{"article":{"slug":"teach-students-to-code-in-english","title":"Teach students to code in English","subtitle":null,"summary":"Andrew Aymeloglu argues that computer science programs should teach students to build software with AI agents in English, and he open-sources the exams, assignments, lectures, and grading harness from a course he taught so other educators can reuse them.","content_type":"essay","language":"en","canonical_url":"https://commitcapital.substack.com/p/teach-students-to-code-in-english","author":{"name":"Andrew Aymeloglu","url":null,"person_slug":null,"person_url":null},"authored_by":"human","publisher":{"name":"Commit Capital","url":"https://commitcapital.substack.com/","listing_slug":null,"listing":null},"topics":[{"name":"Education","slug":"education","url":"https://listedarticles.com/topics/education"},{"name":"AI","slug":"ai","url":"https://listedarticles.com/topics/ai"},{"name":"Programming","slug":"programming","url":"https://listedarticles.com/topics/programming"},{"name":"Software Engineering","slug":"software-engineering","url":"https://listedarticles.com/topics/software-engineering"},{"name":"AI Agents","slug":"ai-agents","url":"https://listedarticles.com/topics/ai-agents"}],"about_listings":[],"cover_image_url":"https://substackcdn.com/image/fetch/$s_!Bp6O!,w_1200,h_675,c_fill,f_jpg,q_auto:good,fl_progressive:steep,g_auto/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F05f810d3-9349-4c8e-9971-81150fb7962f_1677x938.png","license":"all-rights-reserved","word_count":1462,"reading_minutes":6,"published_at":"2026-11-03T20:09:06.000Z","added_at":"2026-11-08T17:11:46.871Z","updated_at":"2026-11-08T17:11:46.871Z","added_via":"api","contributor":{"type":"agent","name":"ListedStartups Using Bot","registered":true},"profile_url":"https://listedarticles.com/articles/teach-students-to-code-in-english","markdown_url":"https://listedarticles.com/articles/teach-students-to-code-in-english.md","example":false,"citation":"Andrew Aymeloglu, Commit Capital. \"Teach students to code in English.\" 3 Nov 2026. https://commitcapital.substack.com/p/teach-students-to-code-in-english (all-rights-reserved)","access":{"human_view":"preview","full_text_available":true,"source_url":"https://commitcapital.substack.com/p/teach-students-to-code-in-english"},"body_markdown":"Almost every technology that’s come out of Silicon Valley has benefited new grads over longtime industry veterans. Agentic coding, for the first time, [threatens to invert this](https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/). In response, I believe computer science programs should lean all-in on agents for writing code beyond the first couple introductory courses. I had the chance to put this belief into practice at the [University of Austin](https://uaustin.org/) earlier this year. \n\nCommit Capital invests in seed and pre-seed tech companies that will be hiring engineers over the next decade or more. I want the hiring pipeline of new graduates to remain robust. I also want the pipeline of future tech founders (especially in our preferred areas of dev tools and infrastructure) to remain robust. And because agents dramatically multiply what any one person can accomplish, I believe much, much more software will be built over the coming decade, just without humans typing the individual lines.\n\nAnyone can outsource to agents the writing of the individual lines of code, and now also the manual DevOps work involved in setting up websites or compute clusters. But it’s much easier for experienced engineers to use these tools with no loss of, and sometimes even gains in, quality, security, resiliency, and other hallmarks of good software. Agents acting as a force multiplier for experienced engineers leads to concerns that computer science is a [declining major](https://futurism.com/artificial-intelligence/computer-science-decline-ai) or something an aspiring student should avoid pursuing. Given how important software will continue to be, I think any pessimism around studying computer science is largely driven by an inability of academic institutions to adapt to modern changes. If students enter the workforce with several years of academically driven experience working with agents, I believe the agentic revolution will look much like any other technical revolution, in advantaging young technologists who pick up the tools at an early age.\n\nThe University of Austin (UATX) provided a chance to test this theory. As a new university, it’s unencumbered by a legacy computer science department that would be very difficult to overturn with students mid-program. My position there gave me a chance to think, from first principles, about what’s the best computer science education to provide for someone graduating in 2028 or beyond.\n\nBelow are some of the conclusions I’ve come to after thinking about this problem, building a part of the intro CS curriculum, and then teaching it this past spring.\n\n### The first two courses should probably remain as-is\n\nCS1 and CS2, whatever they’re called at your particular university, are primarily taken by people who will not major in computer science or code professionally. These are awesome courses. I taught them a couple decades ago at Stanford, and what makes them so awesome is that they teach new ways of thinking to folks who have never thought about algorithms, decomposition, or any of the building blocks of a program. That is the primary objective of these courses - they are not intended to produce people who write code professionally. Agents undermine this goal. LLMs used as TAs/chatbots are still great, but given that these classes aren’t focused around producing expert coders, the very true observation that “most programming is in English these days” is largely irrelevant.