Vibe Coding and the Limits of Copyright

AI now writes the code. Anyone with an idea and a prompt can build an app.

But code an AI wrote may be hard to own, and the real costs move downstream — where Portugal and the Greater Bay Area part ways.

While AI democratises the writing of code, only supportive ecosystems and adaptable IP rules can democratise the building of a livelihood.

Vibe coding lets anyone build software by describing it in plain language. In April 2026, Google said 75% of its new code was AI-generated.

Software-as-a-Service companies lost around $300 billion in market value within 48 hours: the “SaaSpocalypse”.

But the cost did not disappear — it moved downstream. One experiment found 26% more tasks completed with AI; another found experienced developers 19% slower, and industry data reported a 441% increase in review time.

In Portugal, copyright needs a human author’s creative choices: purely AI-generated code may remain unprotected. Functionality can be copied, the interface protects only the surface, and AI services that retain prompts and code can undermine trade secrets.

Add copyleft licences, Cyber Resilience Act documentation and contract-dependent platform terms, and these hurdles can constrain scalable entrepreneurship.

Shenzhen subsidises exactly those downstream costs: compute, model and data vouchers, half-day registration and shared compliance services.

The national registry now restricts AI-written applications — but in Hong Kong, s.11(3) may make the person who made the arrangements necessary the author, for 50 years.

Three fixes: a compliance sandbox and free SBOM scanner for Portugal; an AI-assisted registration track at the CPCC; a cross-border GBA OPC IP guide.

While AI democratises the writing of code, only supportive ecosystems and adaptable IP rules can democratise the building of a livelihood.

1Introduction

Vibe coding is the use of AI, prompted in natural language, to write computer code — a term attributed to Andrej Karpathy in February 2025 and Collins Dictionary’s Word of the Year. In April 2026, Google said “75% of all new code at Google is now AI-generated”.

Financial analysts called it the “SaaSpocalypse”.

−$300 billion in market value, within 48 hours.

Yet the evidence is a context-dependent productivity paradox. One field experiment found 26% more tasks completed with AI tools; a 2025 study of experienced developers found a 19% slowdown on specific tasks; industry telemetry reported a 441% increase in pull-request review time. Building a prototype is cheaper than ever; verifying, securing and maintaining a commercial product is not.

So: can vibe coding help people without coding skills or capital climb the social ladder — and do the IP rules and ecosystems of Portugal and the Greater Bay Area help or hold them back?

2Portugal

Portugal has 5,091 active start-ups, but youth unemployment stands at 20.22% and nearly 30% of Portuguese-born people aged 15–39 live abroad. Some graduates may be pushed toward necessity self-employment — and when a vibe-coded app hits the debugging wall, local developers are scarce: ICT vacancy tightness is 288% higher than for the average Portuguese occupation.

A computer program is protected only if it is the author’s own intellectual creation, reflecting free and creative choices (Infopaq). A general idea is not protected, and purely AI-generated code without human creative contribution may remain unprotected. Google’s engineers review and integrate AI code; for a solo founder, uncertain copyright complicates stopping copying, granting exclusive licences and investor due diligence.

The workarounds are partial.

Functionality cannot be copyrighted (SAS Institute): competitors may observe, study and test a vibe coder’s platform to replicate its exact functional workflows.

An interface can be protected if it reflects creative choices (BSA, Cofemel) — but competitors can easily alter the layout while copying the tool.

Trade secrets require ‘reasonable steps’ to keep information confidential — which may be undermined if the AI service retains or discloses confidential prompts and code.

Downstream, AI models output open-source code without its licence notice; a copyleft snippet can trigger source-code disclosure obligations. From December 2027 the Cyber Resilience Act brings SBOM-related requirements for products in its scope. Indemnities, liability and API prices vary by provider, contract and tier.

In Portugal, these combined hurdles can constrain scalable entrepreneurship and its potential contribution to social mobility.

Same coder. Same app. A different system.

3The Greater Bay Area

Nine Guangdong cities plus Hong Kong and Macao: 86 million people, three legal jurisdictions, and R&D spending from 6.7% of GDP in Shenzhen to 0.31% in Macao.

