Briefs
Course by course
CSC455 Software Engineering
Catalogue today“agile software development methods such as Extreme Programming (XP), Agile Modeling (AM), Scrum, ASD, DSDM, Crystal, Feature Driven Development (FDD), Incremental Funding Method (IFM), DevOps, and Site Reliability Engineering”
The description is a catalogue of process methods. It does not mention how software is now written with AI coding agents, or how AI-written code is specified, tested, reviewed and secured.
Keep- Software life cycle: specification, design, implementation, testing, maintenance
- Agile and Scrum, DevOps, Site Reliability Engineering
Compress- The list of older process methods (XP, AM, ASD, DSDM, Crystal, FDD, IFM, RUP, Open Process Framework, CMM, PSP) into one session on how software process evolved. None appears in the 7 leading catalogues compared.
| Unit | Learning outcome | Assignment | Evidence |
|---|
| Building software with AI coding agents | Students plan a feature, direct an AI coding agent to build it, and explain every change it made. | A feature built with an agent, submitted with the agent's session log and the student's review notes. | CMU 17-316, MIT 6.1040, UPenn CIS 2500; Anthropic and OpenAI academy courses; agents named in 35.6% of frontier-lab technical postings. |
| Specifications an agent can execute | Students write a specification precise enough that an agent's output can be checked against it. | Specification graded before any code is written. | DeepLearning.AI: Spec-Driven Development with Coding Agents. |
| Testing and evaluation as the core skill | Students write the tests that decide whether AI-written code is accepted. | Test suite graded on the defects it catches in a seeded, AI-generated codebase. | Best model fixes 0.83 of real bugs (SWE-bench Verified); evaluations named in 13.6% of frontier-lab technical postings; DeepLearning.AI: Evaluating and Improving Agentic Applications. |
| Reviewing AI-written code, including for security | Students find and justify defects and security flaws in code they did not write. | Structured code review of a pull request produced by an agent. | Code review in leading catalogues went from 5 to 7 course entries; DeepLearning.AI: AI Code Review. |
| Production basics | Students ship to a container with automated tests running on every change. | Team project deployed with continuous integration. | Kubernetes 14.8% and distributed systems 25.0% of frontier-lab technical postings; cloud and DevOps 38% at leading firms; LAU CSC244 teaches Docker. |
AssessmentBecause AI now writes most routine code, grade the specification, tests, review notes and an oral walkthrough, not only the final code.
Full evidence and coverage grid: first two briefs.
CSC441 Cybersecurity Fundamentals
Catalogue today“Students will learn about different types of cyber-attacks, security policies and standards, risk assessment, and security controls.”
The description covers classic fundamentals. It does not mention AI as an attacker, AI as a defender, or the security of AI systems themselves.
Keep- Security concepts, controls, attack types
- Policy, standards and risk assessment
- Legal and ethical issues
| Unit | Learning outcome | Assignment | Evidence |
|---|
| AI as attacker | Students explain what AI tools can now find and exploit, and what that changes for defenders. | In a sandbox, run an AI-assisted vulnerability scan, then verify and triage each finding by hand. | ExploitBench 0.14 to 0.74 in a year; Columbia AI for Software Security; Berkeley: AI for vulnerability detection. |
| Securing AI systems | Students identify the main attacks on AI systems and the controls against them. | Threat model of a chatbot or agent that uses tools and company data. | Columbia Machine Learning Security; Illinois CS 562; CMU 17-331. |
| AI for detection and response | Students run an incident from detection to post-incident review, using AI to speed triage and judging its output. | Timed incident simulation with a written post-incident review. | CMU 17-739 (machine learning as weapon and shield); detection and response roles at frontier labs. |
| Policy for AI security | Students connect security policy and risk assessment to AI use inside an organization. | Short AI-use security policy for a real Lebanese organization type. | CMU 17-331 Information Security, Privacy and Policy in the Age of AI. |
AssessmentGrade the triage judgment and written reasoning, not only tool output; AI tools now produce much of the raw output.
