This as-told-to essay is based on a conversation with Andrew Hsu, the cofounder and CTO of Speak, an AI-powered language learning app. The following has been edited for length and clarity.
I’m a cofounder at Speak, a growth-stage AI-native startup. We’re building an AI language tutor using the power of AI and speech recognition models. The whole company is about 150 people, and we have about 60 engineers.
Coding agents weren’t viewed as a central tool for actual software engineering teams until early December of last year. At that point, the models became good enough that 80% of the coding work software engineers had to do could be handled by an agent. Their day-to-day changed from writing code to primarily communicating through coding agents.
Over the past three months, the rate of change has been unbelievable, and the fundamental job of software engineers has totally changed.
We’re rethinking how we hire and interview engineers to look for what we call agentic engineering skills, in addition to, many of the skills that we’ve always looked for in strong software engineers.
We’ve been deeply rethinking how we do interviews. In the past, engineering has always been the main bottleneck in the product development cycle. The most expensive, high-paid part of the pipeline has changed, and engineering is no longer the bottleneck. We have a much larger capacity now for people who know how to use these coding agents to the maximum extent.
We had a two-week slowdown in individual contributor engineering hiring while we figured out how to evaluate people and how we think about leveling given a young engineer can have more output than a principal engineer. Now we have a new process.
We have not revised our headcount goals for this year. We’re not planning to shrink our engineering team; we’re just looking for engineers who can work in this new style.
The technical part of the interview
We’ve never done the LeetCode, algorithmic tech screens that Big Tech companies are known for. We try to do more real-world coding problems.
All the traditional tech screen coding questions are gone entirely, because the models are good enough that they can answer every one of those questions. It’s no longer a test of engineering skills.
The more interesting test is: Can you build a feature using Claude Code or OpenAI’s Codex? Can you talk through and solve problems and plan and debug through these agents? Can you do it in a way that amplifies your output, while keeping the quality bar high?
We still assign take-home projects to candidates, but they now use coding agents to the maximum extent. We’re also retooling the on-site process to incorporate agentic coding in a live room. Then we ask them about what they’ve built, and why they made certain choices or trade-offs.
An agentic engineering mindset
There’s a new agentic engineering skillset that we’re testing for, and we only want to hire people who are good at it.
We’re looking for engineers who have the agentic mindset. I talk about this internally with a specific framework. “Engineer one” is a software engineer who uses Claude Code or Codex for about 90% of what they do. Fundamentally, though, they’re using it to move faster, but nothing has actually changed in their mindset for how they build.
“Engineer two,” on the other hand, actually realizes that their job has changed, from working on the implementation of the feature to building the environment and systems to make an AI agent work better. They’re building that feedback loop and adding capabilities to the agenda so that it can do all of the work, including verifying the feature that it just built.
It’s a very different job, and I think that people who understand the “Engineer two” mindset will be much more productive than “Engineer one.” I think that the gap will continue to grow.
The profile that we’ve seen really succeed in this agentic engineering world is engineers who are extremely proactive and intellectually curious.
We’ve been pretty lucky, because we’ve always looked for engineers who are more builders than programmers, and who don’t have their identity attached to the beauty of the code and the system, but whether what they’re building is actually solving a problem for users.
In the interviews, we want to see that the candidate is — in combination with all the traditional strengths of a human software engineer — all-in on building with agents. We’ve also placed even more emphasis on traditional skills, like system design and architecture. The higher-level aspects of building a technical system are even more important now.
The entire industry is moving in this direction, and the people with the agentic engineering mindset are going to be the ones who thrive.
