The Scarce IP Is Leaving the Code Artifact

I keep catching myself staring at a pull request that would have taken me a day five years ago and now arrives in twenty minutes. The code is often fine. Sometimes it is better than fine. What is scarce is not the typing. What is scarce is whether anyone can still say why this shape won, what we refused, and how the next person is supposed to change it without summoning the original author from memory.

That is the shift I care about. Intellectual property is moving off the code artifact.

The first inversion already happened

Ocean Tomo’s Intangible Asset Market Value study now covers fifty years of S&P 500 data. In 1975, tangible assets were about 83% of market value and intangibles about 17%. By the end of 2025 the relationship had inverted: intangibles around 92%, tangibles around 8%.[1] Value migrated from what you can touch to what you can think.

I think we are inside a second inversion, nested inside that first one. Among software intangibles, the valuable thing used to look like the artifact: the repo, the binary, the licensed codebase. AI compresses the cost of producing another artifact. The scarce layer becomes the system around the artifact: the process that decides what to build, the documentation that preserves intent, the operating cadence that keeps production honest, and the people who can still hold the why when the model that wrote the file is gone.

We have seen “process over sealed code” before

IBM did not pay about $34 billion for Red Hat because the Linux sources were locked in a vault.[2] The valuable company was the one that could support, distribute, certify, and operationalize open software for enterprises that needed a throat to choke. The code was visible. The business was the system around the code.

TSMC’s foundry model makes a related cut. Design IP and manufacturing capability can live in different firms. Fabless companies keep the crown-jewel designs. The foundry specializes in making silicon without competing as a product company.[3] Software is drifting toward a rhyme of that split. Generation is getting cheaper. The hard part is specifying what should be generated, verifying it, integrating it, and owning the consequences.

I do not mean “people do not matter.” I mean the opposite. When generation is cheap, judgment, taste, teaching, and accountability get more expensive. Devaluing engineers because an agent can emit functions is a category error. The job moves toward architecture, review, product sense, and the social work of keeping a system coherent.

Conception is becoming the legal mirror of the same idea

In November 2025 the USPTO published revised inventorship guidance for AI-assisted inventions. It rescinded the February 2024 guidance and restated a traditional point: only natural persons invent, AI systems are tools, and conception remains the touchstone.[4] Conception is the formation of a definite and permanent idea of the complete invention, not a vague wish that a model might figure it out.

I am not a patent lawyer, and I am not turning this into legal advice. I am noticing the rhyme. When the office says the human must conceive, it is pointing at the same scarce layer operators already feel: prompts, design docs, ADRs, and review notes are how you show that a person held a specific solution in mind. The artifact can be machine-assisted. The conception story still has to be human.

What I am changing on my machines and in my companies

I write this from Omarchy, which has been my daily driver for about eight months. Not a Mac with a Linux VM for weekends. The machine in front of me. That detail matters only as a reminder that I am willing to change the boring substrate of work when the old defaults stop matching the job.

The job, for me, is no longer “protect the precious lines.” The lines will be rewritten. The job is:

1. Write the decision while it is cheap. If an agent drafted the change, the human still owes a short record of intent: what problem, what options, why this one.
2. Teach the system, not only the syntax. Raise people into ownership of interfaces, failure modes, and customer reality. Typing speed was never the scarce talent.
3. Measure survival, not emission. Merges are easy to count. Escaped defects, time-to-understand, and whether a teammate can modify the module next month are harder and more honest.
4. Keep dignity in the loop. AI can clear ceremony. It should not become an excuse to treat engineers as expensive keyboards waiting to be removed.

This is a sequel in spirit to an argument we made on the Wawandco blog: shipping code is not the job.[5] That piece was about partnership and delivery quality. This one is narrower. The artifact is becoming abundant. Documented systems, processes, and decisions are becoming the IP. People who can hold that layer are not a cost center to shrink. They are the compounding asset.

If your org still celebrates who typed the most, you are optimizing for a world AI already discounted. Celebrate who made the system clearer. That is what remains scarce.

References

1. Ocean Tomo, “Ocean Tomo Releases 2025 Intangible Asset Market Value Study Results,” February 2026. https://oceantomo.com/insights/ocean-tomo-releases-2025-intangible-asset-market-value-study-results/

2. Red Hat / IBM, “IBM Closes Landmark Acquisition of Red Hat for $34 Billion,” July 9, 2019. https://www.redhat.com/en/about/press-releases/ibm-closes-landmark-acquisition-red-hat-34-billion-defines-open-hybrid-cloud-future

3. Wikipedia and industry histories of the semiconductor foundry model; TSMC as pure-play foundry separating design from manufacturing. https://en.wikipedia.org/wiki/Foundry_model

4. U.S. Patent and Trademark Office, “Revised Inventorship Guidance for AI-Assisted Inventions,” Federal Register, November 28, 2025. https://www.federalregister.gov/documents/2025/11/28/2025-21457/revised-inventorship-guidance-for-ai-assisted-inventions

5. Wawandco, “Shipping Code Is Not the Job,” August 28, 2026. https://wawand.co/blog/posts/shipping-code-is-not-the-job/