Shenzhen’s 2026–2027 OPC Action Plan is built for the ‘super individual’ (超级个体): one founder whose AI agents handle coding, design and operations.

And it pays for exactly the downstream costs: up to RMB 10 million in Computing Power Vouchers (算力券), RMB 2 million in Model Vouchers (模型券) and RMB 2 million in Corpus Vouchers (语料券) — plus half-day registration and a shared open-source compliance platform.

Courts are flexible on AI art: in Li v Liu the Beijing Internet Court granted copyright to an AI image because 150+ prompts reflected human intellectual achievement.

But since March 2026 the CPCC requires a declaration that no AI wrote the code. The paradox: the city promotes AI-assisted development, while the national registration procedure restricts such declarations.

Two ways out. Shenzhen courts strongly protect source code as a trade secret — and subsidised local models may help keep it in-house.

And Hong Kong: under Section 11(3) of Cap 528, the author of a computer-generated work may be ‘the person by whom the arrangements necessary for the creation of the work are undertaken’ — for 50 years.

4Comparison

Prototyping is equally accessible in Lisbon and Shenzhen. Stage 2 — legal and commercial capability — diverges sharply.

Five structural factors help explain the divergence.

Rules: Portugal follows the European droit d’auteur tradition, which requires human creative choices. The GBA pairs Mainland civil law with Hong Kong’s common law, which expressly addresses some computer-generated works.

Practice: EU safety and transparency rules can bring compliance costs for solo founders. Shenzhen governs through industrial policy: half-day registration, free 3m² desks and compute vouchers.

Interests: With 20.22% youth unemployment, self-employment in Portugal may provide an alternative livelihood. Shenzhen backs 1,000+ AI OPCs to absorb graduates and boost local cloud and hardware demand.

History: Portugal’s start-up scene grew around software services. Shenzhen grew from a 1980 Special Economic Zone into a global innovation cluster tied to hardware supply chains.

The World Outside: Portugal depends on US cloud platforms. The GBA invests in domestic compute and open-weight models, and uses Hong Kong’s common law system as a bridge to global markets.

5Policy

Portugal & the EU: A two-year support sandbox helping solo founders and micro-enterprises meet Cyber Resilience Act requirements, and a free public scanner that checks AI code for copyleft licences and generates SBOMs.

NCAC & CPCC: An “AI-Assisted Software Registration Track”, where founders declare the AI tools and prompts used and seek registration for their original human-designed architecture, workflow and interface.

Shenzhen & Hong Kong: A cross-border “GBA OPC IP Guide” through Qianhai, with templates to document authorship and ownership, structure licensing and understand Hong Kong’s s.11(3) rules.

6Conclusion

Vibe coding clearly democratises software prototyping, but cheaper code does not automatically yield sustainable businesses. The bottlenecks shift downstream — to verification and IP compliance — and there, IP frameworks and ecosystems make the decisive difference. While AI democratises the writing of code, only supportive ecosystems and adaptable IP rules can democratise the building of a livelihood.

Abstract

Shenzhen’s 2026–2027 plan for AI “one-person companies” presents AI-enabled entrepreneurship as a route to innovation and commercialisation. Can vibe coding democratise access to the digital economy, enabling people without formal programming skills or substantial start-up capital to build products, create livelihoods and open new paths to social mobility? This potential also raises important IP challenges: beyond copyright ownership, how do code provenance, third-party code and open-source licences, API dependencies, trade secrets and AI platform terms affect the ability to commercialise and scale AI-generated software? Through a comparison of the GBA and Portugal, this paper will examine whether IP frameworks and entrepreneurial ecosystems can enable this new form of business creation while widening access to economic opportunity.

