If code is easier to generate, what roles become the real bottleneck now?
by•
Being an engineer myself, I see that many people are outputting more code than ever. Some of it being generated vs. written, the volume of code being outputted has definitely risen.Â
There seems to be a shift from "is this something we can build?" to "should we build this and ship it?"
For people who have been recruiting or looking to recruit recently, what roles are you hiring for?Â
Role title (I've seen a rise in PM's and Designers personally)
What exactly can they do that AI can't (yet)
Specific signal that you look for when hiring
Are you moving headcount from one type of role to another?Â
Would be interested to hear from other growing teams! 🚀
433 views
Replies
Cal ID
I see this happening in real time: writing code is easy, but building the right thing is hard.
Teams are hiring more PMs and designers because AI doesn’t bring clarity yet. Even the PMs today are judged on how quickly they can validate ideas, not just ship specs.
If you’re hiring, look for people who can prototype fast, break down ambiguity, and rally teams around a shared north star. In my opinion, designers are the last line of defense against features that work but feel wrong. AI can’t replicate that instinct for what feels good in the hands of real people. (I've experienced this myself)
Onyx
@sanskarix We focus on hiring engineers that take full ownership and can talk to our customers to figure out the pain points and really understand what it is that they are looking for
@sanskarix @justin_tahara It should be pretty hard to find such, is it?
Onyx
@sanskarix @kumala3 Well hiring is never easy but when you are small and need to find the right people, you take extra time and care in finding the right engineer that fits your team
@sanskarix @justin_tahara Definitely!
Triforce Todos
For us, the signal we look for is judgment: someone who can say 'no' to 9/10 good ideas and focus on the one that moves the needle.
Onyx
@abod_rehman Given the AI will most definitely agree to your statements, that makes a lot of sense
@abod_rehman @justin_tahara AI agrees to most of your statements and ideas no matter how good/bad they are by default without giving them a persona.
It sucks and the solution to this I us providing the persona at the beginning explaining the logic behind every judgement and context to know what aspects to consider when judging.
Onyx
@abod_rehman @kumala3 Yup definitely agree here. Even within our tool Onyx, we have prompting that allows you to ensure that the persona of the chat is captured and correctly interacts with your suggestions
Cal ID
AI makes code easy.
The real bottleneck is now making the right decisions, building things people want, and shaping product experience.
If you’re watching what’s happening inside fast-growing teams, the hottest hires are PMs and designers. Because while AI can generate code, it can’t sense what’s valuable, spot user needs, or make tough calls about what not to build.
PMs are crucial for framing the why behind features, validating ideas, and focusing teams on problems that actually move the needle. Designers are essential because they bring the instinct for what feels right to real users, something AI can’t replicate, at least not yet.
When we hire today, the signals we care about are product sense, judgment, and an impact bias – can this person say “no" to most ideas, spot gaps in workflows, and really understand what drives adoption and retention? It’s less about “can we build this?” and more about “should we build this?”
So, the big shift is this: AI writes code, but people still make products. The new bottleneck is product, not engineering– that’s where the real leverage is now.
Onyx
@sanskarix Do you think hiring for engineers that know when to say no and when to not hit the autocomplete button and actually make critical decisions will be impactful as well?
I agree that PM's are what are going to continue to drive the next set of decision making but there is also a bit of play where engineers that are good and understand long term engineering practices will know when and what decisions will make or break a couple years down the line.
Cal ID
@justin_tahara Absolutely... Even with AI, someone still has to draw the line between “can we ship this?” and “should we ship this?”
Engineers who understand architecture and the long-term impact of their choices are the hidden leverage.
It's pretty evident that the best teams invest in engineers with strong product sense and the confidence to question the AI’s suggestions.
Onyx
@manu_goel2Â I see, so you're saying that we still need engineers but engineers that have built systems that scale and understand the decisions and suggestions that the AI model is giving them
@justin_tahara Yes, lesser number of highly experienced engineers ...to review. However, the junior guys have to be very high on reasoning/logical skills but can be low on coding experience...Minimal coding experience still needed.
CalPulse
AI has the potential to assist us, but it's not yet mature enough to effectively handle intricate details. Most AI models still lack the necessary depth and nuance in their execution. 🤔
Onyx
@peilan_qin So you're still in the camp that we have yet to unlock the full potential of AI code assist and that it could become even better and even more of a helper when it comes to coding
CalPulse
@justin_tahara I believe that AI still has significant room for growth in code assistance. While it performs well for simpler tasks, it remains limited when handling complex details. As a marketer, I focus more on how to leverage these technologies to enhance efficiency and user experience. I look forward to a future where AI can better support developers, allowing them to focus on more creative work :)
I've been a Quality Analyst for 12 years, and still going. Code is becoming easier to generate, but "easy" often means it has gaps in quality. Pieces failing to work, back-end and 3rd party connections dropping auths and preventing data from being transmitted.
