examples of each strategy.

Higher Revenue (Same Timeline)

Expanded service capacity. Your support team of 10 people can now handle the inquiry volume that used to require 25. But here’s the thing—you don’t fire 15 people. You serve three times as many customers. Your addressable market just tripled without tripling your headcount. The humans are still there; they’re just handling the hard cases while AI handles the routine ones.

Personalization at scale. A human salesperson can deeply understand say 50 accounts. An AI-augmented salesperson can maintain genuine, contextual on several times that many. The genie remembers that this customer’s CFO cares about security compliance and that customer’s CTO is skeptical of vendor lock-in. Higher conversion, higher retention, higher revenue per rep.

Previously impossible features. Some product capabilities simply weren’t feasible before. Real-time translation. Intelligent search across unstructured data. Automated analysis of documents that would take humans hours. These aren’t cost savings—they’re new value propositions that customers will pay for.

Earlier Revenue (Same Amount)

Faster time to market. If your development team can ship features in two weeks that used to take six, you start earning revenue on those features four weeks earlier. That’s not a cost reduction. That’s the same revenue arriving sooner—which, thanks to the time value of money, is worth more.

Accelerated sales cycles. AI can generate proposals, customize demos, answer technical questions, and handle objections while the human salesperson is asleep. A deal that used to take 90 days now closes in 60. Same deal size, but you’re earning and compounding that revenue a month earlier.

Compressed customer onboarding. New customers who used to take three months to reach full productivity now get there in one. They start generating the usage (and the fees) that justify the relationship two months sooner. Meanwhile, your customer success team can take on the next cohort.

Costs Later (Same Amount)

This one’s subtle, but it’s real.

Deferred hiring. Your current team is handling growth that would normally require two new hires. You’ll probably still make those hires eventually—but six months from now instead of today. Those six months of salary stay in your pocket, earning interest, available for other investments.

Delayed infrastructure. Better optimization and more efficient resource usage means you can push that major infrastructure investment into next year’s budget instead of this year’s. Even if the cost is the same it arrives later.

Extended training runway. AI-assisted onboarding means new employees become productive faster with less senior-employee time. The training cost is spread out, and some of it shifts from expensive human time today to cheaper AI time tomorrow.

Optionality

Here’s where it gets interesting.

New markets become accessible. Real-time translation and localization used to require a dedicated team for each market. Now you can experiment with entering new geographies without committing the full resources. The option to expand exists where it didn’t before. Even if you never exercise that option, its existence has value.

New business models emerge. A professional services firm that couldn’t scale because every engagement required senior talent can now productize some of that expertise. A company that couldn’t offer a lower price tier because the unit economics didn’t work can now create an AI-assisted self-service option. These aren’t cost reductions—they’re entirely new ways to make money.

Faster experimentation. If you can prototype, test, and iterate three times faster, you can run three times as many experiments. Most experiments fail, of course. But the ones that succeed create options you wouldn’t have discovered otherwise. The ability to try more things is itself valuable.

The Bigger Picture

I don’t want to be too hard on Harry here. The labor-replacement story is compelling because it’s legible. You can point at a budget line, point at an AI service, and do arithmetic. “We paid X, now we pay Y, the difference is Z.” Clean. Fundable.

But the other sources of value are just as real, even if they’re harder to measure. More revenue, earlier revenue, costs delayed, options created—these are all ways AI can make a company more valuable.

Software design is an exercise in human relationships. So is AI adoption. The question isn’t just “how do we do the same work with fewer people?” It’s “what can these people do now that they couldn’t do before? What becomes possible that wasn’t possible? What options do we create?”

The pinhole view sees one thing clearly. But there’s a whole landscape out there.

This post is sponsored by Resolve AI.

AI has changed how software is built. With Resolve AI, customers like Coinbase, MongoDB, MSCI, Salesforce, and Zscaler are changing how software is run.

Resolve AI just announced its $125M Series A to accelerate how they deliver AI for prod—a first-class, programmable interface that enables engineering teams to operate fluently without being bottlenecked by tools, expertise, or context.
Learn more →