AI Freelancer, Permanent Hire or Agency: What Fits When?

Your company is going to work with Artificial Intelligence: that has been decided. The harder question remains open. Who builds this capability? Write a job advert for an AI expert? Bring in a freelancer for six months? Or hand the whole topic over to an agency? Each of these three routes works. But each works for a different starting position. Choose the wrong one and you only notice months later: in an unfilled position, in a project without an internal point of contact, or in a dependency that becomes more expensive than planned. This guide compares the three routes along four criteria that genuinely matter for the decision: cost logic, speed, knowledge retention and risk.

The three routes at a glance

Before the comparison, a clean distinction, because the terms are often mixed up:

• AI freelancer: An AI freelancer is a self-employed specialist who works for a company on a project basis, for example as a machine learning engineer, AI consultant or developer for Agentic AI. Billing is by day rate or as a contract for work, that is, with a defined outcome as the subject of the contract. The freelancer remains legally an entrepreneur and is not integrated into your organisation.
• Permanent hire: You create a dedicated position, from AI developer to „Head of AI“. The knowledge becomes part of your organisation, the person part of your payroll. The commitment is long-term, in both directions.
• Agency: An external service provider takes on the project as a whole: conception, implementation, often operation too. You are not buying personnel but a result, including responsibility for it. This includes consultancies, AI agencies and systems integrators.

A fourth variant should be mentioned for completeness: the interim manager, that is, an experienced executive on a temporary basis. It sits between freelancer and permanent hire and is suitable when what is missing is not implementation but leadership and capability-building.

Cost logic: it is not the price that decides, but the structure

The most common error of reasoning in this decision is the pure day-rate comparison. A freelancer day rate looks high next to the pro-rata daily salary of an employee. But this calculation ignores how the costs arise in each case.

Permanent hire: low rate, high fixed costs

On top of the gross salary come employer contributions (typically 20 to 30 percent on top) plus recruiting effort, workspace, training and management time. Above all, however, the costs run regardless of utilisation. An AI position pays off when AI work arises permanently and full-time. That is precisely what is rarely the case with the first AI initiative in a mid-sized company. At the start there is analysis, a pilot project, then a pause for evaluation and a budget decision. An employee who is not fully utilised during this phase is the most expensive option of all.

Freelancer: high rate, no idle costs

With a freelancer you pay only for the days you call off. Social security, acquisition risk and training are priced into the day rate, which explains its level. The costs scale with demand: three days a week during the project, zero days during the evaluation phase. For selective, clearly defined or still uncertain demand, this is the economically most honest structure. A freelancer only becomes expensive when a „six-month project“ quietly turns into two years of permanent occupation. By then a permanent hire would long since have paid off.

Agency: highest effective rate, but in return responsibility for the result

Agencies factor in team overhead, project management and margin. The effective rate per person-day is therefore higher than that of a single freelancer. In return you are buying something different: a team with well-rehearsed processes and, with a clean contract for work, responsibility for the result. If no one internally can or wants to steer the project, this premium is not a luxury. It is the price for relief.

Speed: how quickly does someone actually work on your problem?

Here the differences are greatest and most often underestimated.

• Permanent hire: The labour market for experienced AI specialists is tight. From posting through selection and contract negotiation to the first working day, several months typically pass, including the candidates’ notice periods. For a mid-sized company without a well-known tech brand, the search often takes even longer, because it competes with corporates and start-ups for the same profiles.
• Freelancer: Through specialised placement or an existing network, a vetted AI specialist is typically on the project within one to three weeks. This is the central structural advantage of this route: experience is available immediately, without you having to buy it in permanently.
• Agency: The proposal phase, scoping and contract paperwork require lead time, realistically a few weeks. After that, however, an agency can work with several people in parallel and is therefore, for large, clearly defined undertakings, often the fastest option in terms of implementation.

One honest caveat applies to all three routes: speed is of no use if the task is unclear. Whoever does not yet know which use case is worthwhile only accelerates the confusion with additional personnel. In this phase a structured assessment, for example an AI readiness assessment, is the faster route than any personnel decision.

Knowledge retention: who ultimately owns the capability?

For the management this is, in the long run, the most important criterion, and the one with the most illusions.

