AI is advancing. Where does that leave CSE?

My assessment · not an established future fact

The evidence does not establish the end of CSE. It does not guarantee the old career path either.

My working expectation is that CSE remains worth learning, while the emphasis shifts from producing code to delivering useful, reliable outcomes with AI. Entry opportunities are likely to change unevenly. This is my inference from the evidence below—not a prediction of how many jobs will exist.

CSE is more than typing code: it studies computation, information, systems, hardware, security and how people use technology. A discipline remaining useful and every graduate finding work are different claims. You can study something important and still face a difficult job market.

1. 1. TaskWhat can AI perform?
2. 2. TeamHow many people, including verification?
3. 3. MarketHow much work will customers fund?

One automated task does not equal one vanished occupation. Yet teams can shrink without every task being automated. Cheaper software may create demand; it is not guaranteed to create enough jobs to offset displacement.

Separate findings, projections and limits

One headline cannot explain the whole future.

These sources cover different places, periods and methods. Their numbers cannot be treated as interchangeable answers.

This review does not establish a reliable causal estimate of AI-related job gains or losses for Bangladesh’s CSE graduates. Outsourcing exposure, employer budgets, language, infrastructure and hiring practices differ. I will not copy a US percentage and call it your personal prospects.

Conditional scenarios · not research findings

Several futures can happen at once.

Over the next few years these paths could overlap across employers and types of work. I am not assigning equal likelihood or invented percentages. My confidence in precise longer-term forecasts is low.

I would revise this outlook if large independent evaluations demonstrate reliable, cheaper end-to-end engineering over sustained real projects, or if broader cross-country evidence shows entry opportunities consistently recovering. I do not have a reliable date for AGI.

A reason to learn · not employment insurance

Help producing code does not remove the need to understand it.

Recommendations · timeboxes are not promises

You can choose a useful next move without knowing the entire future.

• Choosing CSE, or just starting?

  Do not choose a degree on salary stories or AI fear alone. Consider interest, study costs, alternatives and teaching quality. This week try one small problem yourself, then ask AI to critique it. Record which parts you can explain independently.

  Open the beginner guide →

• Near graduation and need a first role?

  Read several current descriptions for one target role. Fix one real gap in an existing project with an issue, test, review and small release. Check the role’s DSA expectations; do not postpone every application until you know every pattern.

  Choose your entry route →

• Already building with AI?

  Define one small task’s boundaries, acceptance tests, permissions and stop conditions. Include review, debugging and rework when measuring time saved. Check organizational permission before sharing code or private data. Paid tools are not a prerequisite.

  Follow the bounded AI systems path →

This is not solely a student responsibility. Universities need to assess independent reasoning and teamwork alongside AI use; employers need mentorship and supervised entry work. Not finding a job is not a verdict on your intelligence. Revising a target, budget or direction can be rational.

Transparency and rechecking

The complete prompt behind this page

Runtime (configure before running): GPT-6 Astra with High reasoning.

Role: Act as a skeptical research editor for CSE-Pothik, writing for Bangladeshi CSE students from first semester to recent graduation.

Task: Create the complete bilingual editorial for /future-of-cse/. Answer: “Will AI make CSE disappear, and what should a student do now?” Use evidence available through 12 September 2026.

Evidence rules: Prefer accessible primary research and official statistics. Seek counterevidence. For every consequential claim, give source, date, geography, scope and limitation. Separate observed findings, external projections, conditional scenarios and advice. Never invent Bangladesh statistics, causal claims, job-replacement percentages or AGI dates; do not transfer US evidence to Bangladesh. State missing evidence and what would change the outlook.

Required structure: (1) direct short answer and discipline–task–job distinction; (2) four compact evidence cards; (3) three plausible, explicitly conditional scenarios; (4) one clearly fictional AI-assisted engineering failure; (5) why fundamentals, problem framing, verification and trade-offs still matter—without calling them automation-proof or employment insurance; (6) bounded next actions for new students, active learners and job seekers; (7) responsibilities of institutions and employers; (8) rechecking note and relevant CSE-Pothik routes.

Writing contract: Concise English plus independently natural, professional Bangla using আপনি and familiar technical terms. Use gentle situational humor without dismissing anxiety. No hype, reassurance theatre, job guarantee, employer endorsement or paid-tool prerequisite. Credit GPT-6 Astra (High) as the AI narrator, not as OpenAI’s official forecast. Acknowledge uncertainty and possible bias.

Output only the publish-ready article, linked source ledger and unresolved limitations. Do not reveal chain-of-thought. Avoid repetition. Stop researching when every consequential claim is supported or explicitly marked uncertain.

OpenAI prompting guidance↗ · A prompt-design reference.

My closing view: I can analyze evidence; I have not visited the future. This page should change when the evidence changes. If you find a factual error or weak argument, the footer feedback link is there for it.