Here's What I'd Learn Instead of Coding in 2027

Summary

The video argues that software engineers should prepare for a changing role rather than focus only on learning the newest programming languages and frameworks. AI is writing more code, but companies still need engineers to decide what to build, check AI-generated work, manage AI workflows, and coordinate with people.

Hiring and work trends

The speaker cites data suggesting that engineering remains comparatively resilient despite layoffs. In the figures presented, hiring at 12 major technology companies was down 25% overall compared with 2019, while engineering hiring was down 11%. Engineers’ share of new hires rose from 46% to 55%. Early-stage startups reportedly hired 7% more engineers than in 2019.

AI is also changing how engineers spend their time. A developer survey found that more respondents were using AI to write substantial portions of their code, alongside increased time spent reviewing AI-generated code and fixing issues it introduces. An Anthropic workforce report is cited as finding that AI agents generate many work hours but depend heavily on human responses and oversight. The speaker’s takeaway is that human attention and judgment are becoming bottlenecks.

Four skills to develop

1. Verification — Read code you did not write and assess whether it is correct. The speaker recommends checking AI-generated changes for edge cases, race conditions, timeouts, and other potential failures. Fundamentals such as data structures, databases, and networking provide the knowledge needed to evaluate code.

2. Judgment — Turn an ambiguous problem into a defined piece of work, choose an approach, and explain the trade-offs. One suggested exercise is to keep a decision log for a project, recording alternatives and why a particular choice was made.

3. AI agent management — Design workflows that let agents work reliably with less manual supervision. For example, one agent could implement a written specification, another could review the result, and tests could be required to pass before changes are merged. The aim is to catch routine errors early and reserve human attention for important decisions.

4. Communication — Write clear specifications, design documents, and decision records that align both people and agents. Practical exercises include drafting a pull-request description before coding, explaining a technical trade-off to a non-technical person, and documenting decisions where the team can find them.

Career guidance and exercises

• Students: Build a portfolio that demonstrates reasoning, not just code volume. Include specifications, tests, and a decision log, and practice interviews with AI tools allowed.
• Early-career developers: Seek out ambiguous work, such as operational response or migrations, where the problem has not already been fully defined.
• More experienced engineers: Look for roles where expertise makes AI output more useful, such as system design, usability, or security.

Sources and main speaker

• Main speaker: Maddy Zhang, who introduces herself as a senior software engineer and former Google employee.
• Sources cited: SignalFire’s State of Talent report; a developer survey of more than 700 developers; an Anthropic workforce report; a 2026 Stack Overflow developer survey; hiring assessments attributed to “Entropic” in the subtitles; PwC’s 2026 AI Occupations Barometer; and comments attributed to Jensen Huang.