The most important hiring fable in tech is that AI is making junior engineers out of date. The logic appears sound: AI can write code, so why rent folks to write down extra of it? Junior engineers have at all times discovered on the job, fixing bugs and transport small options till they construct actual judgment. Now that coding brokers can deal with a lot of that work, corporations assume they’ll lower off the underside rungs of the profession ladder.
At Snowflake, we’re making the alternative guess. Snowflake is a knowledge and synthetic intelligence platform that helps 1000’s of organizations like Goldman Sachs, Hyatt, Kraft Heinz, and United Leases construct with and get worth from their knowledge, apps, and AI. Junior engineers make up 70%–80% of our present hiring.
We construct and run infrastructure that international enterprises rely on. As a result of reliability and efficiency matter a lot, we will’t hand these duties to a junior engineer or a coding agent and stroll away.
The expertise hierarchy is being rewritten
I’ve spent most of my profession in environments that rewarded uncooked implementation velocity. At Google and YouTube, and later at my startup, Neeva, the engineers who stood out had been those who might maintain numerous context of their head, determine options rapidly, and churn out code quicker than anybody else.
Whereas deep system reasoning nonetheless issues, velocity is now not the bottleneck. I’ve seen firsthand how coding brokers make workers considerably extra productive: 95% of Snowflake’s engineers use coding brokers weekly, and we’re pushing towards 100% day by day utilization. I usually inform my groups that coding brokers are like Slack and Gmail. You wouldn’t be okay with “I solely use Slack as soon as every week,” and the identical needs to be true for coding brokers. They need to turn into that elementary in day-to-day engineering work.
With the proliferation of coding brokers, the scarce talent is now not writing code. It’s realizing what needs to be constructed, decomposing the issue, making a plan an agent can execute, catching when the agent is confidently fallacious, and connecting the implementation again to the client downside. This is the reason a 10x engineer doesn’t mechanically turn into a 100x engineer simply by adopting AI instruments. Engineers of all ranges might want to rethink their workflow.
For senior engineers, this would possibly imply pairing deep methods data with AI-native methods of working. For engineering managers, good efficiency now means setting the technical course and serving to groups break down issues, not writing probably the most code. Junior engineers usually have an edge right here, since they’re choosing up these expertise from day one, however skilled engineers convey judgment that solely will get extra useful when paired with these instruments.
Curiosity is turning into extra useful than coding velocity
When interviewing engineers in the present day, I search for curiosity and robust engineering judgment. I need individuals who cause from first rules, ask sharp questions, and have deep buyer empathy. These instincts don’t map neatly to tenure. We’ve discovered internally that it’s a mistake to correlate AI effectiveness with seniority. Many junior engineers are transferring with astonishing velocity just because they’re naturally curious and studying AI-native workflows from the start.
Nonetheless, it’s additionally true which you can’t drop a junior engineer into a large code base with an agent and hope for one of the best. Now we have to explicitly educate them a brand new manner of constructing software program with brokers to maximise their affect.
I usually say that English is cheaper than code. Engineers ought to spend time explicitly drafting the method and assumptions, and defining interfaces earlier than producing the implementation. It’s vital to interrupt work into activity graphs, not checklists, and drive the agent to write down the verifiers and exams earlier than writing the code.
We’ve institutionalized this mindset. I requested our engineering group to take a full week off from regular work to focus solely on AI upskilling. Our leaders went workforce by workforce, serving to them study these instruments and patterns. Some folks pushed again, saying they had been too busy with launches and deadlines. My response was that that is the highest-leverage factor you could be doing proper now to remain related for the subsequent decade. Nearly universally after that week, folks got here again saying they want they may do it extra usually.
The engineer turns into the agent workforce lead
The perfect engineers I see in the present day aren’t writing each line of code. As a substitute they’re tech leads for groups of brokers. Inside Snowflake’s engineering workforce, I can level to a small however rising group — round 5%–10% of the group — who’re working at this frontier and managing a number of brokers in parallel.
The human engineer remains to be accountable. They resolve what issues, sequence the work, validate the outcomes, and make tradeoffs, however the brokers are writing a lot of the code.
We’re attempting to develop this group. My aim is to multiply fivefold, in a matter of months, the variety of engineers working this manner. We’re doing this by empowering these engineers to show the folks subsequent to them. That is the subsequent frontier, and we shouldn’t wait 10 years to show it.
Smaller groups, quicker loops
Once I was at my startup, we might transfer rapidly as a result of we had been small. At Snowflake, we have now 1000’s of engineers and prospects who rely on us for crucial workloads. Traditionally, that meant slower iteration cycles, and extra processes.
AI adjustments that. We are able to now take a small group of engineers—4 or 5 folks with the correct mix of expertise—and transfer with startup-like velocity whereas sustaining enterprise-grade high quality. Inside these items, the composition of expertise issues greater than the roles. You want somebody with robust product instinct, somebody who understands the system deeply, and somebody who’s on the chopping fringe of utilizing brokers. Typically, these roles blur. Product managers write code, engineers interact with prospects, and designers construct working prototypes.
To thrive on this atmosphere, engineers want alternatives to develop these cross-functional muscular tissues all through their careers. Beginning early offers folks extra time to compound these expertise.
The final word threat
Many corporations are about to make an enormous mistake. In the event you cease hiring junior engineers as a result of AI can do a few of their work, your short-term margins would possibly look nice. However 5 years from now, when the trade desperately wants senior leaders who’re native to this manner of working, these leaders received’t exist. We’re already seeing a return of the “graybeard” engineers, as a result of judgment and methods instinct are nonetheless paramount. The distinction is that tomorrow’s most precious leaders will mix these qualities with AI-native methods of constructing.
These future leaders have to come back from someplace. You possibly can’t skip the formative, early-career years the place judgment, methods considering, and buyer empathy are honed. You possibly can’t construct a sturdy engineering group by eliminating the underside of your expertise pipeline.
I’m not fearful that AI will change junior engineers; I’m fearful that corporations will resolve they’re now not price creating simply after we want them most.