Keywords: Vibe coding; AI-generated software copyright; Social mobility; One-person companies (OPC); Greater Bay Area

1Introduction

Vibe coding is the use of artificial intelligence prompted by natural language to write computer code, a term attributed to the former Director of AI at Tesla and founding engineer at OpenAI, Andrej Karpathy, in his post on X on 2 February 2025¹. Software creation is thus accessible to anyone with an idea and a prompt, regardless of formal coding skills, earning the term Collins Dictionary Word of the Year for 2025². The disruption extends deep into enterprise markets: in April 2026, Sundar Pichai, CEO at Google, stated that “75% of all new code at Google is now AI-generated and approved by engineers, up from 50% last fall”³. A parallel race is unfolding in the Greater Bay Area, where Chinese tech giant Tencent is actively building and integrating its own Hunyuan-powered AI coding model into its development workflows⁴. Concurrently, Software-as-a-Service (SaaS) companies lost around $300 billion in market value within 48 hours, prompting financial analysts to dub the phenomenon the “SaaSpocalypse”⁵.

Financial analysts called it the “SaaSpocalypse”.

−$300 billion in market value, within 48 hours.

Yet empirical studies reveal a context-dependent productivity paradox. One field experiment found a 26% increase in tasks completed with AI coding tools, whereas a 2025 study of experienced developers using then-available tools found a 19% slowdown on specific repository tasks. Industry telemetry separately reported a 441% increase in pull-request review time; these different metrics do not establish a universal causal effect⁶. Vibe-coded software often lacks clean architecture, introducing security flaws and hallucinated software packages that non-technical creators cannot easily fix. Building an initial prototype is cheaper than ever, but verifying, securing, and maintaining a commercial product remains costly.

Given this shift, it is necessary to examine whether natural-language software generation can help individuals climb the social ladder, and how specific jurisdictions support this emerging market. Notably, Shenzhen’s 2026–2027 plan for AI “one-person companies” (OPCs) promotes AI-enabled solo entrepreneurship as a route to commercialisation⁷. However, this potential raises major intellectual property (IP) challenges: beyond copyright ownership, where protection generally turns on identifiable human creative contributions⁸, how do code provenance, third-party open-source licences, API dependencies, trade secrets, and AI platform terms affect the ability to commercialise and scale AI-generated software?⁹ Accordingly, this paper asks: Can vibe coding democratise access to the digital economy for individuals without formal programming qualifications or substantial capital, and how do IP frameworks and entrepreneurial ecosystems in Portugal and the Greater Bay Area (GBA) facilitate or constrain this process?

Section 2 examines Portugal as the home jurisdiction; Section 3 analyses the GBA; Section 4 compares both systems; Section 5 offers policy recommendations; and Section 6 concludes.

2Portugal

2.1Macroeconomic Context: Necessity, Entrepreneurship and Tech Talent

Evaluating whether vibe coding can drive social mobility in Portugal requires situating the solo founder within the national ecosystem and EU single market. By 2025, Portugal surpassed its national target with 5,091 active start-ups generating €2.86 billion in turnover¹⁰. However, structural macroeconomic hurdles persist. In 2025, Portugal’s youth unemployment rate (aged 15–24) stood at 20.22%, while estimates suggest that nearly 30% of Portuguese-born people aged 15–39 live abroad¹¹. Meanwhile, R&D expenditure was approximately 1.7% of GDP in recent comparable data, reflecting moderate innovation intensity¹².

Economic studies distinguish between “opportunity entrepreneurship” – scalable ventures backed by capital – and “necessity entrepreneurship”, where people turn to self-employment because regular jobs are scarce. In Portugal, youth unemployment may push some non-technical graduates toward necessity self-employment, making vibe coding an appealing way to build digital products without start-up capital. However, turning an AI prototype into a reliable commercial product requires human engineering verification, colliding with Portugal’s severe shortage of ICT specialists: ICT vacancy tightness is 288% higher than that of the average occupation in Portugal (against 117% across the OECD). Where a vibe-coded app requires extensive debugging¹³, hiring local developers can be expensive for low-capital founders.

2.2Copyright Ownership: Human Authorship and the Public Domain Problem

Portuguese vibe coders also face a direct IP hurdle: uncertainty over copyright in AI-generated code. In Portugal, software copyright is governed by Decree-Law No. 252/94 (originally transposing Directive 91/250/EEC, subsequently codified by Directive 2009/24/EC) and the Code of Copyright and Related Rights (CDADC)¹⁴. Under Article 1(3) of Directive 2009/24/EC, a computer program must be the “author’s own intellectual creation”; Article 1(2) of Decree-Law No. 252/94 refers to the program’s creative character. The Court of Justice of the European Union (CJEU), in Infopaq (Case C-5/08) and subsequent cases, connected originality with free and creative choices reflected in the protected expression¹⁵. Portuguese courts have applied originality and creative-character requirements to computer programs, though no Portuguese case has yet directly ruled on AI-generated software¹⁶.