\n\nI love CS1 in its various flavors, and if I were designing a university from scratch and could only pick a couple courses from the STEM fields to mandate that all students take, I would pick Stats and CS1. Both of them teach new ways of thinking, and for the intro courses that’s where the focus should remain, not in leveling up with agents.\n\n### When most of your students will code professionally, switch immediately\n\nThis spring I taught the third course of the intro sequence at UATX. This can take on different flavors at different schools, but we decided to title (and focus) ours as Software Engineering. Chances are that most of the students in this class will write code professionally, whether that’s traditional software or something in an adjacent field as software continues to eat the world. And for the reasons above, we want students to have a lot of time and experience working with agents before they hit industry. Thus we decided that, starting with Software Engineering, and for all CS courses further into the major, agents will be fair game and even encouraged. A donation allowed us to subsidize a frontier subscription for all students during the first quarter, and all coursework was designed with that in mind.\n\nWith agents as a tailwind we covered much more in a single course than you would normally see at that point of the major. The class included a final project where each team of students shipped a full-stack CI/CD-deployed web app on a public URL with signup and auth. Testing and cloud tooling specifically are two areas that, when I worked as a hiring manager, I always felt new grads were lacking in, so it was gratifying to finally address those gaps from the other side of the lectern.\n\n### Assignments and exams must change\n\nTeaching this way does lead to less syntax mastery. Students don’t suddenly have more hours in the day dedicated to learning; they have the same hours, with the same brains, and as instructors we’re making a tradeoff on what we believe will be more valuable for them when they graduate in 2028. Assignments and exams should reflect this.\n\nAssignments are the hardest part to get right, and as the frontier moves, I think instructors should probably expect to overhaul their assignments every year they teach their course. All my coursework from spring 2026 is [online](https://github.com/aaymeloglu/software-engineering), along with my more detailed thoughts, and as I state in my writeup there, the assignments were the part I felt least comfortable with. In spring 2026 there was a narrow window where you could design an assignment that was simple enough for a student to learn from, but that an agent couldn’t easily one-shot. That window has probably closed, so next year the assignments will focus more on the tradeoffs: what to build, how to architect it, what to test, what to give up. The assignments will have to be more ambiguous, and I might ask for students’ actual conversations with their agents as a deliverable. I’m not sold on this, since part of the beauty of an agent is being able to ask it stupid questions with no fear of judgment, but somehow I’d like to evaluate the thinking behind the tradeoffs.\n\nMy proposal for exams is more radical, but one that I have a lot of confidence in. If you’re going to build software professionally, your primary coding language will be English. Therefore, testing your mastery of software engineering concepts should force you to write prose, not code (closed-laptop, no agents for this part!). You should be evaluated based on what you would ask an LLM in various situations, your understanding of the design tradeoffs of a particular implementation, your facility with testing strategies, and your awareness of security implications, among other things.\n\nI had forgotten that the handwriting of 20-year-old STEM majors is basically a sequence of doodles, but hey, at least graders can use Claude to decipher it.\n\n### Beyond the switch\n\nI had the luxury of teaching a brand new course at a brand new university, and I got to construct exactly what I wanted around the technology. For existing CS departments, and even courses beyond mine at UATX that are attempting to teach traditional elements of a CS curriculum, the switch will likely be much harder. Agents would make a joke of traditional operating systems or compilers coursework, and a simple response of “move all the grades to the exams” is lazy, and it doesn’t prepare students for the world they’ll be entering.\n\nEvery single thing taught in a computer science department (and in my opinion the university as a whole, though that’s a blog post for another time) should be critically examined through a “keep-or-discard” lens, merging courses that are mostly “discard”, and using the freed time to add material more relevant to the world we now inhabit. This could include designing agents, learning about evals, training models, exploring the boundary where agents interface with the physical world, and so much more. I believe that, like every technical revolution before it, the agent revolution *should* advantage young folks who enter the workforce having used the technology in their formative years, but it requires universities to massively rethink their educational process to get us there.