The modern artisan: gaining independence through AI without surrendering judgement

Between the salaried specialist and the technology empire, another figure is becoming possible.

This person may research across disciplines, build software, design products, run operations, publish, sell, and coordinate specialised agents without first assembling a conventional organization around every function. They do not own a foundation model, a cloud, a payment network, or a distribution platform. They are not a technolord. Their power is smaller, more practical, and potentially more humane: they can use large systems without allowing those systems to decide what their work is for.

The modern artisan is our name for that figure. It is an editorial concept, not an established labour-market category. Richard Sennett uses craftsmanship in a similarly broad sense to describe the human commitment to making something well, extending the idea beyond traditional manual trades to fields such as programming and medicine. (Sennett, The Craftsman, 2008) The modern artisan adds a new layer: they do not only make the work. They increasingly design the workshop through which the work is made.

AI can make this form of independence economically possible for more people. It can reduce the cost of expertise, iteration, administration, and coordination. But it does not create independence by itself. A person who leaves one hierarchy only to become completely dependent on one model provider, marketplace, ranking algorithm, or payment rail has changed the location of control, not abolished it.

Real independence is therefore not the absence of systems. It is the ability to choose their direction, understand their contribution, preserve what is learned through them, and replace them when they no longer serve the work.

AI changes the minimum viable organization

In 1937, Ronald Coase asked why firms exist when markets can, in principle, coordinate buyers and sellers. His answer began with a practical observation: using the price mechanism has costs. Finding suppliers, negotiating agreements, transferring information, and coordinating repeated transactions can make internal organization more efficient than contracting for every task separately. (Coase, Economica, 1937)

Generative AI does not remove those costs, but it can lower some of them. A founder can obtain a first pass at research, code, analysis, design alternatives, documentation, customer communication, or operational planning without opening a new hiring or procurement process for every bounded task. Agents can also pass structured work between stages and keep several independent workstreams moving at once.

A preregistered field experiment with 791 Procter & Gamble professionals makes part of this effect visible. Participants worked alone or in pairs, with or without AI, on real product-innovation challenges. Individuals using AI matched the performance of two-person teams working without it and produced proposals that crossed technical and commercial boundaries more effectively than unaided specialists. Yet human judgement still added value when choosing which ideas were worth pursuing. (Dell’Acqua et al., Organization Science, 2026)

That is strong evidence for a bounded claim: AI can supply some of the breadth and generative capacity that previously came from another person in a particular innovation exercise. It is not evidence that relationships, institutions, or whole companies have become optional.

The practical change is still substantial. In some forms of knowledge and creative work, technology can reduce how many specialties must be hired before a serious product or operating system can begin to exist. It lowers the minimum viable organization.

Lower entry is not durable independence

Early evidence suggests that AI may already be changing who attempts entrepreneurship. It also warns against confusing more attempts with more durable success.

A May 2026 preprint analysed more than 160,000 Product Hunt launches around the public release of ChatGPT. By 2025, solo entry had grown nearly 90% more than team entry relative to 2022 in the authors’ difference-in-differences model. The increase was especially pronounced in categories that had previously been team-heavy. But the share of solo launches receiving no update within twelve months rose from 95.1% to 97%, while teams’ share of top-ten rankings increased from 50% to 53%. (Kim, Kang and Song, preprint, 2026)

The study covers one technology-focused platform. Product Hunt rankings are an early attention measure, not revenue, survival, social value, or long-term quality. The release of ChatGPT also coincided with other changes, so the research design cannot turn a historical event into a perfectly controlled experiment. Its most defensible implication is narrower: generative AI appears to have lowered the cost of solo experimentation much more clearly than it eliminated the advantages of strong teams.