This is consistent with the prevailing view that autonomously generated output without sufficient human creative contribution may fail copyright originality requirements. When a vibe coder merely supplies a general idea, that idea is excluded from software copyright under Article 1(2) of Directive 2009/24/EC. Yet the absence of manual syntax editing does not itself rule out protectable human contributions to preparatory design or expressive selection and arrangement. Purely AI-generated code without such contributions may remain unprotected, although any embedded third-party material can still be protected¹⁷.

What happens, then, to large companies like Google where 75% of new code is AI-generated, and how does that differ from solo vibe coders? Large tech firms might be more protected because their human engineers review, modify, and integrate AI code into massive existing codebases, potentially allowing copyright in original human contributions to the software; Article 3(2) of Decree-Law No. 252/94 presumes collective-work status for software created within an enterprise but does not dispense with originality¹⁸, backed by private cloud servers and strong market scale. Conversely, for a solo founder whose app is entirely AI-generated, uncertain copyright complicates protection against copying, the grant of exclusive licences and investor IP due diligence.

2.3Can Vibe Coders Protect Platform Appearance or Functionality Instead?

Without source-code copyright, can vibe coders rely on alternative IP rights covering functionality, Graphical User Interfaces (GUIs), or trade secrets? EU and Portuguese case law shows these workarounds offer only partial protection:

First, software functionality cannot be copyrighted. In SAS Institute Inc v World Programming Ltd (C-406/10), the CJEU held that neither the functionality of a computer program, nor its programming language or data file formats, constitute protectable expression under Directive 2009/24/EC, as monopolising functionality would amount to patenting ideas. Under SAS Institute and Articles 2(2) and 6(1)(b) of Decree-Law No. 252/94, lawful users may, subject to statutory conditions, observe, study and test a program to identify underlying principles, without thereby acquiring a right to copy protected expression¹⁹.

Second, protecting the visual appearance or GUI is possible, but only covers surface design. In BSA (Case C-393/09), the CJEU ruled that a GUI is not covered by the Software Directive, but can be protected as an independent work under the InfoSoc Directive (2001/29/EC) if it reflects the author’s own creative choices. Similarly, in Cofemel v G-Star Raw (Case C-683/17, referred from Portugal), the CJEU confirmed that designs qualify for copyright if they reflect free and creative human choices²⁰. While founders whose custom interfaces satisfy the relevant originality or registered-design requirements may challenge protected visual copying, competitors can easily alter the layout while copying the underlying tool.

Third, SaaS founders can try to protect backend code as a trade secret under the EU Trade Secrets Directive (2016/943), transposed into the Portuguese Industrial Property Code. Yet trade secret protection requires ‘reasonable steps’ to keep information confidential, which may be undermined if the particular AI service retains or discloses confidential prompts and code without adequate contractual safeguards²¹.

2.4Beyond Authorship: Provenance, Copyleft, APIs, and EU Compliance

Commercialising AI software in Portugal also brings four downstream IP and regulatory risks. First, regarding code provenance and open-source licences: AI models often output third-party open-source code without its licence notice. Accidentally including code governed by a strict ‘copyleft’ licence, such as GPLv3 or AGPLv3 (whose network-use conditions may be relevant to SaaS), can trigger source-code disclosure obligations in defined circumstances; neither licence automatically requires the entire product to be provided for free. Second, the EU Cyber Resilience Act (Regulation (EU) 2024/2847) will impose cybersecurity and technical-documentation duties, including SBOM-related requirements²², on products within its scope, with general application from December 2027; standalone SaaS is not automatically covered, and the AI Act depends on the nature of the system. Non-technical founders rarely have the capital to hire specialists to audit dependencies and compile SBOMs. Third, solo founders rely on AI platform terms under which indemnities, infringement liability and API prices vary by provider, contract and service tier²³. In Portugal, these combined hurdles can constrain scalable entrepreneurship and its potential contribution to social mobility.