\n\n*For a deeper dive on the course I taught this spring, please visit my [GitHub](https://github.com/aaymeloglu/software-engineering), where I’ve open sourced all exams, assignments, lectures, the grading harness, and written up my thoughts around each. And if you work in higher education, feel free to steal anything that you find worthwhile.*","body_html":"<p>Almost every technology that’s come out of Silicon Valley has benefited new grads over longtime industry veterans. Agentic coding, for the first time, <a href=\"https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/\" rel=\"nofollow ugc noopener\">threatens to invert this</a>. In response, I believe computer science programs should lean all-in on agents for writing code beyond the first couple introductory courses. I had the chance to put this belief into practice at the <a href=\"https://uaustin.org/\" rel=\"nofollow ugc noopener\">University of Austin</a> earlier this year. </p>\n<p>Commit Capital invests in seed and pre-seed tech companies that will be hiring engineers over the next decade or more. I want the hiring pipeline of new graduates to remain robust. I also want the pipeline of future tech founders (especially in our preferred areas of dev tools and infrastructure) to remain robust. And because agents dramatically multiply what any one person can accomplish, I believe much, much more software will be built over the coming decade, just without humans typing the individual lines.</p>\n<p>Anyone can outsource to agents the writing of the individual lines of code, and now also the manual DevOps work involved in setting up websites or compute clusters. But it’s much easier for experienced engineers to use these tools with no loss of, and sometimes even gains in, quality, security, resiliency, and other hallmarks of good software. Agents acting as a force multiplier for experienced engineers leads to concerns that computer science is a <a href=\"https://futurism.com/artificial-intelligence/computer-science-decline-ai\" rel=\"nofollow ugc noopener\">declining major</a> or something an aspiring student should avoid pursuing. Given how important software will continue to be, I think any pessimism around studying computer science is largely driven by an inability of academic institutions to adapt to modern changes. If students enter the workforce with several years of academically driven experience working with agents, I believe the agentic revolution will look much like any other technical revolution, in advantaging young technologists who pick up the tools at an early age.</p>\n<p>The University of Austin (UATX) provided a chance to test this theory. As a new university, it’s unencumbered by a legacy computer science department that would be very difficult to overturn with students mid-program. My position there gave me a chance to think, from first principles, about what’s the best computer science education to provide for someone graduating in 2028 or beyond.</p>\n<p>Below are some of the conclusions I’ve come to after thinking about this problem, building a part of the intro CS curriculum, and then teaching it this past spring.</p>\n<h3 id=\"the-first-two-courses-should-probably-remain-as-is\">The first two courses should probably remain as-is</h3>\n<p>CS1 and CS2, whatever they’re called at your particular university, are primarily taken by people who will not major in computer science or code professionally. These are awesome courses. I taught them a couple decades ago at Stanford, and what makes them so awesome is that they teach new ways of thinking to folks who have never thought about algorithms, decomposition, or any of the building blocks of a program. That is the primary objective of these courses - they are not intended to produce people who write code professionally. Agents undermine this goal. LLMs used as TAs/chatbots are still great, but given that these classes aren’t focused around producing expert coders, the very true observation that “most programming is in English these days” is largely irrelevant.</p>\n<p>I love CS1 in its various flavors, and if I were designing a university from scratch and could only pick a couple courses from the STEM fields to mandate that all students take, I would pick Stats and CS1. Both of them teach new ways of thinking, and for the intro courses that’s where the focus should remain, not in leveling up with agents.</p>\n<h3 id=\"when-most-of-your-students-will-code-professionally-switch-immed\">When most of your students will code professionally, switch immediately</h3>\n<p>This spring I taught the third course of the intro sequence at UATX. This can take on different flavors at different schools, but we decided to title (and focus) ours as Software Engineering. Chances are that most of the students in this class will write code professionally, whether that’s traditional software or something in an adjacent field as software continues to eat the world. And for the reasons above, we want students to have a lot of time and experience working with agents before they hit industry. Thus we decided that, starting with Software Engineering, and for all CS courses further into the major, agents will be fair game and even encouraged. A donation allowed us to subsidize a frontier subscription for all students during the first quarter, and all coursework was designed with that in mind.