# I was in OpenAI’s first intern cohort. Here’s what it taught me about becoming an AI-native engineer > Source: > Published: 2026-06-16 17:00:45+00:00 **TL;DR:** AI is making it easier than ever to build software that looks impressive in a demo. But after working in OpenAI’s first intern cohort, I learned that the real challenge is not just speed. It is judgment: knowing what to trust, what to test, and when a human still needs to stay in the loop. AI has made it easier than ever to create software that looks impressive. A prototype can be built faster. A codebase can be explored faster. A test can be generated faster. A confusing document can be summarized faster. For engineers, this is an enormous shift. But speed is not the same thing as judgment. That was one of the biggest lessons I took from my time at OpenAI. I was part of the company’s first intern cohort, and the experience changed the way I think about software engineering in the AI era. Before that, I thought a lot about [becoming a stronger programmer](https://thenextweb.com/news/difference-between-junior-mid-level-senior-developer-syndication). Afterward, I started thinking much more about becoming a better judge of systems: what works, what fails, what only looks correct, and what can actually be trusted. ## My path into AI was not linear I did not start my career with a perfect plan. I was born in Cairo and moved to Canada when I was 10. For a long time, I thought I would pursue medicine or forensic science. Computer science became interesting to me when I realized software was becoming one of the highest-leverage ways to build products, solve problems, and participate in the future of technology. Once I made that switch, I tried to put myself in environments where I could learn faster. My first software internship was unpaid. It was not at a famous technology company or a well-known AI lab. It was with a very early software project led by a Waterloo senior. I could not get a paid software role at first, so I took the opportunity I had and tried to turn it into the next one. That path eventually took me to Whatnot, then Verkada, and finally OpenAI. At the time, OpenAI was preparing its first intern cohort. I applied the day applications opened after a friend sent the link in a group chat. Because the cohort was new, there was no established playbook. Nobody could tell me exactly what the interviews would look like or what kind of background they wanted. The process surprised me. I expected the interviews to focus heavily on artificial intelligence. Instead, most of the process tested core software engineering: algorithms, system design, speed, clarity, and judgment. The bar was not just whether I could solve the problem. It was whether I could think clearly, communicate well, and move quickly under pressure. That ended up matching the culture I experienced inside the company. At OpenAI, the pace felt faster than what I expected from a large technology company. I shipped code on my first day. There was not a long onboarding period where every detail was explained before I could contribute. I had to learn quickly, ask good questions, and take ownership. The lesson was clear: speed matters, but judgment matters more. ## AI-native does not mean AI-dependent I use [AI coding tools](https://thenextweb.com/news/developers-refuse-work-without-ai-coding-productivity-paradox) regularly now. Earlier in my career, I was skeptical of how useful they could be for serious engineering work. That changed once the tools became good enough to help with real tasks: writing code, generating tests, exploring unfamiliar systems, summarizing information, and reducing tedious work. But the better these tools become, the more important fundamentals become. If you understand the system you are building, AI can make you much faster. If you do not understand it, AI can create the illusion of progress. It can produce code that looks right. It can give explanations that sound confident. It can solve the narrow example in front of you while missing the deeper issue in the system. That is why I think the next generation of engineers needs to become AI-native, but not AI-dependent. An AI-native engineer is not someone who blindly accepts model outputs. It is someone who knows how to use AI tools with taste and discipline. They know when to trust the output, when to question it, when to test it, and when to slow down. The engineer’s job does not disappear. It changes. More of the work becomes about asking better questions, designing better tests, understanding the architecture, spotting subtle errors, and knowing when a system is reliable enough to use. ## Agents raise the stakes This matters even more as AI moves [from chatbots into agents](https://thenextweb.com/news/google-cloud-next-ai-agents-agentic-era). Chatbots answer questions. Agents can take actions. They can use tools, navigate software, retrieve information, write code, review documents, and complete tasks. That makes them more powerful, but also much riskier. A chatbot giving a bad answer is a problem. An agent taking the wrong action can be a much bigger problem. That is why the future of AI engineering cannot only be about making models more capable. It also has to be about trust. Engineers need to think about evaluation, testing, transparency, oversight, and when a human should step in. They need to understand not only whether an AI system can do something once, but whether it can do it repeatedly across messy real-world situations. It is easy to be impressed by a demo. It is harder to build something that works when the inputs are unclear, the data is incomplete, the user changes their mind, or the environment behaves differently than expected. That is where engineering judgment matters most. ## The demo is not the product One mistake I think people make with AI is confusing demos with systems. A demo is designed to show what is possible. A system has to survive what is probable. Real users do not follow the perfect path. Real workflows contain missing information, edge cases, unclear instructions, old systems, conflicting requirements, and unexpected constraints. The model is only one part of that. The surrounding system determines whether the AI becomes useful: the interface, the evaluation loop, the tool access, the error handling, the escalation path, and the human oversight. This is why strong software engineering still matters. Algorithms, systems, databases, networking, and software design are not suddenly irrelevant. They give engineers the foundation to understand when AI-generated work is correct and when something feels off. The future is not fundamentals versus AI. It is fundamentals plus AI. ## Students should get close to real problems For students trying to break into AI, my advice is practical. Build real things. Use AI tools every day. Break them. Compare models. Try to automate parts of your own workflow. Notice where the tools help and where they fail. You do not need to be a machine learning researcher to contribute to AI. That is one path, but it is not the only one. The industry also needs strong software engineers, infrastructure engineers, product thinkers, designers, security experts, and people who understand how real users behave. The models are powerful, but the systems around them determine whether they become useful products. I also think students should get close to [ambitious environments](https://thenextweb.com/news/asked-tech-execs-what-advice-give-younger-selves) as early as possible. Moving to San Francisco changed my perspective. Being around serious builders made the pace of the industry feel real. I started to understand what people were working on, what problems they cared about, and [what skills were becoming valuable](https://thenextweb.com/news/demand-skills-developers-2025). But proximity alone is not enough. You still have to do the work. ## Take imperfect opportunities seriously My path to OpenAI looked much cleaner from the outside than it felt while I was living it. I was rejected many times. I cold-emailed recruiters. I practiced interviews with friends. I applied early. I tried to put myself in environments where I could learn faster. Before entering frontier AI, I also worked jobs far from the world of advanced technology, [including as a janitor](https://x.com/hamostaf04/status/1957465218793328780?s=20). I later [shared that part of my story publicly](https://www.linkedin.com/posts/hamza-mostafa_i-was-debating-posting-this-for-a-while-activity-7325281932219461632-oinC/) because I think people often see only the final outcome, not the uncertainty that came before it. When people see OpenAI on a resume, they assume the path was straightforward. It was not. The unpaid internship mattered. The rejections mattered. The cold emails mattered. The jobs outside tech mattered. Each step gave me a little more evidence that I could operate in a more demanding environment. OpenAI showed me what the frontier of AI looks like. The path there taught me something just as important: progress often comes from taking the next opportunity seriously before you feel fully ready. The future of AI will not belong only to people who can build impressive demos. It will belong to people who can turn powerful technology into systems that work reliably in the real world. That is the lesson I am carrying forward. ## Get the TNW newsletter Get the most important tech news in your inbox each week. Provided by Hamza Mostafa