What’s Really Like to Be an AI/ML Engineer

The newsletter presents firsthand notes from AI/ML engineers about what the role looks like in practice. It focuses on daily work, the tools they use, the problems they face, and how the job differs from software engineering. It also points to the need for structured interviews and real depth in the problems the engineer wants to work on.

Reading notes

#

• The piece gathers real-world experience from three AI/ML engineers working at Meta, Everseen, and contexxt.ai.
• The questions it centers on are daily work, technologies used, specific challenges, and differences from software engineering.
• Shivam Anand’s work focuses on large-scale machine learning systems and adversarial ML.
• At Meta, his day changes with the phase of the work, moving between planning, technical strategy, mentoring, and hands-on execution.
• His stack includes PyTorch and Meta’s internal systems, with work on deploying LLAMA models.
• One hard problem he describes is scaling LLMs to billions of unknown examples with only a few positive labels.
• The text frames that problem as an extreme class imbalance case that depends on data pipelines, evaluation, and inference performance.
• Compared with software engineering, ML has more uncertainty because the outcome is often unclear until the team experiments.
• In adversarial settings, iteration speed, measurement discipline, and expectation management become central parts of the work.
• The piece says progress in AI is usually empirical and iterative, with trade-offs and tuning rather than magical breakthroughs.
• For Big Tech ML interviews, it recommends deliberate preparation and real depth in the domain of interest.