Permanent hire promises the best knowledge retention: the capability sits in-house, knows your data, your processes, your customers. That holds true as long as the person stays. A single AI specialist is at the same time a concentration risk. If they leave, the knowledge is lost more completely than with any external party, because no one else ever held it. Knowledge retention through permanent hiring only works from two people onwards, or with consistent documentation.

Freelancers take their knowledge with them again by default. That is their business model. The transfer does not happen by itself, but it can be organised: documentation as a contractually agreed deliverable, joint working with an internal employee (pairing), a structured handover at the end of the project. Whoever engages a freelancer should determine from day one who learns internally. In addition, AI workshops can broaden the foundational knowledge in the team, so that the handover falls on fertile ground.

Agencies have the structurally weakest knowledge retention. The experiential knowledge of why decisions were made the way they were, which dead ends there were, stays with the agency. What remains in the company is the result. If the AI solution is then developed further internally, precisely this knowledge is missing. What can be secured contractually: full handover of all artefacts (code, prompts, configurations, model decisions), an obligation to document, and defined handover dates with your own people.

Risk: what can go wrong with each route?

Permanent hire: the risk of a bad hire

A bad hire in the AI field is doubly expensive: through the personnel costs themselves and through the lost time during which the topic was seemingly covered. Making matters worse, managements without their own AI knowledge find it hard to assess the qualifications of applicants. The field develops quickly, and impressive buzzwords are no proof of competence. On top of this comes the technology risk: a job profile that fits today may, in two years, miss the mark of actual demand.

Freelancer: bogus self-employment and the bus factor

The most important legal risk is bogus self-employment. If a person who is formally self-employed in fact works like an employee, that is, bound by instructions, integrated into teams and working hours, without their own entrepreneurial risk, social security contributions can become payable retroactively. The risk can be reduced through project-based contracts with defined outcomes, the freelancer’s self-directed organisation of work, and limited contract terms. On top of this comes the „bus factor“. A freelancer is a single person: in the event of illness, absence or a more attractive follow-on assignment, the project stalls. Reputable intermediaries cushion this through the provision of a replacement. In a direct contract you should at least keep it in mind.

Agency: dependency and standard solutions

The agency risk is vendor lock-in, the dependency on a provider whose solution no one else understands or can maintain. Every change then runs through the agency, on its terms. The second risk: agencies have an economic interest in deploying tried-and-tested standard solutions. That is often even correct, but it can lead to your specific features being forced into a template that does not fit. Both risks are manageable: through handover obligations, open technologies and an internal point of contact who can engage on the substance.

The direct comparison

Which route fits when? A decision aid

From the four criteria, three typical starting positions can be derived:

• You are facing your first AI project and the demand afterwards is open: freelancer. You buy experience precisely for the duration you need it, and in doing so learn which capability you really need in the long term. A permanent hire based on a hunch is the riskiest option in this phase.
• AI is already part of your core business or is meant to become so: permanent hire, built up with external guidance. Whoever permanently operates their own AI systems or integrates AI agents into their processes needs internal knowledge. The realistic route there often runs via an intermediate stage: an experienced external party builds it up, the first in-house people learn alongside, then the role is filled internally.
• A large, clearly defined undertaking, and internally both capacity and steering are missing: agency, with contractually secured knowledge transfer. Here the responsibility for the result pays off. Insist on handover obligations and designate one person internally who accompanies the project on the substance.

In practice the routes are not mutually exclusive. They often follow one another: a freelancer for the pilot project, from this the job profile for the permanent hire is derived, an agency for the later scaling. The only thing that matters is that the sequence fits the maturity of your company and not the other way round. Where this maturity stands and which use case comes first is a strategic question. If it is still open at your company, start there, for example with an AI strategy, before you decide on personnel. Our glossary explains the central terms around AI roles and technologies.

Conclusion

There is no best route to AI capability, there is the fitting route for your starting position. The short formula: a freelancer for getting started and for selective demand, a permanent hire for permanent core business, an agency for large undertakings without internal steering. For most mid-sized companies facing their first AI personnel decision, the freelancer is the rational starting point. Quickly available, without a commitment to fixed costs, and with the option of deriving the right long-term decision out of the project itself. One condition always applies here: organise the knowledge transfer from day one. Otherwise, with each of the three routes, you are only buying a result and not a capability.