Full evidence and coverage grid: first two briefs.
CSC210 Introduction to Programming
Catalogue today“Students will learn how to write simple programs using a high-level programming language such as Java.”
Closest peer courses say- Washington CSE 121: “Students write programs to express algorithmic thinking and solve computational problems motivated by modern societal and scientific needs”
This week's module picks- AI Python for Beginners DeepLearning.AI, Beginner, 11h30m, updated 12 days ago
Optional lab track: the same constructs taught in Python with an AI tutor.
Keep- Core constructs, control flow, functions, debugging
| Unit | Learning outcome | Assignment | Evidence |
|---|
| Problems drawn from real data and real needs | Students solve small computational problems from science and society, not only textbook drills. | Weekly problems built on real data sets. | Washington CSE 121 frames the first course this way. |
| Learning to program with an AI assistant, under rules | Students use an AI assistant to explain errors and suggest fixes, and can explain every line they submit. | Each lab submitted with a short log of what the assistant suggested and what the student changed. | AI now writes standard exercises; first-year students will use these tools whatever the policy. |
| Python alongside Java | Students can write the same small program in both. | Two labs repeated in Python. | Python is named in half of frontier-lab technical postings. |
AssessmentMove weight to in-class coding without AI tools and short oral explanations of submitted code; take-home exercises alone no longer show what a student can do.
- The best model fixes 0.83 of real software bugs (SWE-bench Verified) and completes tasks of 17.4 hours on its own at a 50% success rate (METR). Standard programming exercises can now be done by AI.
- Python is named in 50.0% of 1,001 frontier-lab technical postings; Rust 17.8%; Kubernetes 14.8%; distributed systems 25.0%; agents and tool use 35.6%.
CSC211 Discrete Methods
Catalogue today“set theory, relations, and functions, logic and proofs, combinatorics and discrete probability, graph theory, number theory, the growth of functions, and complexity of algorithms”
Closest peer courses say- Georgia Tech CS 2050: “Proof methods, strategy, correctness of algorithms over discrete structures. Induction and recursion.”
This week's module picks- Mathematics for Machine Learning and Data Science DeepLearning.AI, Beginner, 94h29m, updated 44 days ago
Supplement for the probability unit, for students heading to data and AI courses.
Keep- The full topic list; it matches leading first-year courses
| Unit | Learning outcome | Assignment | Evidence |
|---|
| Correctness of algorithms | Students prove a simple algorithm correct by induction. | One correctness proof per algorithm studied in CSC315. | Georgia Tech frames discrete math around proof strategy and algorithm correctness. |
| Checking machine-written proofs | Students find the flaw in a plausible but wrong proof produced by an AI model. | Critique of three AI-generated proofs, one of them flawed. | Models now answer expert-level science questions at 0.96 (GPQA Diamond); the skill that remains is judging whether an argument holds. |
AssessmentAdd whiteboard or oral proofs; written take-home proofs can now be produced by AI.
- The best model fixes 0.83 of real software bugs (SWE-bench Verified) and completes tasks of 17.4 hours on its own at a 50% success rate (METR). Standard programming exercises can now be done by AI.
CSC212 Computer Organization and Assembly Language
Catalogue today“The course focuses on the Intel x86 Assembly language, number representations, logic circuits, Boolean algebra and logic simplification”
Closest peer courses say- Washington CSE 469: “Memory systems and caching. Performance modeling of microprocessors.”
- Stanford CS 180: “introduction to operating systems and compilers”
This week's module picks- No current MOOC or lab module in the index fits this course. Use the Washington CSE 469 and Stanford CS 180 syllabi.
