Tech Career Decisions: What to Know Before 2026

Marina Wyss, a Senior Applied Scientist at Twitch who's mentored nearly 200 people into tech, has a theory about career advice: most of it answers the wrong question. People ask "What's the best tech job?" when they should be asking "What kind of day do I actually want to have?"

In a recent video, Wyss walks through a decision tree for choosing among seven major tech roles—software engineer, data engineer, machine learning engineer, AI engineer, data analyst, data scientist, and applied scientist. What makes her framework interesting isn't the roles themselves (those are well-documented elsewhere) but the questions she uses to separate them. They're psychological more than technical.

The Fundamental Split: Building vs. Discovering

Wyss starts with what she calls "the most important question," and it has nothing to do with programming languages or math requirements: Do you prefer building stuff or discovering insights?

"Builders create things," she explains. "They write code that becomes a feature, a pipeline, or an application. At the end of the day, something exists that didn't exist before, and it's more concrete." Discoverers, meanwhile, "dig through data, test hypotheses, and find patterns. At the end of the day, there's an insight or a model that helps someone make a better decision."

This framing is useful because it cuts through the fog of overlapping job descriptions. Every role on her list involves coding. Many involve data. But the fundamental orientation—are you making things or learning things?—differs in ways that compound over a career.

If you lean toward building, the next question concerns your tolerance for ambiguity. Some builders want deterministic systems where the same input always produces the same output. Others are "comfortable or maybe even energized by uncertainty," working on problems where "you might do everything right and the system still doesn't behave the way you expect."

This is where Wyss separates the traditional engineering roles (software and data engineering) from the AI-adjacent ones (machine learning and AI engineering). It's not primarily about technical difficulty—it's about whether unpredictability drains you or energizes you.

The Clarity Track: Software and Data Engineering

For those who value deterministic work, Wyss distinguishes between software engineering (building products users interact with) and data engineering (building infrastructure that supports everyone else's work).

Software engineering remains the most accessible entry point. "While degrees definitely help, some software engineers do start through boot camps or self-study," Wyss notes. But she's clear-eyed about current market conditions: "Entry-level positions are harder to get into than during the COVID boom. And with AI coding tools improving, employers are looking for everyone to have more senior architecture and system design skills."

The Bureau of Labor Statistics still projects 15% growth through 2034, but that number predates the current AI coding assistant wave. The real question is whether junior roles continue to exist in their traditional form, or whether the bar for "entry-level" keeps rising.

Data engineering, meanwhile, might be "the most AI proof role on this entire list," according to Wyss. "Every AI system requires clean, well structured data." McKinsey specifically names data engineers among the most in-demand AI-related hires. The work is infrastructure—unglamorous, often invisible when done well, and essential.

"One of the realities of data engineering is that their best work goes completely unnoticed," Wyss says. "Nobody comments when the pipeline works perfectly." That's either a dealbreaker or exactly what you want, depending on what motivates you.

The Ambiguity Track: ML vs. AI Engineering

For builders comfortable with unpredictability, the fork comes down to math appetite and education commitment. Machine learning engineers need "strong software engineering fundamentals, deep experience with machine learning, and system design for machine learning systems." They take models and make them work at scale—the classic example being an Uber ML engineer ensuring a surge pricing model "returns predictions in milliseconds" and "works 24/7 without crashing."

The demand is real: ML engineer postings grew 40% year-over-year in 2025, on top of 78% growth in 2024. But the entry barrier is steep—typically requiring a master's or PhD—and the field moves fast enough that "you will never be done learning."

AI engineering, by contrast, builds applications on top of foundation models rather than training models from scratch. "People often say AI engineers are just software engineers calling a different API," Wyss acknowledges. "And honestly, if you're a bad AI engineer, that's true."

But good AI engineers need to understand embeddings, evaluation metrics, ranking algorithms, and how to build systems where "the output is non-deterministic. LLMs don't always return what you expect, and you need to design around that."

The role ranked number one on LinkedIn's jobs on the rise for two consecutive years. The on-ramp is faster than ML engineering—fewer positions require advanced degrees, and the math expectations are lighter. It's a bet on a role that's still defining itself, which means both opportunity and risk.

The Discovery Side: Analytics to Applied Science

For those who chose "discovering" at the initial fork, Wyss uses a different question to navigate the options: Do you prefer variety and stakeholder-driven work, or long focus blocks and self-directed exploration?

Data analysts sit on the reactive end—business detectives who answer questions from marketing teams and product managers. The role is the most accessible entry point into data careers ("No PhD required, and you can learn the core toolkit, SQL, Excel, and a visualization tool in a couple months"), but it's also the most vulnerable to AI automation. Job postings declined 15% year-over-year.

Wyss believes "people with strong business intuition and subject matter expertise will remain in demand," which is probably true but also the kind of reassurance that gets repeated until suddenly it isn't. The skills that make you good at this role—translating vague questions into concrete analysis, storytelling with data—are exactly what large language models are rapidly improving at.

Data scientists get more autonomy and tackle more ambiguous problems. The BLS projects 34% growth through 2034, making it one of the fastest-growing occupations. But the entry barrier is high—most roles want advanced degrees or significant experience—and "the career progression isn't as well defined as it is for engineers."

Applied scientists, Wyss's own role, combine "the analytical rigor of data scientists, the engineering capabilities of machine learning engineers, and the innovative mindset of researchers." It requires the widest skill range on the list: production-level coding, deep ML and statistics knowledge, stakeholder communication, and the ability to take research all the way to deployment.

The job market is more specialized—fewer positions exist, but also fewer qualified candidates. "For many people like myself, applied science is a role you grow into over time," Wyss notes. "Many if not most applied scientists started as data scientists or machine learning engineers."

What The Framework Doesn't Answer

Wyss's decision tree is clarifying, but it leaves some questions deliberately unresolved. What happens when the boundaries between roles blur? How do you choose when your personality fits multiple paths? And perhaps most urgently: how much weight should you give to market conditions versus personal preference when those are in tension?

"Your career is a journey, not a fixed destination," Wyss concludes. Which is both true and unhelpful if you're trying to decide what to learn next month. The framework maps the terrain well. It just can't tell you which path to take when several look equally viable—or equally uncertain.

Marcus Chen-Ramirez is a Senior Technology Correspondent at Buzzrag