Integrating Artificial Intelligence and Large Language Models into an organization requires more than just purchasing software licenses or API tokens. Anyone wanting to create structural value cannot avoid setting up dedicated capacity. But how do you build an AI team when the market is tight, technology is changing rapidly, and specific roles are still very much in flux?
As a hiring manager or founder, you face complex choices. Should you immediately bring in an expensive AI researcher, or is it better to start by retraining existing software engineers? And which roles do you actually need first? To answer these questions properly, it is useful to understand how the different AI roles relate to each other within a modern organizational culture.
The fundamental roles: What do you really need first?
Many organizations make the mistake of looking for a mythical unicorn: a developer who is skilled in advanced machine learning, understands product management, masters data architecture, and directly translates business goals. In practice, this person virtually does not exist. A functioning AI initiative rests on a number of clearly distinguished pillars:
- The AI Engineer / LLM Integrator: This is the technical engine of your team. This person builds the bridges between business applications and foundation models via APIs, frameworks like LangChain or LlamaIndex, and masters vector databases.
- The AI Product Owner / Translator: A tech specialist without business context builds solutions that no one uses; a company without technical insight gets lost in pipe dreams. This role bridges the gap and translates operational challenges into concrete AI use cases.
- The Data Engineer: Without clean, reliable data, even the best language model underperforms (garbage in, garbage out). This specialist provides the pipelines that make data accessible for Retrieval-Augmented Generation (RAG) and fine-tuning.
Depending on the complexity of your issue, you can add roles at a later stage, such as a dedicated AI ethicist or a specialized MLOps engineer. Anyone seeking broader insight into national trends and wage structures would do well to study the current salaries for AI roles in the Netherlands to set realistic expectations.
Combining roles in one person: When can you and when can't you?
In smaller organizations or early startup phases, there is often no budget or headcount for a full team of four or five specialists. Combining roles is then necessary, but it has hard limits.
What often works well in practice:
- Full-stack developer with AI affinity: An experienced backend developer (for example, in Python or TypeScript) can often be trained within a few weeks to build LLM integrations.
- Product Owner with data background: Someone who already has experience with data-driven product development can excellently combine the role of AI Translator with product responsibility.
What almost never works:
- Expecting a data scientist without software engineering experience to build robust, scalable production applications.
- Expecting a junior developer to handle the architecture, security, and prompt optimization without senior guidance.
In-house training versus external recruitment
The market for experienced AI talent is extremely competitive. Job openings often remain vacant for a long time, and salary demands are no joke. Therefore, it pays to look critically at internal talent. Many organizations discover that their best AI engineers are already working in their own Software Development department.
The benefits of in-house training (upskilling)
The biggest advantage of training existing employees is context. An internal developer already knows the legacy systems, security guidelines, and business objectives inside out. You only need to teach them the specific skills of modern AI technology. This also significantly increases employee retention.
When should you recruit externally anyway?
Sometimes, the necessary fundamental knowledge is simply lacking within the organization. If your company is building a core product that relies entirely on unique AI algorithms or advanced fine-tuning, you need a senior expert who can act as a catalyst. You bring that person in from the outside to set standards and mentor the internal team. For a broader perspective on the structure of the labor market, it is useful to follow developments in the AI job market in the Netherlands.
Realistic expectations about seniority in a tight market
A common mistake made by hiring managers is requiring a candidate to meet every conceivable requirement. Because the technology only exploded in its current form after the breakthrough of modern LLMs, there are simply no seniors with ten years of demonstrable experience in generative AI.
Anyone looking for a "senior AI engineer with fifteen years of experience in LLMs" is chasing an illusion. Everyone in this field is learning at a rapid pace. Therefore, when evaluating candidates, do not blindly focus on job titles or years of experience, but on:
- Learning capacity and fundamental foundation: A strong foundation in computer science, mathematics, or solid software engineering is far more valuable than temporary tricks with a specific framework that will be outdated next year.
- Demonstrable personal projects: What has someone built? Has the candidate experimented with open-source models, agents, or RAG architectures in their own time or previous roles?
- Pragmatism: The best AI professionals understand that AI is a means to achieve a business goal, not an end in itself. They choose the simplest solution that works, rather than maximizing complexity.
Conclusion
Building a successful AI team requires realism, focus, and a willingness to invest in your own people. Start small with the core roles you need immediately, accept that the market requires flexibility, and steadily build a future-proof organization from there.