The biggest hiring myth in tech is that AI is making junior engineers obsolete. The logic seems sound: AI can write code, so why hire people to write more of it? Junior engineers have always learned on the job, fixing bugs and shipping small features until they build real judgment. Now that coding agents can handle much of that work, companies assume they can cut off the bottom rungs of the career ladder.
At Snowflake, we’re making the opposite bet. Snowflake is a data and artificial intelligence platform that helps thousands of organizations like Goldman Sachs, Hyatt, Kraft Heinz, and United Rentals build with and get value from their data, apps, and AI. Junior engineers make up 70%–80% of our current hiring.
We build and run infrastructure that global enterprises depend on. Because reliability and performance matter so much, we can’t hand those responsibilities to a junior engineer or a coding agent and walk away.
The talent hierarchy is being rewritten
I’ve spent most of my career in environments that rewarded raw implementation speed. At Google and YouTube, and later at my startup, Neeva, the engineers who stood out were the ones who could hold a lot of context in their head, identify solutions quickly, and churn out code faster than anyone else.
While deep system reasoning still matters, speed is no longer the bottleneck. I’ve seen firsthand how coding agents make employees significantly more productive: 95% of Snowflake’s engineers use coding agents weekly, and we’re pushing toward 100% daily usage. I often tell my teams that coding agents are like Slack and Gmail. You wouldn’t be okay with “I only use Slack once a week,” and the same should be true for coding agents. They should become that fundamental in day-to-day engineering work.
With the proliferation of coding agents, the scarce skill is no longer writing code. It’s knowing what should be built, decomposing the problem, creating a plan an agent can execute, catching when the agent is confidently wrong, and connecting the implementation back to the customer problem. This is why a 10x engineer doesn’t automatically become a 100x engineer just by adopting AI tools. Engineers of all levels will need to rethink their workflow.
For senior engineers, this might mean pairing deep systems knowledge with AI-native ways of working. For engineering managers, good performance now means setting the technical direction and helping teams break down problems, not writing the most code. Junior engineers often have an edge here, since they’re picking up these skills from day one, but experienced engineers bring judgment that only gets more valuable when paired with these tools.
Curiosity is becoming more valuable than coding speed
When interviewing engineers today, I look for curiosity and strong engineering judgment. I want people who reason from first principles, ask sharp questions, and have deep customer empathy. These instincts don’t map neatly to tenure. We’ve learned internally that it’s a mistake to correlate AI effectiveness with seniority. Many junior engineers are moving with astonishing speed simply because they’re naturally curious and learning AI-native workflows from the beginning.
However, it’s also true that you can’t drop a junior engineer into a massive code base with an agent and hope for the best. We have to explicitly teach them a new way of building software with agents to maximize their impact.
I often say that English is cheaper than code. Engineers should spend time explicitly drafting the approach and assumptions, and defining interfaces before generating the implementation. It’s important to break work into task graphs, not checklists, and force the agent to write the verifiers and tests before writing the code.
We’ve institutionalized this mindset. I asked our engineering organization to take a full week off from normal work to focus entirely on AI upskilling. Our leaders went team by team, helping them learn these tools and patterns. Some people pushed back, saying they were too busy with launches and deadlines. My response was that this is the highest-leverage thing you can be doing right now to stay relevant for the next decade. Almost universally after that week, people came back saying they wish they could do it more often.
The engineer becomes the agent team lead
The best engineers I see today aren’t writing every line of code. Instead they are tech leads for teams of agents. Inside Snowflake’s engineering team, I can point to a small but growing group — around 5%–10% of the organization — who are operating at this frontier and managing multiple agents in parallel.
The human engineer is still accountable. They decide what matters, sequence the work, validate the results, and make tradeoffs, but the agents are writing most of the code.
We’re trying to expand this group. My goal is to multiply fivefold, in a matter of months, the number of engineers operating this way. We’re doing this by empowering these engineers to teach the people next to them. This is the next frontier, and we shouldn’t wait 10 years to teach it.
Smaller teams, faster loops
When I was at my startup, we could move quickly because we were small. At Snowflake, we have thousands of engineers and customers who depend on us for critical workloads. Historically, that meant slower iteration cycles, and more processes.
AI changes that. We can now take a small group of engineers—four or five people with the right mix of skills—and move with startup-like velocity while maintaining enterprise-grade quality. Within these units, the composition of skills matters more than the roles. You need someone with strong product intuition, someone who understands the system deeply, and someone who’s at the cutting edge of using agents. Often, those roles blur. Product managers write code, engineers engage with customers, and designers build working prototypes.
To thrive in this environment, engineers need opportunities to develop these cross-functional muscles throughout their careers. Starting early gives people more time to compound those skills.
The ultimate risk
Many companies are about to make a big mistake. If you stop hiring junior engineers because AI can do some of their work, your short-term margins might look great. But five years from now, when the industry desperately needs senior leaders who are native to this way of working, those leaders won’t exist. We’re already seeing a return of the “graybeard” engineers, because judgment and systems intuition are still paramount. The difference is that tomorrow’s most valuable leaders will combine those qualities with AI-native ways of building.
Those future leaders have to come from somewhere. You can’t skip the formative, early-career years where judgment, systems thinking, and customer empathy are honed. You can’t build a durable engineering organization by eliminating the bottom of your talent pipeline.
I’m not worried that AI will replace junior engineers; I’m worried that companies will decide they’re no longer worth developing just when we need them most.