Code generation and completion
OpenAI Codex / Claude CodeWrite boilerplate, debug, and refactor faster. The skill is reviewing and directing the output, not accepting it blindly.
dear [Software Engineer]
AI writes the code now. Now what?
The question isn’t whether to use AI. It’s whether you understand the system well enough to review what it built.
What does a software career actually look like now?
AI tools now generate, test, and refactor code faster than most of us can type. Engineering teams are 30-40% more productive, meaning companies produce more with fewer people. Entry-level roles that once trained new grads have been cut significantly across Big Tech and startups alike. The engineers being hired today are expected to design systems, review AI output, make product decisions, and ship end-to-end.
The catch: if you never wrote the code in the first place, you don’t yet know what good looks like. New grads face a double burden: leverage AI effectively while building the judgment to know when it’s getting it wrong. The bar for proof of work has gone up, not down.
Skills that stay valuable
- Systems design and architecture
- Code review, debugging, and quality assurance
- AI-assisted development and prompt engineering
- DevOps, deployment, and infrastructure
- Product thinking and cross-functional collaboration
- Security and reliability engineering
- Data engineering and ML pipelines
26%
drop in entry-level software engineering job postings between 2022 and 2024
LinkedIn Economic Graph, 20243x
more code is being written in companies using AI coding tools, with fewer engineers
GitHub, 202571%
of hiring managers now prioritize AI and systems thinking skills over raw coding ability
Microsoft x LinkedIn, 2024“If someone is applying for a software development role and has never used an agentic coding tool like Claude Code or Codex to build a product, that’s an immediate red flag.”
The skills that got you hired aren’t the skills being asked of you now.
Ivan Lee is CEO of Datasaur, a 60-person startup that has raised $8M. He estimates his engineers are now up to 40% more productive with AI. In the past, he would have hired more people to produce the same output. He is still hiring new grads, with an eye to the future: if he doesn’t build the pipeline now, he has no one to grow into senior roles later.
What he looks for has changed. He no longer tries to stop candidates from using AI in interviews. He assumes they will and is looking to see whether they understand what good looks like, and whether they can guide the AI to get there. His engineers are energized by their newfound leverage and ability to ship more than they ever thought possible.
“New grads have to learn to use AI while also learning what good looks like, then use AI to get there. It’s almost double the work. But I’m seeing young engineers figure this out and inspire the rest of our team.”
Proof of work checklist for entry-level engineers
Leetcode scores get filtered. Here are the things that actually get you hired.
A project you shipped end-to-end: what the problem was, what you built, what you used AI for, and what the outcome was
Evidence you can read, review, and improve AI-generated code, not just accept it as-is
A GitHub profile with real commits, not just tutorial clones
A stated point of view on how you use AI coding tools: what you trust them for, where you check everything
Demonstrated systems thinking: can you explain why you made the architectural choices you made, not just what they were
At least one end-to-end ML project: data, model, evaluation, and deployment, even at small scale
Understanding of model evaluation and when a model is good enough versus when it isn’t
Evidence you can work with APIs and foundation models, not just build from scratch
A clear point of view on responsible AI: what guardrails matter and why
Strong Python fundamentals and the ability to debug when AI-generated code fails silently
Hands-on experience with at least one cloud platform: AWS, GCP, or Azure
Evidence you understand the full deployment lifecycle: build, test, ship, monitor
A concrete example of automating something that was manual before
Familiarity with containers, CI/CD pipelines, and infrastructure as code
A point of view on AI-assisted infrastructure management and where you still want human checkpoints
Something real and deployed that users can interact with, however small the user base
Evidence you can move across the stack: you understand why decisions at one layer affect another
A concrete example of using AI to accelerate development without creating tech debt
Strong product instincts: can you explain why you built what you built, not just how
Demonstrated ability to ship iteratively and respond to real user feedback
AI as leverage
System design and architecture
Claude / ChatGPTExplore architectural tradeoffs, draft RFCs, and pressure-test your decisions with an AI thought partner.
Testing and QA
Cursor / QodoGenerate test cases and edge cases you might have missed. Review everything, especially for security-critical paths.
Documentation
Claude / Notion AITurn code into clear documentation and READMEs in minutes. The bottleneck becomes having something worth documenting.
DevOps and infrastructure
AWS Q / CopilotAutomate repetitive infrastructure tasks and troubleshoot faster. Keep human eyes on anything that touches production.
Learning and upskilling
Claude / ChatGPTAI is a patient tutor that knows what you already know. Use it to go deep on concepts, not just copy code.
Support our work
dearCC is an initiative of New Work Foundation, a 501(c)(3) nonprofit (EIN 42-1999114). Help us equip Gen Z grads with AI tools, skills, and community to land their first job.
Our leadership team takes $0 salary.