So this actually creates a quality bottleneck. The act of code becoming easier to generate means you have to use a fine-tooth comb strategy in refactoring... Or you need to implement QA processes to keep it functioning, your demos rolling, and your clients happy before even reaching production.
But that also leads into an engineering bottleneck. Time fixing issues after a production launch means you don't have time to work on new features and improvements because you're running back to put fires out. And if the engineers are also expected to be the QA then that time has to be used testing and analyzing the system, not pumping out features and functionality.
Product Management, UI/UX Design, and Business Analyst all typically focus on building requirements, strong user interfaces, and analyzing the business and project flow. But the one missing in the person devoted entirely to the quality of the product, not the project. So a QA/SDET is a strong addition to any team like that.
@ryan_from_ashlight 100% agree about the low quality of AI Generated code because I have been using GitHub Copilot with Claude Sonnet 4 for the past 4 months and noticed that code quality even for frontend is really poor. Weak, ugly UIs, mixed up auth logic and duplicated code.
I am constantly working to make it better by creating new rules for Agents that write code but they still lack that sense of what humans have built-in and is quite hard to explain.
Onyx
@ryan_from_ashlight @kumala3 Have you tried using Codex or Claude Code instead? I think there are definitely ones that perform better for me personally.
@ryan_from_ashlight @justin_tahara I had tried both Claude Code and Codex, however, I am still sticking to GitHub Copilot because of the great usability that I like with it. I want to use CC more, though - get used to it and compare how it does in more detail. Do you have any insights on GitHub Copilot vs Claude Code? I'd love to hear some if you have any!
@kumala3 @justin_tahara I don’t have a strong take on Claude vs. Copilot. I track outcomes: first-pass CI rate, flake trends, and escaped defects. If those improve, awesome—I still keep human review in the loop.
The end consumer is human, after all.
Onyx
@sheraz_abdul_hayee We are seeing some AI products building out end to end systems but how do you think this will impact teams?
Onyx
@stanleyamaziro Can engineers stand out more if they can also work with customers and figure out the needs themselves? Especially in a smaller team where you may not have a PM, I'm sure the engineer who can talk with customers and communicate ideas clearly will stand out more in the future
@stanleyamaziro @justin_tahara If the engineer is capable of talking with customers, listening and then translating it into points that enable them to understand what customers need/want/would be nice to see.
Onyx
@stanleyamaziro @kumala3 I think this is just going to be the skill set that is required for engineers moving forward. No more just coding but more understanding what you are coding + directly understanding the needs of the customer
@stanleyamaziro @justin_tahara I agree on this because PMs (Project/Product Managers) can communicate with AI to build some features (it's not there yet, however, moving forward, looking 1-5 years ahead, AI will create really good software because it will be trained on the best Software Engineers, practices, and workflows). Understanding what you need to do comes from understanding the needs of the customer, so it would become the essential trait that Software Engineers will need to acquire, which is by default in all Entrepreneurs is.
Not hiring right now, but if I were starting a company today, this is the team I'd build from day one. Everyone would need to ship. No passengers.
A designer who pushes UX boundaries. Someone who doesn't just polish, but experiments with new interaction patterns. Think poke.com personality, energy: curious, playful, opinionated. They prototype in code.
An engineer who understands systems. Not just writing code, but shaping the architecture so an LLM can step in and help. They share mental models with the machine: name things clearly, structure code predictably, and design so future agents (or teammates) can move fast without friction.
A data analyst with product instincts. Someone who doesn't wait for dashboards. They get creative about what to measure, how to validate whether something works, and what "working" even means. They feed the loop.
Someone who owns storytelling and social. You can't afford to be invisible. This person isn't just scheduling posts; they're shaping the narrative as you go. They help the team stay legible to the outside world while you're still figuring things out.
We ran an internal hackathon with a similar mindset. The goal was to get more people building end-to-end. That shift from relying on others to rigging things yourself completely changes how you see the product.
So yeah, if I were hiring, I'd optimize for people who take initiative, learn fast, and aren't afraid to touch the whole stack. Titles matter less than momentum.
Totally agree — code is abundant, clarity is rare. The real edge isn’t shipping faster, it’s deciding what’s worth shipping at all. That’s why PMs and designers who can cut through noise, validate quickly, and bring focus are becoming even more valuable.