An experiment with 640 small-business owners in Kenya provides a second caution. Entrepreneurs were randomly offered access to a GPT-4-powered business adviser through WhatsApp. The researchers could not reject the possibility of no average effect on revenue and profit. Their subgroup analysis found that lower-performing businesses at baseline did nearly 10% worse with the adviser, while higher-performing businesses may have improved by more than 15%. The difference appeared to arise not from the questions or advice alone, but from which recommendations the entrepreneurs selected and implemented. (Otis et al., working paper, revised 2025)

This is one country, one population of microenterprises, and one AI intervention; the subgroup result should not be turned into a universal rule. It does show why access is not enough. AI expands the available action space, but the person still has to identify the right problem, evaluate advice against local reality, and carry a decision into the world.

A workshop becoming cheaper to enter is not the same as a craft becoming easy to master.

Independence is a control stack

Working for oneself is a legal or commercial status. Independence is a question of control.

Albert Hirschman’s classic framework distinguished exit—the ability to leave a deteriorating relationship—from voice—the ability to influence it from within. (Hirschman, Exit, Voice, and Loyalty, 1970) For an AI-augmented independent, both matter. A person needs enough voice to shape the tools and channels they use, and a credible exit when those systems change against their interests.

That control has at least five layers.

Direction

The independent chooses which problem is worth solving, which customer promise is acceptable, what quality means, and when growth would damage the work. AI may propose objectives or strategies, but convenience must not silently become purpose.

Knowledge

The independent keeps the domain knowledge and evaluation ability required to recognise a plausible failure. Important decisions, customer understanding, quality cases, and operating lessons survive outside a transient conversation. As the previous essay in this series argued, assisted performance is not the same thing as retained competence.

Technical exit

Data can be exported. Workflows are separated into replaceable modules. Evaluation cases belong to the organization rather than to one model. A provider can be changed without erasing the standards by which the old and new systems are compared.

Market relationship

The brand, domain, catalogue, and consented customer relationship do not exist only inside a marketplace account. Platforms may remain valuable channels, but they are not allowed to become the sole memory of who the business is, what it sells, or whom it serves.

Economic and relational resilience

No single customer, distribution channel, model, or collaborator should be able to end the entire practice through one decision. This does not mean owning every layer. It means having enough diversity, liquidity, human trust, and recovery capacity to survive a change in one layer.

These are not binary achievements. Independence can be stronger in direction and weaker in distribution, or strong in technical portability and weak in finance. The useful question is not “Am I independent?” but “Which decision can another party make today that I cannot meaningfully resist or recover from?”

The invisible boss

Digital platforms demonstrate why this distinction matters.

Donato Cutolo and Martin Kenney call sellers and producers whose market access depends on a digital intermediary platform-dependent entrepreneurs. Their analysis argues that the power imbalance is not an accidental defect: it follows partly from the platform’s control over technical architecture, governance, data, and access to participants. (Cutolo and Kenney, Academy of Management Perspectives, 2021)

An interview study of 77 high-performing eBay business sellers shows the mechanism at human scale. Even infrequent negative reviews created anxiety and vulnerability. Sellers faced power asymmetries both in transactions with customers and in the governance relationship with eBay, while trying to recover agency through practical knowledge of the platform’s algorithm. (Curchod et al., Administrative Science Quarterly, 2020)

The broader platform-work evidence contains the same tension. An International Labour Organization report drawing on surveys and interviews with about 12,000 workers and representatives of 85 businesses found that digital labour platforms can create income opportunities, flexibility, market access, and lower operating costs. It also documented algorithmic control, insufficient well-paid work for many workers, and gaps in social protection and bargaining power. (ILO, World Employment and Social Outlook, 2021)

A marketplace seller, a ride-hailing driver, and an independent studio are not the same legal or economic case. Nor is a model provider an employer. The shared mechanism is dependence on an intermediary that can change access, visibility, price, or rules while the smaller participant bears much of the adaptation cost.

AI can therefore produce a misleading form of freedom. A person may have no manager, set their own schedule, and still find that their livelihood is governed by systems they cannot inspect, negotiate with, or leave.