Keep- Number representations, Boolean logic, assembly, memory and I/O
| Unit | Learning outcome | Assignment | Evidence |
|---|
| Caches and performance | Students predict and measure how memory access patterns change run time. | Measure a loop under different access patterns and explain the result. | Washington CSE 469 covers caching and performance modeling. |
| Accelerators and GPUs, at survey level | Students explain why AI workloads run on GPUs and what that means for memory and parallelism. | Short comparison of a CPU and a GPU on one workload. | GPUs and high-performance computing appear in 10% of leading-firm technical postings; performance optimization in 10.9% of frontier-lab postings. |
AssessmentKeep hands-on assembly and measurement labs; they test understanding that AI cannot supply for the student.
- Of 1,356 technical postings at four leading AI firms: cloud and DevOps 38%, data engineering 31%, human-computer interaction and UX 15%, GPUs and high-performance computing 10%.
CSC314 Object Oriented Programming
Catalogue today“The course provides an in-depth understanding of object-oriented programming and design principles using UML”
Closest peer courses say- Georgia Tech CS 1331: “Emphasis on software development and individual programming skills.”
- MIT 6.1040: “taught to make effective and intentional use of LLMs in design and coding”
This week's module picks- Claude Code: A Highly Agentic Coding Assistant DeepLearning.AI, Intermediate, 2h, updated 87 days ago
Lab on refactoring and extending an object-oriented code base with an agent.
Keep- Encapsulation, inheritance, polymorphism, design principles
| Unit | Learning outcome | Assignment | Evidence |
|---|
| Designing and refactoring with an AI coding agent | Students direct an agent to refactor a class design and judge whether the result is better. | Refactor an existing project with an agent; submit before and after designs with reasons. | MIT 6.1040 teaches intentional use of LLMs in design and coding. |
AssessmentGrade the design decisions and their justification; generated code is not evidence of design skill.
- The best model fixes 0.83 of real software bugs (SWE-bench Verified) and completes tasks of 17.4 hours on its own at a 50% success rate (METR). Standard programming exercises can now be done by AI.
CSC315 Data Structures and Algorithms
Catalogue today“The course emphasizes the use of object-oriented programming techniques to implement and manipulate data structures.”
Closest peer courses say- Washington CSE 373: “Linked lists, stacks, queues, directed graphs”
This week's module picks- No current MOOC or lab module in the index fits. Use the Washington CSE 373 and Stanford CS 106B syllabi.
Keep- The core structures and analysis of algorithms
| Unit | Learning outcome | Assignment | Evidence |
|---|
| Choosing structures for real workloads | Students pick and justify a structure from measured behavior, not from memory. | Benchmark two structures on the same task and defend the choice. | AI can produce standard implementations; the judgment of which to use and why stays with the engineer. |
AssessmentLive problem-solving with spoken reasoning about complexity; take-home implementations alone can be produced by AI.
- The best model fixes 0.83 of real software bugs (SWE-bench Verified) and completes tasks of 17.4 hours on its own at a 50% success rate (METR). Standard programming exercises can now be done by AI.
CSC320 Database Management Systems
Catalogue today“Entity-relationship (ER) modeling and database design, SQL programming and query optimization”
Closest peer courses say- UT Austin C S X86D: “distributed database management systems”
This week's module picks- Vector Databases: from Embeddings to Applications DeepLearning.AI, Intermediate, 1h5m, updated 31 days ago
A two-week unit on vector search, the storage layer behind AI search and agents.
Keep- Relational model, ER design, SQL, normalization, indexing, transactions
| Unit | Learning outcome | Assignment | Evidence |
|---|
| Vector databases and retrieval | Students store embeddings, run similarity search, and compare it with SQL search on the same data. | Small search feature built two ways. | Retrieval and search are named in 9.9% of frontier-lab technical postings; data engineering in 31% of leading-firm postings. |
| Distributed databases | Students explain replication and partitioning trade-offs. | Design note for scaling a schema across regions. | UT Austin's database systems course covers distributed database management systems. |
AssessmentKeep design projects; add a written defense of schema and index choices.
