Software strategy and IP
Is software IP dead in the age of AI?
AI is making code abundant. The strategic question is what still makes enterprise software proprietary, defensible, and worth building.
By Kesho Partners
10 minute read
Software's old scarcity is collapsing. AI can generate interfaces, features, and working code faster than many companies can approve a procurement request. That does not make software IP dead. It makes undifferentiated code a weaker asset. Defensible value now sits in provable ownership, proprietary workflows, governed data, technical architecture, customer knowledge, and the ability to operate software reliably at enterprise scale.
Free tool for investors
Portfolio AI Readiness Score
See which portfolio companies can win with AI and which are exposed, with a clear next step for each.
No. But code alone is a weaker moat.
Copyright can protect original human-authored code, but it does not give one company ownership of a software function or business idea. A competitor may build similar functionality without copying protected expression or infringing another right.
AI lowers the cost of implementation. It does not create exclusive data rights, deep customer workflows, sound architecture, security evidence, trade secrets, trusted distribution, or dependable production operations.
The response is not to stop building software. It is to build software that becomes a genuine enterprise asset: clearly owned, traceable, embedded in valuable work, difficult to replace, and capable of producing measurable outcomes.
Who owns code produced with AI?
Sometimes, but not automatically. AI assistance does not prevent ownership of the human-authored parts of software. It can make the boundary between human and machine contribution harder to prove.
The UK government states that code may receive copyright protection where it expresses human creativity, while software functionality itself is not protected by copyright. The U.S. Copyright Office also requires human authorship. It says prompts alone generally provide insufficient control, although original human selection, arrangement, and modification may qualify.
The commercial answer therefore depends on evidence: who designed the system, who selected and changed generated output, whether those people assigned their rights, and whether third-party terms restrict use or transfer.
What intellectual property can protect software?
Software is rarely protected by one right. Buyers should test the complete IP and contractual position.
| Protection | What it may protect | What to verify |
|---|---|---|
| Copyright | Original human-authored code, documentation, and some interface expression | Human contribution, authorship records, assignments, and jurisdiction |
| Contract | Ownership allocation, licences, warranties, restrictions, and transfer rights | Employment terms, contractor assignments, customer terms, and AI vendor terms |
| Open-source licences | Permission to use third-party code subject to licence conditions | Dependencies, notices, source-disclosure duties, modifications, and distribution model |
| Trade secrets | Confidential source code, methods, prompts, data processes, and know-how | Access controls, confidentiality terms, disclosure history, and secrecy procedures |
| Patents | Qualifying technical inventions, subject to jurisdiction and examination | Ownership, filing status, claims, validity risk, and freedom to operate |
| Database and data rights | Certain databases, licensed datasets, and contractual data access | Source, consent, permitted purpose, transfer rights, retention, and territory |
| Trade marks | Product and company identifiers | Registration, ownership, territory, conflicts, and transfer |
Functionality, ownership, and freedom to operate are different questions. Owning copyright in source code does not by itself prevent others from building similar functionality or prove that the product does not infringe third-party rights.
What does AI change?
AI increases provenance risk. More code can enter a product without a clear record of its source, licence, or human contribution.
AI increases contract risk. Tool terms may differ for inputs, outputs, model training, indemnities, confidentiality, and enterprise use.
AI increases valuation risk. A large codebase is not automatically a large proprietary asset if material parts lack protectable human authorship, rely on restrictive licences, or can be reproduced from public functionality.
What evidence should a buyer request?
Request evidence, not a general warranty that the company owns its IP.
| Evidence | What it proves |
|---|---|
| Employee and contractor IP assignments | The company received rights from the people who created the software |
| Repository history and named approvals | Human design, selection, modification, review, and release decisions are traceable |
| AI tool and model register | The company knows which systems materially generated or transformed code |
| AI vendor terms in force at the time | Output use, training, confidentiality, warranties, and indemnities were understood |
| Software bill of materials and licence scan | Third-party components and obligations are identified |
| Input and reference-material record | Prompts did not improperly include customer data, confidential code, or unlicensed material |
| Patent and trade-secret register | Claimed inventions and confidential know-how are defined and controlled |
| Claims, disputes, and takedown history | Known challenges to ownership or permitted use are visible |
Sources: [3]
What are the main red flags?
Treat any of these findings as a reason to investigate ownership before approving the transaction.
- No contractor assignments
- Payment for development work does not always transfer all intellectual-property rights.
- No record of AI tools
- The company cannot reconstruct which terms, models, or data-handling rules applied.
- Unknown dependencies
- The repository lacks a current software bill of materials or reliable licence review.
- Copied reference code
- Teams prompted with third-party source code or reproduced material without documenting permission.
- Prompts as the only authorship evidence
- There is no traceable human selection, modification, architecture, or review.
- Trade secrets without controls
- The company calls information confidential but cannot show access restrictions or confidentiality obligations.
How should the board decide?
Approve the IP position only when the company can identify what it owns, what it licenses, what it must keep confidential, and what may prevent use or transfer.
Reduce the valuation, require remediation, seek specific warranties or indemnities, or walk away when a material asset depends on unverified authorship, missing assignments, incompatible licences, restricted data, or unresolved claims.
Do not value AI-generated code by volume. Value the rights the company can prove, the know-how it can lawfully protect, and the product it can continue to operate and transfer.
This is a due-diligence framework, not legal advice. Copyright, patent, trade-secret, contract, database, and open-source questions require qualified counsel in the relevant jurisdictions.
Related service
Technical due diligence for software and AI
Kesho helps investors and acquirers test software defensibility, ownership, security, architecture, operations, data, and the evidence behind management claims.
Explore the service