Thesis
Software Is Becoming a Commodity. Agent Labor Isn't.
The software was the asset for two decades. That bet is eroding, and the value is moving to the work the software performs.
For two decades the bet was simple: write software once, sell it many times, and let the margins compound. The software was the asset. That bet is eroding, not because software stopped being useful, but because it stopped being hard to produce. When a competent model can generate a working app in an afternoon, the app is not the moat. It is the table stakes.
The commodity trend
Code generation got good enough that the marginal cost of a feature is approaching the cost of describing it. That is wonderful for builders and brutal for anyone whose defensibility was "we built the software." If the thing you sell can be cloned over a weekend, the price of that thing trends toward the cost of cloning it, which is now close to zero.
Labor is the scarce thing
Here is what did not get commoditized: actually doing the work, reliably, over and over, without dropping it. A tool that can do a task once in a demo is common. A system you can hand a recurring responsibility to and trust to own it is rare. The scarcity moved from the software to the labor the software performs.
This is why "AI agent" and "AI app" are not the same category. An app is a surface you operate. An agent is labor you delegate to. You buy software to use it. You hire labor to offload something. The second relationship is far more valuable and far harder to build.
What changes when you buy outcomes
Once you frame it as labor, the questions change. You stop asking "what features does it have" and start asking "what can I stop doing myself." You stop measuring it in tokens or seats and start measuring it in work completed. The unit is the outcome, not the interaction.
Companies do not buy software because they want software. They buy it because they want the work the software does. Agents collapse the distance between the two.
The lean company
The downstream effect is a different shape of company. If labor is something you can buy as agents instead of headcount, the founder-to-output ratio changes by an order of magnitude. A handful of people, or one person, can run an operation that used to require a department, because the recurring work is owned by a team of agents that close their own loops.
That is the company Nerve is built for and built as. Not software you log into. A team of agents that does the work.
FAQ
Are you saying software has no value?
No. Software is more useful than ever. The point is that the software itself is no longer scarce, so durable value moves to reliably performing the work, which is much harder to replicate.
How do you price agent labor versus software?
You price it against the work it replaces or the outcome it produces, not against token cost or per-seat licensing. The buyer is comparing it to getting the job done, not to another app.