Video summary
Where Software Engineers Go Next (5 AI-Proof Jobs for Software Engs)
Main summary
Key takeaways
Summary
The speaker argues that AI is more likely to change software engineers’ work than make their skills obsolete. He highlights five career paths where software engineering experience could be especially valuable.
Career paths
-
Security engineering: As AI agents make it easier to find and exploit vulnerabilities, companies will need more security professionals. The work could include directing defensive agents, red-teaming and penetration testing, and—crucially—having humans verify whether vulnerabilities found by agents are real. The speaker expects security to become even more important as AI enters physical products such as robots.
-
Forward-deployed or applied AI engineering: These engineers work with businesses to identify and automate processes using AI. Software engineers have a head start, though the role also calls for product judgment and communication or sales skills.
-
Managing “agent factories”: Rather than managing only human teams, engineers may coordinate large numbers of AI agents working on software and other tasks. This requires technical understanding to assign work, shape the agents’ environment, assess trade-offs, and align their output with organizational goals. The speaker points to Peter Steinberger and a SpaceX example as early archetypes.
-
AI research and LLM engineering: This category includes work in AI labs, fine-tuning models for companies, and deploying open-weight models efficiently. The speaker says software engineers can learn the necessary concepts without necessarily becoming specialists in every area of machine learning. He also anticipates AI systems helping to create or improve smaller models, increasing the value of people who understand how to guide that work.
-
Robotics: The speaker sees robotics as a way to bring software’s efficiency into the physical world. Much of the work may happen first in simulation, where teams can train and test systems more cheaply. The process then cycles between simulation and physical testing: build and refine in software, test in the real world, feed the resulting data back into the simulation, and iterate.
Skills to build
The speaker recommends developing three foundations that support all five career paths:
- Computer systems: Understand hardware and operating systems, including CPUs and GPUs, compilation, kernels, virtualization, and how software maps to hardware. This helps engineers assess performance and security trade-offs.
- LLMs: Learn the basics of pre-training, training, post-training, reinforcement learning, data labeling, and model deployment. These concepts are useful for building, evaluating, securing, or integrating AI systems.
- Software engineering: Retain a strong grasp of APIs, networking, databases, sockets, and distributed systems so you can judge an agent’s recommendations rather than simply accept them.
He recommends using LLMs as interactive tutors—for example, to create visual explanations, exercises, or step-by-step walkthroughs. He also emphasizes that combining expertise in another field, such as biology, with programming can unlock new applications.
Guides and resources mentioned
- Andrej Karpathy: A roughly two-to-two-and-a-half-hour video on LLMs, described as a useful introduction.
- Sebastian Raschka: YouTube videos for a deeper exploration of the subject.
- The speaker suggests watching educational material alongside an LLM that can explain concepts and adapt its instruction.
Main speaker and sources
The main speaker is Lyubomir, sharing his own predictions and learning plans. He also references Peter Steinberger, a SpaceX example identified in the subtitles as “Potato” or Lauren, Andrej Karpathy, and Sebastian Raschka.
Rate this summary
Your feedback will help improve summaries.
Improve this summary
Reprocess with a stronger model when the summary feels incomplete or inaccurate.
Translate summary in another language
Ask questions to this video
Chat for follow-up questions, clarifications, and source-backed answers.