AI · Careers
Shaping the future: what new graduates need for a career in AI.
In 2024 I wrote that a career in AI would depend on collaboration, a future-ready education, ethics and lifelong learning. Two years later, all four still hold, but the day-to-day work has changed. AI assistants now write much of the first draft of code, so the skills that stand out are framing problems, evaluating results and taking responsibility for what ships.
Learn the fundamentals (Python, data, PyTorch, statistics), learn to build with foundation models (retrieval, tool calling, structured outputs), and learn to prove your system works with evaluations. Use AI assistants every day, but own every line you ship. Build one real, end-to-end project and write about it.
What has changed since 2024
When I wrote the first version of this article, most AI work for graduates meant training or fine-tuning models. Today, many AI products are built on top of large foundation models that someone else trained. The job has moved up a level:
- From training models to building systems. Much of the value is in connecting a model to the right data, tools and guardrails, not in the model itself.
- From writing code to directing and reviewing it. Assistants like Cursor, Windsurf, GitHub Copilot and Claude Code can scaffold a service, write tests and refactor in minutes. The engineer's job is to specify the work clearly and judge the result.
- From "it works on my prompt" to evaluation. Model outputs vary. Teams now expect engineers to measure quality with test sets and automated evaluations, the same way we measure latency or uptime.
1. Collaboration is the job
AI products touch data, software, design, law and the business at once. No single person covers all of that, so the most effective graduates are the ones who work well across disciplines.
- Work in mixed teams. Strong AI features usually come from engineers, designers, domain experts and people who think about ethics working on the same problem.
- Communicate in writing. Clear design notes and pull request descriptions matter more now, because they are also what you hand to an AI assistant as instructions.
- Learn from peers. Pair programming and code review are still the fastest way to grow, with or without AI in the loop.
2. An education built on real projects
Courses struggle to keep up with the field, so real projects carry more weight than grades. When I review candidates, one end-to-end project tells me more than a list of courses. In interviews, I ask candidates to walk me through it: the different approaches they considered, why they chose one, and what they would do differently now. How someone reasons about those trade-offs tells me more than the final result. A strong project:
- solves a real problem with real, messy data;
- has a clear baseline and a way to measure whether the AI version is better;
- runs somewhere other than a notebook, with an API, a simple interface and basic monitoring;
- is written up, including what did not work.
Soft skills still matter as much as technical ones: teamwork, communication and problem-solving are what make a project land inside a real organization.
3. The frameworks worth learning now
Tools change quickly, so learn the concepts first and the tools second. These are the layers most AI applications are built from today, with examples of widely used tools:
| Layer | What to understand | Tools to know |
|---|---|---|
| Foundations | Python, data wrangling, statistics, how neural networks learn | Pandas, NumPy, scikit-learn, PyTorch |
| Models | Using and fine-tuning pre-trained models | Hugging Face Transformers, model provider APIs |
| Retrieval | Giving a model the right context from your own data | Vector search in PostgreSQL (pgvector) or Elasticsearch, LlamaIndex |
| Agents and tools | Letting a model call functions and take multi-step actions safely | LangGraph, Pydantic AI, Model Context Protocol (MCP) |
| Structured outputs | Making model responses predictable and machine-readable | JSON Schema, Pydantic, Zod |
| Evaluation and monitoring | Measuring quality, cost and failures over time | Test sets, automated evaluations, tracing and experiment tracking such as MLflow |
4. Use AI assistants, but own the result
Graduates who use AI assistants well get far more done. Those who accept output they don't understand create problems someone else has to fix. A few habits make the difference:
- Write down what you want before you ask for it: inputs, outputs, edge cases.
- Ask the assistant for tests first, then the code.
- Read every change as if a colleague wrote it. If you can't explain it, don't merge it.
- Never paste secrets, customer data or confidential code into a tool your company hasn't approved.
5. Ethics is part of the engineering
As AI becomes part of daily life, its effects on people grow, and so does regulation, such as the EU AI Act and privacy laws like Canada's PIPEDA. Graduates need to treat ethics as engineering work, not a separate topic.
- Bias and fairness. Test how your system performs across different groups of users, and fix the data or the design when it doesn't.
- Privacy and security. Collect only the data you need, protect it at every stage, and watch for new risks such as prompt injection in tools that read outside content.
- Transparency. Tell users when they are dealing with AI, and keep a human in the loop for decisions that seriously affect people.
6. Keep learning, on purpose
The pace of change makes lifelong learning part of the job description. What works for me:
- Structured learning. Courses and certifications for foundations. I completed the Certified AI Engineer program at Lambton College in 2024.
- Community. Meetups, conferences and open-source projects, to hear what actually works in production.
- Research and building. Small experiments and side projects. My own research project at Lambton was published in the STEM Fellowship Journal.
Conclusion
The future of AI application development brings real challenges and real opportunities. Graduates who collaborate well, learn by building, measure their results, treat ethics as part of the work and keep learning will do more than adapt to change. They will lead it.
Get my next case studies and articles in Google Search.