The answer is not to abandon platforms. Their distribution, trust mechanisms, infrastructure, and demand can make independence possible in the first place. The answer is to treat them as rented leverage and convert part of that leverage into portable assets: a recognisable brand, direct knowledge of customers, a canonical catalogue, reusable operating methods, and more than one route to market.

The artisan owns the standard, not every tool

The historical artisan did not mine every metal, grow every fibre, or build every instrument. Independence never meant creating the whole world alone. It meant possessing enough skill and control to remain answerable for the work.

That is the useful part of the craft metaphor in an AI setting.

The modern artisan has three connected roles:

1. Maker: they understand the material or domain well enough to recognise quality, diagnose failure, and make consequential choices.
2. Workshop builder: they turn repeated work into tools, records, evaluations, and bounded automations that make future work more reliable.
3. Principal: they decide what the work is for and remain accountable to the customer, collaborator, and public affected by it.

An eight-month ethnography inside a French mould-producing company found that workers in formally low-skilled roles used latent craft skills to create pockets of greater autonomy and earn recognition from colleagues and supervisors. The study does not describe AI entrepreneurship, but it supports a broader point: craft is not merely a product category. It can be a way of recovering judgement, authorship, and meaning inside systems that otherwise narrow them. (Rostain and Clarke, Organization Studies, 2025)

Automation is compatible with that idea. The artisan does not prove authenticity by doing every repetitive operation manually. They preserve authorship by deciding what should be automated, defining what a good result looks like, inspecting the cases where the system is likely to fail, and accepting responsibility for what leaves the workshop.

Build a workshop that can change tools

Independence has an architectural cost. A tightly coupled system is often faster to assemble at first. A replaceable one requires interfaces, records, and tests before the need to switch becomes urgent.

Research on modular design describes this extra work as the creation of options. When components interact through explicit design rules, an inferior module can be substituted without redesigning everything around it. Those options have value, but modularity is not free; its architecture has to be designed and maintained. (Baldwin and Clark, Design Rules, 2000)

For an AI-augmented independent, the equivalent workshop includes:

• canonical, exportable records for products, customers, decisions, and rights;
• task definitions and data contracts that do not assume one model’s private interface;
• evaluation cases drawn from real work, including known failures;
• versioned outputs and decision records that make change visible;
• human approval gates for money, rights, security, publication, and irreversible state;
• a release and recovery path when an automation fails;
• an owned domain and brand alongside external distribution channels;
• a lawful, consent-based way to maintain customer relationships outside any single marketplace.

None of this makes a business sovereign. Cloud providers, banks, payment networks, suppliers, regulators, and customers remain real dependencies. The goal is not autarky. It is replaceable dependence: knowing what is rented, keeping what must endure, and preserving enough choice to move.

Independent does not mean alone

The romantic image of the lone genius is a poor model for the modern artisan.

Research across four countries found that entrepreneurs systematically changed how they used discussion networks through different stages of forming a business, speaking with more people during planning and continuing to draw on personal ties while operating. (Greve and Salaff, Entrepreneurship Theory and Practice, 2003) A later meta-analysis covering 31 independent samples and 5,259 observations found positive associations between venture growth and both network size and tie strength, while network density was not significantly related to growth. These are associations across heterogeneous studies, not proof that adding contacts causes success. They do reinforce that entrepreneurship is socially embedded even when ownership is individual. (Peng, Li and Liu, SAGE Open, 2022)

A qualified independent should therefore be understood as a strong node in a network, not a sealed unit. They may own the direction while collaborating with specialists, peers, suppliers, customers, and future employees. They seek human disagreement where incentives, lived experience, or professional accountability matter. They use AI to reduce the cost of preparation and coordination, not to simulate away every relationship.

Small human headcount is a possible starting structure. It should never become an ideology that prevents the organization from adding a person whose judgement, care, trust, or responsibility improves the work.

Independence is not an easier life

There are good reasons people value self-direction, but it should not be romanticised as automatic well-being.