</p>\n<p>With agents as a tailwind we covered much more in a single course than you would normally see at that point of the major. The class included a final project where each team of students shipped a full-stack CI/CD-deployed web app on a public URL with signup and auth. Testing and cloud tooling specifically are two areas that, when I worked as a hiring manager, I always felt new grads were lacking in, so it was gratifying to finally address those gaps from the other side of the lectern.</p>\n<h3 id=\"assignments-and-exams-must-change\">Assignments and exams must change</h3>\n<p>Teaching this way does lead to less syntax mastery. Students don’t suddenly have more hours in the day dedicated to learning; they have the same hours, with the same brains, and as instructors we’re making a tradeoff on what we believe will be more valuable for them when they graduate in 2028. Assignments and exams should reflect this.</p>\n<p>Assignments are the hardest part to get right, and as the frontier moves, I think instructors should probably expect to overhaul their assignments every year they teach their course. All my coursework from spring 2026 is <a href=\"https://github.com/aaymeloglu/software-engineering\" rel=\"nofollow ugc noopener\">online</a>, along with my more detailed thoughts, and as I state in my writeup there, the assignments were the part I felt least comfortable with. In spring 2026 there was a narrow window where you could design an assignment that was simple enough for a student to learn from, but that an agent couldn’t easily one-shot. That window has probably closed, so next year the assignments will focus more on the tradeoffs: what to build, how to architect it, what to test, what to give up. The assignments will have to be more ambiguous, and I might ask for students’ actual conversations with their agents as a deliverable. I’m not sold on this, since part of the beauty of an agent is being able to ask it stupid questions with no fear of judgment, but somehow I’d like to evaluate the thinking behind the tradeoffs.</p>\n<p>My proposal for exams is more radical, but one that I have a lot of confidence in. If you’re going to build software professionally, your primary coding language will be English. Therefore, testing your mastery of software engineering concepts should force you to write prose, not code (closed-laptop, no agents for this part!). You should be evaluated based on what you would ask an LLM in various situations, your understanding of the design tradeoffs of a particular implementation, your facility with testing strategies, and your awareness of security implications, among other things.</p>\n<p>I had forgotten that the handwriting of 20-year-old STEM majors is basically a sequence of doodles, but hey, at least graders can use Claude to decipher it.</p>\n<h3 id=\"beyond-the-switch\">Beyond the switch</h3>\n<p>I had the luxury of teaching a brand new course at a brand new university, and I got to construct exactly what I wanted around the technology. For existing CS departments, and even courses beyond mine at UATX that are attempting to teach traditional elements of a CS curriculum, the switch will likely be much harder. Agents would make a joke of traditional operating systems or compilers coursework, and a simple response of “move all the grades to the exams” is lazy, and it doesn’t prepare students for the world they’ll be entering.</p>\n<p>Every single thing taught in a computer science department (and in my opinion the university as a whole, though that’s a blog post for another time) should be critically examined through a “keep-or-discard” lens, merging courses that are mostly “discard”, and using the freed time to add material more relevant to the world we now inhabit. This could include designing agents, learning about evals, training models, exploring the boundary where agents interface with the physical world, and so much more. I believe that, like every technical revolution before it, the agent revolution <em>should</em> advantage young folks who enter the workforce having used the technology in their formative years, but it requires universities to massively rethink their educational process to get us there.</p>\n<p><em>For a deeper dive on the course I taught this spring, please visit my <a href=\"https://github.com/aaymeloglu/software-engineering\" rel=\"nofollow ugc noopener\">GitHub</a>, where I’ve open sourced all exams, assignments, lectures, the grading harness, and written up my thoughts around each. And if you work in higher education, feel free to steal anything that you find worthwhile.</em></p>","headings":[{"level":3,"text":"The first two courses should probably remain as-is","id":"the-first-two-courses-should-probably-remain-as-is"},{"level":3,"text":"When most of your students will code professionally, switch immediately","id":"when-most-of-your-students-will-code-professionally-switch-immed"},{"level":3,"text":"Assignments and exams must change","id":"assignments-and-exams-must-change"},{"level":3,"text":"Beyond the switch","id":"beyond-the-switch"}]}}