Why AI Startups Are Skipping the Experience Filter

Hiring has a years-of-experience problem. It’s been there for a while, but it’s getting harder to ignore in AI.

The filter exists because it’s a proxy. Hiring managers can’t interview everyone. Filtering for five years of experience cuts the applicant pool fast, and it loosely correlates with someone who’s shipped real things, debugged real problems, and survived at least one production incident. It’s not perfect, but it’s cheap to apply.

In AI specifically, it’s starting to filter out the wrong people.

Some of the strongest candidates for ML engineering roles are graduating from CS programs this spring. They’ve spent two or three years doing research in university labs, interned at AI companies where they shipped real code, and they know the tools and techniques driving the current wave of AI products better than most people interviewing them. They also have zero years of post-graduation industry experience, which means they often don’t make it through the initial screen.

The companies that have noticed this are quietly adjusting.

What “Junior” Means in an AI Context Has Changed

The traditional junior engineer story: someone graduates, joins a team, spends a year or two learning how production systems work, and slowly earns more responsibility. The early years are mostly about onboarding to professional engineering: version control conventions, code review culture, deployment pipelines, incident response. Technical skills usually aren’t the bottleneck.

ML and AI work don’t fit that model cleanly.

There isn’t a standard “how production ML works” curriculum baked into five years of industry experience, because the tools are new enough that senior engineers are often learning the same things new grads are learning. Just slower, because new grads came out of environments where this was the actual focus for years.

A student who spent two years at a university ML lab working on language model training knows things that are useful to an AI startup starting week one. A well-practiced explanation of a project you barely understand will fall apart when someone who knows the field asks follow-up questions. But if you’ve actually done the work, that conversation goes differently.

Some new grads from top programs are functionally mid-level in AI-native work. They just don’t have the industry tenure that would normally get them labeled that way.

The Schools This Is Actually True For

This doesn’t apply uniformly across CS graduates, so let’s be specific.

The programs producing the graduates AI startups are recruiting hardest are the ones with serious research infrastructure: MIT, Stanford, Carnegie Mellon, Columbia, NYU, UC Berkeley, Cornell, UIUC, University of Washington, Georgia Tech. The reputation matters less than what students at those schools have actually done before graduation.

At these programs, a student can spend three years in a lab doing genuine ML research: contributing to real projects, writing code that runs on real hardware, co-authoring papers other researchers actually read. They intern at AI companies two or three times. They ship personal projects that real users use.

That profile isn’t every student at every school on that list. But it’s common enough that hiring managers who have met a few of these graduates now pay more attention to what’s in the portfolio than what the graduation date says.

The question starting to replace “how many years of experience do you have” is: “walk me through the most technically complex ML project you’ve worked on.” A strong new grad from MIT can hold their own in that conversation against someone with five years in industry. Sometimes more than hold their own.

The Comp Structure Is Part of Why This Works

There’s a business case worth being transparent about.

Senior ML engineers at AI-stage startups aren’t cheap. A strong candidate with five or six years of relevant experience, coming from a big tech company, might cost $200k to $220k in base, even more once you factor in total comp expectations.

A new grad with a genuinely strong AI/ML background can come in at $140k to $160k. The output difference for specific AI-native tasks is often smaller than that comp difference. Senior engineers bring judgment and architectural instinct and production experience that matters enormously in some contexts. For the execution layer of ML work, though, the performance gap is narrower than the price gap implies.

Series A and B companies are watching burn closely. Making a few smart new grad hires for AI-core work lets them stretch their runway while still getting the technical caliber they need. That’s not a compromise, it’s a recalibration of what experience actually predicts for the specific work that needs doing.

What Adaptability Buys You at the Early Stage

There’s a version of this conversation that’s only about skills, and it misses something real.

Senior engineers from big tech companies often arrive with strong opinions about how things should be built. Some of those opinions are right and useful. Some are baggage from contexts that don’t apply at a 20-person startup. It’s genuinely hard to know which is which before you hire someone.

New grads are more likely to pick up whatever stack the company runs and just use it. They’re not mentally comparing everything to how a much larger team with much more infrastructure handled it. They show up and figure out how things work here.

At a startup still figuring out its own patterns, that has real value.

The tradeoff is also real: new grads need more mentorship, make mistakes that come from inexperience, and take longer to develop the engineering judgment that comes from shipping a lot of things over time. Good startups hire new grads knowing this and invest in the onboarding. The ones that expect a new grad to perform like a senior from day one are going to have a bad time.

The Pipeline Is More Specialized Than People Realize

If you’re a new grad trying to find these opportunities, one thing is worth knowing: most of the AI startups making bets on strong new grads aren’t posting those jobs broadly.

The hiring happens through tighter channels. Research advisor referrals. Former labmates who joined a company and vouched for you. Talent platforms that specialize in engineering-to-startup matching and have relationships with both sides. Early-stage AI companies have learned that their best candidates don’t show up through Indeed.

Applying cold through a job board, you might be technically qualified and genuinely strong and still hear nothing back, not because you’re wrong for the role, but because you entered through a door that doesn’t connect to where the actual decisions get made.

The network you’re already inside, your lab, your research advisor, the people two years ahead of you who are already at these companies, matters more than it would for a traditional job search. That’s unfair in some ways. It’s also just how the market operates right now.

If You’re Finishing Your Degree This Spring

The old calculus was: get the big company logo on your resume, then you’re hireable. That made sense when experience at Google or Meta was a genuine proxy for whether you could do real engineering work.

AI scrambled that proxy. The relevant skills for AI-native work aren’t built at big companies, they’re built in research labs and personal projects and the process of actually training models. A lot of big tech ML work is specialized enough that it doesn’t transfer directly to what an AI startup needs from a new hire.

If you’re finishing a CS degree from a strong program with real ML work behind you, have the conversation with the AI startup market before defaulting to FAANG. Not because FAANG is a bad choice, it’s genuinely great for a lot of reasons, but because the AI startup path is more viable for strong new grads than most people in your graduating class have been told.

The offers are real. The work is interesting. The equity is a gamble, but not a worthless one. And the path to technically deep problems is a lot shorter than it would have been five years ago.

That’s worth knowing before you sign the first offer you get.

Fonzi AI matches vetted software engineers with VC-backed AI startups in SF and NYC. If you’re a recent grad with a strong ML or AI background and want to explore what’s out there, check out our Match Day here.