In Portugal, these combined hurdles can constrain scalable entrepreneurship and its potential contribution to social mobility.

Same coder. Same app. A different system.

3The Greater Bay Area

3.1Shenzhen’s 2026–2027 AI ‘One-Person Company’ (OPC) Plan: Motives and Scale

The Guangdong-Hong Kong-Macao Greater Bay Area (GBA) handles AI entrepreneurship differently. Linking nine Mainland Guangdong cities with Hong Kong and Macao, the GBA is a mega-region of 86 million people and a GDP above RMB 14 trillion. Operating under ‘One Country, Two Systems’, it combines three legal jurisdictions and wide variations in R&D spending: from 6.7% of GDP in Shenzhen to 0.31% in Macao²⁴.

In January 2026, the Shenzhen Municipal Industry and Information Technology Bureau issued the Action Plan for Building an AI ‘One-Person Company’ (OPC) Entrepreneurial Ecosystem Leading Hub (2026–2027) (深圳市打造人工智能OPC创业生态引领地行动计划)²⁵. The policy motives behind this plan are clear. Whereas software ventures previously required full engineering and operations teams, Shenzhen policymakers recognised that AI agents allow a single founder – a ‘super individual’ (超级个体) – to handle coding, design, and operations without a traditional engineering team. Economically, the OPC plan serves two goals: it aims to create opportunities for university graduates and returning talent, and it drives commercial adoption of domestic AI models across Shenzhen’s hardware and cross-border e-commerce industries.

The Action Plan sets concrete statistical targets for the end of 2027: building over 10 specialised OPC communities (exceeding 500,000 m² of workspace), nurturing over 1,000 high-growth AI OPCs, and attracting 10,000 AI innovation talents. Crucially, Shenzhen subsidises the exact downstream costs that solo founders struggle with, using the ‘Three Vouchers’ (三券) tool: eligible projects can receive up to RMB 10 million in ‘Computing Power Vouchers’ (算力券) for cloud compute, up to RMB 2 million in ‘Model Vouchers’ (模型券) for AI APIs, and up to RMB 2 million in ‘Corpus Vouchers’ (语料券) for eligible data resources. District-level programmes add practical support: Nanshan District’s ‘Model Camp’ (模力营) offers half-day company registration and seven public service platforms, including a dedicated open-source and compliance platform, while Longgang District offers up to RMB 10 million in seed equity, RMB 10,000 cash grants for recent graduates and HK/Macao residents, up to three years of rent-free 3m² workstations, and one year of rent-free housing. Local hubs like the OPC Companion Community further connect vibe coders through in-person events and WeChat networks²⁶.

3.2Mainland China’s Copyright Rules and the 2026 CPCC Registration Restriction

Within Mainland Guangdong, however, solo founders face a sharp tension between flexible court rulings on AI art and strict administrative rules on software registration. In visual art cases like Li v Liu (2023), the Beijing Internet Court granted copyright to an AI-generated image because the user’s 150+ prompts and parameter choices reflected human intellectual achievements, building on earlier reasoning from the Shenzhen Nanshan District People’s Court in Tencent v Yingxun (2019)²⁷.

For software, though, administrative rules tightened in 2026. In Mainland China, a Software Copyright Registration Certificate from the China Copyright Protection Centre (CPCC) can be important evidence for particular incentives and commercial procedures, although registration does not create copyright. After national software registrations reached 3.18 million in 2025 (~330,000 in Guangdong), the CPCC introduced a restrictive March 2026 registration undertaking concerning AI use in writing code and application materials²⁸. Applicants are required to provide detailed supporting materials and declare that their software was independently developed without AI writing the code or application materials. This creates a paradox for Shenzhen vibe coders: the city promotes AI-assisted development, while the national registration procedure restricts such declarations. Registration and the substantive subsistence of copyright must nevertheless be distinguished.