- Of 1,356 technical postings at four leading AI firms: cloud and DevOps 38%, data engineering 31%, human-computer interaction and UX 15%, GPUs and high-performance computing 10%.
CSC331 Web Programming
Catalogue today“Students will learn how to create dynamic and interactive web pages using HTML, CSS, JavaScript, XML, and DTD.”
Closest peer courses say- Illinois CS 409: “data exchange (HTTP, AJAX, JSON, REST)”
- Washington CSE 154: “web security”
This week's module picks- Claude Code: A Highly Agentic Coding Assistant DeepLearning.AI, Intermediate, 2h, updated 87 days ago
Build and test a small web app with an agent. - Building AI Browser Agents DeepLearning.AI, Intermediate, 1h5m, updated 118 days ago
Optional advanced lab: agents that use web pages.
Keep- HTML, CSS, JavaScript, server-side basics
Compress- XML and DTD to one session; peers teach data exchange through JSON and REST.
| Unit | Learning outcome | Assignment | Evidence |
|---|
| APIs: JSON and REST | Students build and call a REST API that exchanges JSON. | Front end that talks to a student-built API. | Illinois CS 409 names HTTP, JSON and REST; APIs are named in 34.5% of frontier-lab technical postings. |
| Web security basics | Students find and fix common web vulnerabilities. | Security review of their own app. | Washington CSE 154 includes web security. |
| A modern front-end framework | Students build an interactive page with a component framework. | Rebuild one lab with a framework. | TypeScript or JavaScript is named in 16.7% and React or front end in 10.4% of frontier-lab technical postings. |
AssessmentGrade the working app, its API design and the security review.
- Python is named in 50.0% of 1,001 frontier-lab technical postings; Rust 17.8%; Kubernetes 14.8%; distributed systems 25.0%; agents and tool use 35.6%.
CSC352 Theory of Programming Languages
Catalogue today“programmers need to understand the different styles of programming promoted by different languages”
Closest peer courses say- UT Austin C S X86L: “formal syntax representations, program correctness, typing, and data abstraction”
This week's module picks- No current MOOC or lab module in the index fits. Use the UT Austin C S X86L and Carnegie Mellon 17-396 syllabi.
Keep- Paradigms and language styles
| Unit | Learning outcome | Assignment | Evidence |
|---|
| Types and correctness | Students explain how a type system prevents classes of bugs. | Port a small buggy program to a stricter language and report what the compiler caught. | UT Austin's course centers program correctness and typing. |
| Rust as a case study in safety | Students explain ownership and memory safety. | Small Rust program with a written comparison to C. | Rust is named in 17.8% of frontier-lab technical postings; USEK's catalogue does not mention it. |
AssessmentWritten analysis of language features, with examples the student runs and explains.
- Python is named in 50.0% of 1,001 frontier-lab technical postings; Rust 17.8%; Kubernetes 14.8%; distributed systems 25.0%; agents and tool use 35.6%.
CSC416 Graph Theory and Operations Research
Catalogue today“shortest path, minimum spanning tree, maximal flow network”
Closest peer courses say- Georgia Tech CS 7510: “Algorithms for graph problems such as maximum flow, matching, network reliability, minimum cuts”
This week's module picks- No current MOOC or lab module in the index fits. Use the Georgia Tech CS 7510 and MIT 6.7210 syllabi.
Keep- The graph and linear-methods core; it matches peers
| Unit | Learning outcome | Assignment | Evidence |
|---|
| Solving real models with a solver | Students formulate a scheduling or routing problem and solve it with an optimization library. | Model a real Lebanese logistics or scheduling case. | MIT's optimization course pairs methodology with the underlying mathematics; practice with solvers is the working skill. |
AssessmentGrade the model formulation and the interpretation of results.
CSC417 Human Computer Interaction
Catalogue today“To facilitate communication between human factors engineers and computer scientists on user interface development projects.”
