Introduction: AI Without the Right Team Is Just an Expensive Toy
The best technology in the world does nothing without the right people. Many AI projects fail not because of bad technology, but because of the wrong team.
In this episode, I will tell you what roles an AI team needs, the minimum number of people required, how to find them, and whether it is better to build an internal team or outsource.
Key Roles in an AI Team
1. Machine Learning Engineer
This person builds, trains, and optimizes the model. The difference from a data scientist is that an ML Engineer focuses more on practical implementation and production, not research.
Required skills: Python, ML frameworks (PyTorch or TensorFlow), familiarity with LLMs, language model APIs, basic mathematics and statistics.
Why it matters: Without this role, you simply cannot build any custom model.
2. Data Engineer
This person collects, cleans, structures, and prepares data. Remember, 80% of AI work is data. So this role is critical.
Required skills: Databases (SQL, NoSQL), ETL (Extract, Transform, Load), data processing tools, Python.
Why it matters: Without clean, structured data, the AI model is essentially operating blind.
3. AI Product Manager
This person is the bridge between the technical team and the business. They understand what AI can and cannot do. They set priorities and ensure the technical team is working on the right problem.
Required skills: Understanding AI principles (no coding needed), strong communication skills, business understanding, project management.
Why it matters: Without this role, the technical team might build an amazing model that nobody needs.
4. Software Engineer
The AI model itself is not a product — it needs to fit into an application, website, or system. The software engineer turns the model into a real product.
Required skills: Web or mobile development, APIs, basic DevOps, working with existing systems.
Why it matters: A model without an application is like an engine without a car.
5. Domain Expert
This person does not know AI — but they know your business domain well. If you work in healthcare, a doctor. In legal, a lawyer. In finance, an accountant.
Why it matters: AI without domain expertise might produce technically correct but practically useless results.
Minimum Viable Team
If budget is limited, what is the minimum number of people needed?
If using APIs (like ChatGPT API):
Minimum: 1 person — a software engineer familiar with APIs who can build a simple RAG system. Many Level 1 projects can be executed with just one person.
If building a custom model:
Minimum: 3 people
- 1 ML Engineer (model building)
- 1 Software/Data Engineer (data preparation + integration)
- 1 Product Manager/Domain Expert (direction)
With 3 people, you can run a real AI project. Not ideal — but it works.
Ideal team for a serious project:
5 to 8 people
- 1-2 ML Engineers
- 1 Data Engineer
- 1-2 Software Engineers
- 1 Product Manager
- 1 Domain Expert (can be part-time)
- Optional: 1 UX Designer (because the UI of AI products is very important)
In-house Team or Outsource?
This is one of the most important decisions you need to make.
In-house Team
Pros:
- Full control
- Organizational knowledge stays in the team
- Direct and fast communication with other teams
- Higher security (data does not leave the organization)
Cons:
- Hiring is hard — AI talent is scarce and competition is high
- High cost — fixed salaries even when the project is done
- Risk of losing key personnel
Outsource
Pros:
- Faster start
- Flexible costs
- Access to diverse experience
- Lower risk for the first project
Cons:
- Less control
- Knowledge leaves with the end of the project
- Communication challenges
- Data privacy concerns
How to Hire AI Talent
1. Use the right job title
“AI Specialist” is too generic. Be specific about what you need: “ML Engineer with NLP and LLM experience” or “Data Engineer with ETL and Python experience.”
2. Practical experience matters more than degrees
Someone with 2 years of real AI project experience usually performs better than someone with a PhD in ML but no industry experience. Look at GitHub projects, technical articles, and portfolios.
3. Set up practical interviews
Give them a real problem from your business and ask how they would solve it. Not necessarily code — but approach and thinking. People who define the problem correctly are usually better than those who immediately start coding.
4. Offer competitive pay
The AI market is competitive. If you pay below market rate, you will not find good talent. Research what the market rate is and at least match it.
5. Sell growth and learning
AI specialists care about learning. If you provide an environment where they can learn and grow, they might come even with lower pay. Budget for conferences, courses, and research time.
Upskilling Your Current Team
You may not need to bring everyone from outside. Many times, people on your current team can take on AI roles with the right training.
Who has potential to learn AI?
Programmers: With 3 to 6 months of training, they can work with APIs and build simple AI systems.
Data analysts: If they work with SQL and Excel, they can become data engineers by learning Python and ML tools.
Product managers: With a 1-2 month course on AI principles, they can become AI product managers — no coding needed.
Team Structure — Who Do They Report To?
Option 1: Under IT
The most common option. But the problem is that the IT team is usually busy maintaining systems, and AI may not get priority.
Option 2: Independent Team
An independent AI team reporting directly to the CEO. The advantage is focus and priority. The drawback is it might become disconnected from the rest of the organization.
Option 3: Distributed
AI specialists embedded in each department — marketing, sales, production. The advantage is AI is close to the problem. The drawback: coordination is hard and knowledge gets scattered.
Recommendation: Start with a centralized team (Option 2) so knowledge and processes can form. After 1-2 years, distribute specialists to departments (Option 3) and keep one person as a central coordinator.
Summary
Building a team for AI is not simple, but with the right approach it is doable:
- Know the roles — ML Engineer, Data Engineer, PM, Software Engineer, Domain Expert
- Start small — even 1 person is enough for API-based projects
- Think hybrid — first project with a contractor, next with internal team
- Upskill your current team — many have the capability to learn
- Value practical experience over credentials
The next episode discusses AI risks and dangers — what can go wrong and how to prevent it.