3.3Trade Secrets in Shenzhen and the Hong Kong SAR Legal Bridge

The GBA ecosystem helps founders overcome this hurdle in two ways. First, in Mainland Guangdong, Shenzhen courts strongly protect software source code, algorithms, and business data as trade secrets under Article 10 and general unfair competition rules under Article 2 of the PRC Anti-Unfair Competition Law (AUCL), as revised in 2025, while local hubs like Nanshan’s Model Camp assist OPCs with open-source checks and relevant algorithm-filing and compliance requirements where applicable²⁹.

Second, founders can use Hong Kong SAR as a legal and financial bridge. With 5,221 start-ups in 2025, Hong Kong operates under common law and explicitly protects ‘computer-generated works’ under Section 11(3) of the Copyright Ordinance (Cap 528). Where a computer program (a literary work under Section 4(1)(b)) has no human author, Section 11(3) may attribute authorship to ‘the person by whom the arrangements necessary for the creation of the work are undertaken’ for 50 years (Section 17(6)), although the Hong Kong Intellectual Property Department acknowledges unresolved questions of originality and identifying that person³⁰. A Shenzhen OPC founder may thus combine municipal compute support with Hong Kong licensing and investment infrastructure, but incorporation in Hong Kong neither automatically determines authorship nor guarantees copyright protection worldwide.

4Comparison and Explanation

4.1Side-by-Side Comparison Across the Commercialisation Process

Comparing Portugal and the GBA across three stages – (1) Technical Democratisation, (2) Legal and Commercial Capability, and (3) Economic Conversion –, highlights the role of local institutions. While Stage 1 (prototyping) is equally accessible in Lisbon and Shenzhen, Stage 2 diverges sharply:

Copyright Ownership & Alternative IP: In Portugal, EU rules (Infopaq; Decree-Law No. 252/94) make copyright in code lacking identifiable human creativity doubtful, and SAS Institute leaves functionality unprotected, leaving founders with narrow GUI protection (BSA, Cofemel). In the GBA, despite Mainland China’s restrictive 2026 CPCC registration undertaking, Hong Kong’s Cap 528 Section 11(3) may attribute authorship to the person making the ‘necessary arrangements’, subject to originality and territorial limits³¹;

Provenance, Open Source & Compliance: Portuguese solo founders may need to audit copyleft licences (GPL/AGPL) and, for products within the scope of the EU Cyber Resilience Act when applicable, prepare SBOM-related documentation in a market where tech developer shortages are 288% above the average Portuguese occupation. In Shenzhen, OPC hubs like Nanshan’s Model Camp provide shared open-source compliance services and RMB 2 million data vouchers.

APIs, Trade Secrets & Platform Terms: Portuguese founders rely on foreign APIs under contract-dependent terms with varying indemnities and confidentiality safeguards, whereas Shenzhen’s RMB 10 million compute vouchers and local open-weight models may facilitate private deployment, potentially improving control over trade secrets.

4.2Explaining the Divergence: Rules, Practice, Interests, History, and the World Outside

Applying the area-studies method³², five structural factors help explain the different conditions for economic conversion in the GBA and Portugal:

Rules: Portugal follows the European droit d’auteur tradition shaped by the CJEU, which treats copyright as a human personal right and requires human creative choices. The GBA combines Mainland civil law (where registration undertakings have tightened, partly in response to integrity concerns) with Hong Kong’s pragmatic common law copyright (Cap 528), which expressly addresses some computer-generated works.

Practice: The EU imposes safety and transparency rules (including the Cyber Resilience Act and AI Act) where applicable, potentially creating compliance costs for some solo founders. Shenzhen governs through flexible industrial policy, offering half-day registration, free 3m² desks, and compute vouchers to reduce early friction.

Interests: With R&D at approximately 1.7% of GDP and 20.22% youth unemployment, self-employment in Portugal may provide an alternative livelihood. In Shenzhen (6.7% R&D/GDP), city authorities actively back 1,000+ AI OPCs to absorb graduates and boost local cloud and hardware demand.

History: Portugal’s start-up scene grew after 2011 mainly around software services without deep manufacturing chains. Shenzhen evolved from a 1980 Special Economic Zone into a top global innovation cluster, as documented by WIPO and observed during our programme’s site visits to Huawei and Tencent and lectures in Shenzhen (October 2026)³³, where software innovation connects directly to hardware supply chains, design protection, and enterprise testing.