Closest peer courses say- Washington CSE 440: “Methods for designing, prototyping, and evaluating user interfaces”
- Stanford CS 147D: “conversational and agentic interfaces”
This week's module picks- No lab or MOOC module in the index teaches interface design for AI. Use the Stanford CS 147D syllabus.
Keep- Human factors, design methods, prototyping
| Unit | Learning outcome | Assignment | Evidence |
|---|
| Designing conversational and agent interfaces | Students design and test an interface where the user works with an AI assistant or agent. | Prototype and user test of an AI feature. | Stanford CS 147D teaches conversational and agentic interfaces; HCI and UX appear in 15% of leading-firm technical postings. |
| Evaluation with real users | Students run and report a usability study. | Usability test with at least five users. | Washington CSE 440 centers designing, prototyping and evaluating. |
AssessmentGrade the user study and the design changes it led to.
- Of 1,356 technical postings at four leading AI firms: cloud and DevOps 38%, data engineering 31%, human-computer interaction and UX 15%, GPUs and high-performance computing 10%.
CSC420 Computer Networks
Catalogue today“It presents the networking protocol models: OSI and TCP/IP by discussing the different layers, their functions, roles, and services.”
Closest peer courses say- Illinois CS 438: “design issues and protocols in the transport, network, and data link layers”
This week's module picks- Introduction to Model Context Protocol Anthropic Academy
Optional application-layer case: the protocol AI tools use to talk to services.
Keep- Layered models and protocols
| Unit | Learning outcome | Assignment | Evidence |
|---|
| Protocol design issues, not only layer descriptions | Students explain the design choices behind transport and routing protocols. | Packet capture analysis of a real exchange. | Illinois CS 438 teaches design issues in each layer. |
| Networks for cloud and distributed systems | Students explain how data-center and cloud networks shape application design. | Short design note for a service deployed across regions. | Distributed systems are named in 25.0% of frontier-lab postings; cloud and DevOps in 38% of leading-firm postings. |
AssessmentHands-on labs with packet captures and configuration.
- Python is named in 50.0% of 1,001 frontier-lab technical postings; Rust 17.8%; Kubernetes 14.8%; distributed systems 25.0%; agents and tool use 35.6%.
- Of 1,356 technical postings at four leading AI firms: cloud and DevOps 38%, data engineering 31%, human-computer interaction and UX 15%, GPUs and high-performance computing 10%.
CSC421 Operating Systems
Catalogue today“fundamental concepts, principles, and algorithms underlying modern operating systems”
Closest peer courses say- Stanford CS 111: “concurrency, synchronization, scheduling, processes, virtual memory, I/O, file systems, and protection”
- Georgia Tech CS 3210: “Projects will cover design and implementation of several operating systems compo”
This week's module picks- No current MOOC or lab module in the index fits. Use the Stanford CS 111 and Georgia Tech CS 3210 syllabi.
Keep- Processes, scheduling, memory, file systems
| Unit | Learning outcome | Assignment | Evidence |
|---|
| Build operating system components | Students implement parts of a teaching kernel. | Scheduler and memory-management projects. | Georgia Tech CS 3210 is built around implementation projects. |
| Containers and virtualization | Students explain how containers isolate processes and run one under limits. | Package a service in a container and set resource limits. | Kubernetes is named in 14.8% of frontier-lab technical postings; cloud and DevOps in 38% of leading-firm postings. |
AssessmentImplementation projects graded on tests and a design walkthrough.
- Python is named in 50.0% of 1,001 frontier-lab technical postings; Rust 17.8%; Kubernetes 14.8%; distributed systems 25.0%; agents and tool use 35.6%.
- Of 1,356 technical postings at four leading AI firms: cloud and DevOps 38%, data engineering 31%, human-computer interaction and UX 15%, GPUs and high-performance computing 10%.
