AI is transforming how engineers get hired — and Speak is proof
Artificial intelligence has come to shake up the hiring process at tech companies, and Andrew Hsu, CTO and co-founder of Speak, is feeling it firsthand. Speak is an AI-powered language learning startup with about 150 employees and 60 engineers, and the company decided to completely rethink how it evaluates and hires software engineering talent — for reasons that go way beyond a simple internal process update.
It’s no exaggeration to say that software engineering has hit a major turning point. Since December of last year, AI models have evolved so much that today 80% of coding tasks are already handled by coding agents, according to Andrew himself. What does that mean in practice? The modern engineer has stopped writing code all day and shifted to communicating with agents to get the work done.
It sounds subtle, but it changes everything.
With this new reality knocking at the door, Speak made a bold decision: it paused all individual contributor engineer hiring for two weeks just to rethink from scratch how to properly evaluate talent in this new landscape.
Goodbye, LeetCode. Goodbye, technical screening questions that any AI model can answer in seconds.
What replaced them? A hiring process that puts candidates face to face with tools like Claude Code and OpenAI’s Codex, evaluating not what they type, but how they think, make decisions, and build alongside an agent.
In this article, you’ll learn how this shift happened, what Speak is now looking for in the engineers it hires, and why this transformation isn’t an isolated case — it’s the direction the entire market is heading. 🚀
What changed in software engineering with coding agents
For years, the hiring process at tech startups followed a well-known formula: the candidate received an algorithmic problem, sat in front of a code editor, and demonstrated they could sort arrays, traverse graphs, and optimize loops. It was the ritual. LeetCode became a kind of universal language in the engineering market. But that language started losing its meaning very quickly, because artificial intelligence agents themselves began mastering that vocabulary better than any human could in a one-hour interview.
Andrew Hsu made it clear that Speak never adopted Big Tech-style algorithmic tests. The company always preferred coding problems closer to real-world scenarios. Still, even that model became outdated. Traditional technical screening questions were completely eliminated because the models got good enough to answer all of them without breaking a sweat. Testing a candidate with those kinds of questions stopped being a test of engineering skill — it became, at best, a test of who could copy and paste the fastest.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
The arrival of coding agents wasn’t gradual — it was a rupture. Tools like Claude Code, Codex, and other solutions powered by large language models started handling not just isolated snippet generation, but complete development tasks: writing entire functions, identifying bugs, suggesting architectures, refactoring legacy codebases, and even creating automated tests. The engineer who used to spend hours coding a feature now spends that same time reviewing what the agent produced, adjusting prompts, validating logic, and making product and architecture decisions that no AI can handle on its own — at least not yet.
This new landscape created a massive gap between what companies were asking for in interviews and what they actually needed day to day. Evaluating a candidate by having them solve an algorithm challenge became almost a joke, because the first thing any professional would do in a real environment is open the agent and ask it to solve the problem. The skill that matters now is no longer writing clean code from memory — it’s working alongside the machine intelligently, knowing when to trust it, when to question it, and when to step in. 🤔
How Speak rethought hiring from the ground up
When Andrew Hsu and the Speak team realized their hiring process had become obsolete, the decision was radical: pause everything. No new engineering hires for two full weeks while the company redesigned every step of the evaluation. That’s rare to see. Growing startups rarely stop to breathe, especially when they have open seats in engineering. But the logic behind the decision makes perfect sense — hiring by the wrong criteria would be worse than not hiring at all, because it would bring someone onto the team with a profile that wouldn’t fit the way the company was already working at full speed.
An important detail Andrew made sure to emphasize: Speak did not reduce its headcount targets for the year. The company isn’t shrinking its engineering team. The goal is to find engineers who know how to work in this new style, not simply cut positions because of automation. That’s a critical distinction, because it shows that demand for professionals still exists — what’s changing is the profile required.
The new process created by the startup puts candidates directly in contact with the AI tools they’d actually use on the job. Instead of an artificial whiteboard exercise or a coding platform challenge, the candidate gets a genuine problem — the kind that comes up in Speak’s daily workflow — and needs to solve it using Claude Code, Codex, or whatever other tools they have at their disposal.
The evaluators aren’t looking at whether they nailed the syntax or used the most efficient data structure. They’re watching how the candidate formulates prompts, how they interpret the agent’s responses, how they decide what to accept and what to discard, and how they drive their reasoning toward a functional, well-thought-out solution.
Take-home projects — the ones candidates complete on their own time — still exist, but now with a crucial difference: candidates are expected to use coding agents as much as possible. On top of that, the on-site stage was redesigned to include live coding with agents in a room with the evaluators. Afterward, the candidate is asked about what they built, why they made certain choices, and what trade-offs they considered along the way.
Beyond that, Speak started placing much greater value on traits that were previously considered soft skills and took a back seat to pure technical performance. Clear communication, critical thinking, the ability to ask the right questions, and speed of learning all moved up the priority ladder. After all, if the agent writes the code, the engineer needs to be excellent at everything the agent still can’t do — and that’s much more about human judgment than memorizing algorithms. 💡
The Engineer 1 and Engineer 2 framework
One of the most interesting parts of Andrew Hsu’s vision is the framework he created to explain internally the difference between two types of professionals in this new landscape. He calls them Engineer 1 and Engineer 2, and the distinction between the two is what defines who will thrive and who will fall behind.