The World Outside: Portugal enjoys EU single-market access but depends on US cloud platforms. Facing US chip export controls, China and the GBA have invested heavily in domestic compute vouchers and open-weight models while using Hong Kong’s common law system as a bridge to global markets³⁴.

5Policy Recommendations

To enable vibe coding to support real entrepreneurship and social mobility while managing IP risks, three realistic measures are recommended:

a) European Commission, Startup Portugal, and the Portuguese Institute of Industrial Property (INPI) – Create a Micro-Enterprise Compliance Sandbox and Free SBOM Scanner: EU and Portuguese authorities should introduce a two-year support sandbox helping eligible solo founders and micro-enterprises meet applicable EU Cyber Resilience Act requirements, using the Regulation’s existing SME-sensitive measures. At the same time, Startup Portugal and INPI – following Nanshan’s Model Camp example – should provide a free public scanning tool so non-technical founders can check AI-generated code for copyleft (GPL/AGPL) licences and generate SBOMs where required before placing relevant products on the market.

b) National Copyright Administration of China (NCAC) and the China Copyright Protection Centre (CPCC) – Introduce an “AI-Assisted Software Registration Track”: To resolve the clash between Shenzhen’s OPC plan and the CPCC’s restrictive 2026 registration undertaking concerning AI-generated code, the NCAC and CPCC should create a dedicated registration track where OPC founders openly declare the AI tools and prompts used, and seek registration based on identifiable, sufficiently original human-designed elements of system architecture, workflow selection and interface integration – aligning software registration with the Beijing Internet Court’s approach in Li v Liu.

c) For the Shenzhen Market Supervision Administration (IP Bureau) and the Hong Kong Intellectual Property Department (IPD) – Publish a Cross-Border “GBA OPC IP Guide”: Through the Qianhai Cooperation Zone, Shenzhen and Hong Kong IP authorities should provide simple legal templates helping Mainland OPC founders document authorship and ownership, structure licensing and understand Hong Kong’s computer-generated-work rules under Section 11(3) of Cap 528 (where copyright is not registered), making it easier to combine Mainland computing support with potential Hong Kong IP protection, subject to territorial limits, when raising investment.

6Conclusion

Vibe coding clearly democratises software prototyping for non-technical founders with limited capital, yet cheaper code generation does not automatically yield sustainable businesses or upward social mobility. As the cost of generating syntax drops – seen in Google’s 75% AI code figure and the ~$300 billion ‘SaaSpocalypse’ – the real bottlenecks shift downstream to code verification (including reported increases in review time in certain datasets) and intellectual property compliance.

At this downstream stage, IP frameworks and local ecosystems make the decisive difference. Because purely AI-generated code without sufficient human creative contribution may lack copyright under EU and Portuguese law (Infopaq; Decree-Law No. 252/94) and faces restrictive 2026 CPCC registration undertakings in Mainland China, vibe coders cannot always rely on traditional source-code copyright. While alternative IP routes, such as protecting manually designed interfaces (BSA, Cofemel) or keeping backend code as a trade secret, offer partial workarounds, they do not protect software functionality under SAS Institute and do not shield solo founders from open-source copyleft contamination, EU Cyber Resilience Act compliance costs, or one-sided AI platform terms. Consequently, in Portugal, where high youth unemployment (20.22%) meets severe tech talent shortages (288% above the average Portuguese occupation) and potential compliance costs, vibe coding risks leaving solo founders in precarious self-employment. By contrast, the GBA, combining Shenzhen’s 2026–2027 OPC Action Plan (RMB 10 million computing vouchers and shared compliance services) with Hong Kong’s conditional framework for computer-generated works under Cap 528 Section 11(3), may provide institutional support for turning technical access into economic opportunity.

Two limits of this study should be noted: first, because vibe coding only recently emerged and both Shenzhen’s OPC plan and the CPCC registration restrictions emerged in March 2026, long-term data on business survival and household income mobility are not yet available; second, comparing an EU country with a mega-region like the GBA involves scale differences, addressed here by comparing how rules and institutions work in practice. Ultimately, while AI democratises the writing of code, only supportive ecosystems and adaptable IP rules can democratise the building of a livelihood.