CSC461 Parallel and Distributed Computing
Catalogue today“the design, implementation, and evaluation of parallel and distributed algorithms and systems”
Closest peer courses say- Illinois CS 484: “shared memory and distributed memory multiprocessors, SIMD processors and co-processors”
This week's module picks- No GPU-programming module is in the index yet (NVIDIA's course catalogue was not readable). Use the Illinois CS 484 syllabus.
Keep- Parallel algorithms, distributed systems, evaluation
| Unit | Learning outcome | Assignment | Evidence |
|---|
| GPU programming | Students write and profile a simple GPU kernel. | Speed up one computation on a GPU and explain the result. | Illinois CS 484 covers SIMD processors and co-processors; GPU kernels (CUDA, Triton) are named in 2.7% and performance optimization in 10.9% of frontier-lab postings. |
| How large AI models are trained across many machines | Students explain data and model parallelism. | Short case study of a distributed training setup. | Large-scale training is named in 4.6% and distributed systems in 25.0% of frontier-lab postings. |
AssessmentMeasured speed-ups with written analysis.
- Python is named in 50.0% of 1,001 frontier-lab technical postings; Rust 17.8%; Kubernetes 14.8%; distributed systems 25.0%; agents and tool use 35.6%.
- Of 1,356 technical postings at four leading AI firms: cloud and DevOps 38%, data engineering 31%, human-computer interaction and UX 15%, GPUs and high-performance computing 10%.
CSC470 Machine Learning
Catalogue today“supervised learning, unsupervised learning, and reinforcement learning”
Closest peer courses say- MIT 6.3900: “with a focus on gradient-based methods and neural-network architectures”
- Washington CSE 546: “Supervised learning and predictive modeling; decision trees, rule induction, nearest neighbors”
This week's module picks- Machine Learning Specialization DeepLearning.AI, Beginner, 94h58m, updated 71 days ago
Self-paced practice for weaker students. - Production ML systems Google ML Crash Course
One-week unit on running models in production.
Keep- Learning types, model building, evaluation
| Unit | Learning outcome | Assignment | Evidence |
|---|
| Gradient-based training and neural networks | Students train a small neural network and explain each step. | Implement gradient descent from scratch, then with a library. | MIT 6.3900 now frames introductory machine learning around gradient-based methods and neural networks. |
| Machine learning in production | Students monitor a deployed model for drift. | Deploy a model and report a week of monitoring. | Google's crash course now includes production ML systems. |
AssessmentGrade the experiment design and error analysis, not only accuracy.
- Python is named in 50.0% of 1,001 frontier-lab technical postings; Rust 17.8%; Kubernetes 14.8%; distributed systems 25.0%; agents and tool use 35.6%.
CSC474 Artificial Intelligence
Catalogue today“search algorithms, game-playing algorithms, knowledge representation, reasoning, and planning”
Closest peer courses say- UT Austin C S X43: “starting from foundations in search and knowledge representation, and then sampling from major AI topics such as probabilistic reasoning”
This week's module picks- Agentic AI DeepLearning.AI, Intermediate, 9h55m, updated 13 days ago
Unit on agents that plan and use tools. - Build Reasoning Models Hugging Face LLM Course
Advanced option on reasoning models.
Keep- Search, game playing, knowledge representation, planning
| Unit | Learning outcome | Assignment | Evidence |
|---|
| Probabilistic reasoning | Students build and query a small Bayesian network. | Diagnosis model with uncertain evidence. | UT Austin's AI course moves from search to probabilistic reasoning. |
| Agents that plan and use tools | Students connect classical planning to modern AI agents and build one. | Small agent that plans steps and calls tools. | Agents and tool use are named in 35.6% of frontier-lab technical postings. |
AssessmentGrade the agent's design, its failure cases and the student's analysis.
- Python is named in 50.0% of 1,001 frontier-lab technical postings; Rust 17.8%; Kubernetes 14.8%; distributed systems 25.0%; agents and tool use 35.6%.