Engineer 1 is the professional who adopted tools like Claude Code and Codex for about 90% of what they do. Sounds advanced, right? But at its core, this engineer is using AI only to move faster. Their mindset hasn’t changed. They still think about the work the same way they did before — the difference is they now have an assistant that types quicker. The results are better, sure, but the potential is being underutilized.
Engineer 2, on the other hand, understood that the work itself has changed. This professional isn’t just using the agent to speed up tasks — they’re building the environment and systems that make the agent perform better. They create feedback loops, add capabilities to the agent, set up automated verification methods, and treat AI as a team member that needs context, tools, and clear instructions to deliver the best possible results.
In practice, Engineer 2 went from feature implementer to agentic systems architect. It’s a very different job, and Andrew believes the productivity gap between these two profiles will only widen over time. Speak wants to hire Engineer 2s — and the hiring process was built to identify exactly that profile.
What AI startups are looking for in engineers now
Speak isn’t alone in this shift. Other startups and artificial intelligence companies are revisiting their hiring criteria and reaching similar conclusions. Companies like Canva and Meta already allow candidates to use AI tools during the interview process, signaling that the market is converging toward a new standard. The profile emerging from this transformation is what some are already calling the augmented engineer — a professional who uses agents as an extension of their own delivery capacity.
In practice, the most valued skills in this new context are quite different from the ones that dominated job requirement lists two or three years ago. Today, what an AI startup wants to see in an engineer includes a combination of technical and behavioral competencies that rarely appeared together in the old evaluations:
- Ability to reason about complex systems — understanding how the pieces fit into the whole, not just how to write an isolated function
- Skill in working with agents productively — knowing how to build effective prompts, interpret outputs, and iterate quickly
- Sharp critical thinking — questioning what the agent produces, catching subtle errors, and not accepting the first answer as absolute truth
- High-quality technical communication — describing problems with precision, both to humans and to machines
- Accelerated adaptability — the evolution cycle for AI tools is extremely fast, and the engineer needs to keep up without losing the thread
- Initiative and intellectual curiosity — according to Andrew Hsu, this is exactly the profile that has stood out the most in the world of agentic engineering
This set of traits completely changes how interviews need to be structured. A conversation about past experiences and a timed technical test reveal none of this. What does reveal it is putting the candidate into action, in an environment close to the real thing, and watching how they behave when they need to make decisions with incomplete information, with an agent by their side, and with a problem that doesn’t have a single right answer.
Systems design and architecture carry even more weight now
A point Andrew Hsu highlighted that deserves special attention: with agents taking over a large chunk of implementation work, systems design and architecture skills have become even more important than they already were. It seems counterintuitive at first — if AI does more things, engineers need to know less, right? Wrong.
The opposite is actually happening. When the agent writes the code, someone needs to make sure that code makes sense within the larger system. Someone needs to define the structure, the contracts between services, the scalability patterns, and the trade-off decisions that will affect the maintenance of that software for months or years. This high-level work is fundamentally human and requires a type of thinking that current AIs still can’t replicate consistently.
Speak has always looked for engineers who were more builders than programmers — people whose professional identity wasn’t tied to the beauty of their code or the elegance of the system, but to the real impact of what they built for users. That philosophy, which used to be a cultural differentiator for the company, has now become a concrete competitive advantage in adapting to this new landscape.
The impact on career leveling
Another fascinating aspect of this transformation is its impact on seniority levels. Andrew mentioned something that probably made a lot of people stop and think: with agents, a junior engineer can have more output than a principal engineer. This completely scrambles the traditional leveling logic that tech companies have used for decades.
If the volume of code produced is no longer a reliable indicator of seniority — because anyone with a good command of agents can produce tons of functional code — then what defines a senior engineer? The answer comes down to judgment, systems thinking, the ability to make architectural decisions under pressure, and the skill to mentor other professionals in working better with these new tools.
Speak had to rethink how it evaluates candidate levels precisely because of this shift. Output volume no longer tells the same story it used to, and traditional productivity metrics are being recalibrated in real time. It’s a transformation still in progress and one that will likely spark a lot of discussion in the market over the coming months.
Why this change isn’t an isolated case
What’s happening at Speak is a barometer for the entire market. Andrew Hsu stated that the whole industry is moving in this direction, and that people with the agentic engineering mindset will be the ones who thrive. Hiring in tech companies is going through one of the biggest transformations in its history, driven by tools that just two years ago were still considered novelties. Now they’re infrastructure.
Over the last three months, according to Speak’s CTO, the rate of change has been unbelievable. The fundamental work of software engineers has shifted entirely, and companies that don’t adjust their evaluation processes will end up hiring professionals misaligned with the reality of the job today. It’s like selecting drivers based on their horseback riding skills — technically related, but completely out of context.
For engineering professionals, the message is clear: mastering generative AI tools and coding agents is no longer a differentiator — it’s a prerequisite. And beyond mastering the tool, what really matters is developing the right mindset — that of Engineer 2, who understands that the work has changed and embraces that change with curiosity, initiative, and systems thinking.
Those who figure this out first will come out ahead — both the companies that redesign their hiring processes and the engineers who learn to truly work side by side with agents. The era of agentic engineering has already begun, and Speak is just one of the first to make that crystal clear. 🚀
