The New Job of a Software Engineer (It's Not Writing Code)

Summary

The video argues that AI is taking over more code-writing, but software engineering jobs are not simply disappearing. Instead, the work is shifting toward deciding what to build, directing AI tools, checking their output, and taking responsibility for what ships.

Evidence cited

• The speaker says Google reports that about 75% of its new code is AI-generated, while Anthropic says more than 90% of its code is written by Claude Code.
• A JetBrains survey of 15,000 developers reportedly found that coding agents write about 47% of the average professional developer’s code. A 2026 Stack Overflow survey found that 65% of developers use coding agents.
• The speaker emphasizes that AI-generated code still requires human review and approval.
• U.S. software development job postings reportedly began rising after a low in May 2025, though they remain below pre-pandemic levels. The market is especially difficult for junior developers.

How the software engineer’s role is changing

1. Define problems clearly. Engineers need to turn vague requests into specific tasks, incorporating business priorities, technical context, constraints, and acceptance criteria that an agent may not know.
2. Break up work and coordinate agents. Rather than prompting one agent at a time, engineers may delegate separate tasks—such as implementation, tests, documentation, and bug fixes—to multiple agents running in parallel.
3. Review and verify code. As AI produces code quickly, checking whether it is correct, secure, and genuinely tested becomes a key bottleneck. The speaker cites an analysis in which AI-generated code passed security tests only about 56% of the time. Understanding code fundamentals remains essential, primarily for judging code rather than writing it from memory.
4. Maintain project context. Teams can use instruction or memory files, such as agents.md or memory.md, to explain codebase conventions, legacy areas, compatibility requirements, and what agents should avoid. System design and architectural decisions also remain important human responsibilities.
5. Be accountable for outcomes. Engineers remain responsible for decisions to ship code, including when AI-generated changes cause outages or data leaks. The speaker says developers commonly prevent agents from making unapproved system changes.

Practical guidance for developers

The video recommends practicing how to:

• Translate ambiguous problems into clear tasks with acceptance criteria.
• Read pull requests and inspect AI-written changes line by line, looking for bugs before running tests or deploying.
• Learn programming fundamentals and system design to evaluate an agent’s work and make sound architectural decisions.
• Use agents regularly on real work, gradually experimenting with multiple agents in parallel.
• Avoid delegating personal judgment to AI. Use the tools to increase leverage while retaining responsibility.

For junior developers, the speaker suggests that AI tools can help them take on work that previously required more experience—but only if they build the skills to understand problems, review code, and make decisions.

Product feature overview

The sponsor segment presents JetBrains Air, which brings coding agents into JetBrains IDEs. The speaker says it supports JetBrains’ Juni agent and Claude Code, allows different agents to run in parallel, and lets agents use IDE capabilities such as finding usages, inspections, and tests.

Users can review changes with familiar IDE tools, run tests or a debugger, and send comments to an agent from a diff before accepting changes. The speaker also mentions a free, lightweight version of Juni called Juni Light.

Main speaker and source

• Tech With Tim — The video’s presenter, providing the analysis and developer advice.
• JetBrains Air — Featured in a sponsored product demonstration.