Research comparing self-employment with organizational employment has often linked the satisfaction advantage of self-employment to autonomy rather than income. (Benz and Frey, Economica, 2008) Longitudinal evidence complicates the picture: one study following transitions into self-employment found that gains in satisfaction with autonomy, schedule flexibility, and the nature of the work were not permanent. (Georgellis and Yusuf, Journal of Small Business Management, 2016)

The modern artisan assumes risks that an employer would otherwise distribute: uneven demand, compliance, security, customer support, capital allocation, illness, and continuity. AI can reduce the cost of some operations. It cannot guarantee demand, absorb liability, create social protection, or give the founder more hours in a day.

Independence should therefore be measured partly by what happens under stress. Can the work pause without disappearing? Can a provider fail without destroying the records? Can a specialist take over a bounded part? Can the founder say no to revenue that would compromise the standard? A system that produces impressive output only while one person works continuously is not independent. It is fragile.

What GeMarkt can honestly say

GeMarkt is not an example of technological sovereignty or an autonomous company. It is a human-managed, agent-assisted one-person studio gradually trying to make its capabilities more portable and its commercial identity less dependent on any single channel.

As of 1 August 2026, the live production site provides art discovery, buyer tools, a growing public catalogue, and privacy-preserving first-party measurement. Purchases still complete through external marketplaces. This gives GeMarkt distribution and payment infrastructure, but it also means the direct commercial relationship is incomplete.

In AWS non-production, a private media-delivery path exists with bounded publishing and verification. In local development and CI, provider-neutral catalogue contracts, versioned public projections, migration and replay safeguards, tests, release gates, and rollback mechanisms form part of the future commerce foundation.

Customer accounts, baskets, direct checkout, a production commerce database, mature fulfilment operations, and a production storefront cutover remain planned, not live.

The independence story is therefore a direction of travel, not an accomplished fact. GeMarkt is using external platforms while building an owned brand, canonical knowledge, reusable evaluations, portable catalogue structures, and safer operating methods around them. AI agents expand the breadth one founder can coordinate. They do not remove dependence on marketplaces, infrastructure providers, suppliers, professional expertise, or customers.

That bounded case is more valuable than a myth of autonomy. It shows what independence looks like in practice: not a declaration, but the slow conversion of rented leverage into retained capability.

Could qualified independents become a serious economic force?

The evidence cannot support a confident forecast that AI-augmented independents will become a coherent new class or displace large firms. The clearest early platform evidence shows more solo entry, not solo dominance at the top. Capital-intensive production, regulated services, large physical operations, and many forms of care still require durable teams and institutions. Even in digital work, trusted relationships and differentiated human knowledge remain valuable.

But a narrower vision is credible.

AI can allow more people to test an idea before raising capital, hire later rather than earlier, retain ownership through a longer stage of development, serve a smaller specialised market profitably, and combine creative and technical work that previously sat in separate organizations. Some of those people will remain solo. Others will build small firms, cooperatives, studios, or networks of specialists. Independence is compatible with growth; it simply changes the terms on which growth begins.

Their collective significance would not come from behaving like miniature technology empires. It would come from many people possessing more credible options: the option to create, to refuse, to change providers, to reach customers directly, to collaborate without surrendering direction, and to build an institution only when the work genuinely needs one.

This will not eliminate concentrated technological power. It may create a more plural economy around it—one in which powerful infrastructure is used by people and small organizations that retain more of their own knowledge, identity, and bargaining capacity.

Systems that remain answerable

AI can help a person gain independence when its output becomes more than a temporary performance. The work must leave behind knowledge, standards, customer trust, portable records, and options that the person or organization continues to control.

The modern artisan does not work without systems. They build systems that remain answerable to them.

That is a quieter ambition than becoming a technolord, and a more durable one. Independence in the AI age is not the fantasy of needing nobody. It is the practical ability to choose, understand, and replace the systems on which one’s work depends—without surrendering judgement about what should be made, how well it should be made, and whom it should serve.