Are AI Labor Replacements a TAM Expansion Vector?

Every software founder is positioning their AI agent as a labor replacement with enormous TAM potential. How realistic is the pitch?

I had a few hours on my flight to SF, so I threw together some thoughts around the most common theme we’ve seen over the last several months (even more common than young founders trying to channel Henry Kravis). I’m again experimenting with shorter pieces and cross-posting to X. Please reach out with feedback, especially if you’re building something; my email is below.

In 2023, nearly every software pitch was some “Copilot for X.” In the spring, on the heels of the Devin / Cognition launch, every founder began pitching “agents.”

Now, as a natural extension of the agent wave, founders are pitching agentic labor replacements, arguing that because their product will capture customers’ labor spend – not SaaS spend – their TAM is enormous.

“We’re going to sell a labor replacement” is an especially effective pitch for vertical software companies. These benefit from a much more coherent narrative around moat trajectory compared to horizontal AI startups (oligopolistic market structures, domain expertise, proprietary data advantages, lower levels of competition) but struggle with TAM considerations. Picking something unique and defensible is often at odds with its potential to become very large (the Fractal Software problem).

There’s real reason to be excited about selling agents instead of tools as a TAM expansion vector. Software spend for large F500 companies represents maybe a couple percentage points of revenue. Labor spend is at least an order of magnitude larger.

With every founder making the TAM expansion pitch, VCs need to thoughtfully interrogate how realistic it is. One of the most attractive opportunities for software VCs today is to identify a business with credible TAM expansion potential that everyone else passes on due to conventional market size concerns. My current view is that five things matter when it comes to driving venture-scale returns.

Models must actually be well-suited to automate work

While it seems obvious, companies should attack markets not only where there’s high levels of labor spend, but also where that spend is on people whose work can actually be automated by language / voice models. An “agent for pool cleaning is not going to make VCs any money

Labor market shortages should create a compelling pain point

The best opportunities will exist in markets characterized by persistent labor shortages. Here, customers whose growth is constrained by their inability to hire people will gladly adopt AI-enabled labor replacements. Companies building for these customers will benefit from attractive sales cycles due to strong customer pull and will likely succeed in charging premium rates for their software.

Build for customers with a BPO muscle

Because buying a labor replacement is akin to outsourcing work to a BPO, companies may (somewhat counterintuitively) find more success selling to customers conditioned to hiring BPOs vs. those conditioned to buying software tools for internal teams. The procurement motion will feel less foreign, and companies will need to do less expensive and time-consuming “re-education” work on behalf of their customers.

Data spread across disparate systems is a huge advantage

If customer data is primarily stored in one dominant system of record, that incumbent vendor has an enormous advantage in building an agentic “system of intelligence.”

If instead customer data is spread across disparate systems, a new vendor can position themselves as “Switzerland,” able to integrate with and work across all data sources rather than trying to make use of data within a walled garden. This is part of the bull case for a company like Glean.

The initial agent product must be well-positioned

Most agentic labor replacements will start as point solutions. It’s critical that the initial agent be situated within a workflow that is “top of funnel” for customers and / or has access to lots of structured and unstructured data. This gives a vendor the “right to win” additional workflows and affords it valuable data to build the highest-quality agents, both of which are critical to offering a platform product (rather than a simple point solution) with venture-scale potential.

This is part of our thesis for Vooma: order entry and quoting not only sits near the beginning of a freight broker’s workflow, but also allows a broker to capture 20-25 critical data points that can power agents for downstream tasks.

If you have questions, comments, or feedback, please reach out: andrewziperski [at] gmail [dot] com.

The views expressed herein are the author’s own and are not the views of Craft Ventures Management, LP or its affiliates.