Video summary

Kod Daha Hızlı, İş Daha Ağır: AI Sonrası Yazılımcılık

Main summary

Key takeaways

News and Commentary

Overview

The speaker argues that AI has not made software development “easier” overall. While it can accelerate certain parts of coding, it often increases total workload, context switching, and the complexity of requirements.

AI’s Promised Productivity vs. Real-World Outcome

  • What improved: Early chat-based coding tools reduced the speaker’s time spent writing code and testing.
    • He cites reductions from roughly “half an hour” to much faster cycles.
    • He also references timeframe improvements such as 34 days → 23 days → “half a day” with later tools (e.g., Cursor / cloud coding).
  • What didn’t improve (or worsened): The overall effort and amount of work didn’t decrease. Instead:
    • Developers must handle more iterations
    • Developers end up responsible for more tasks

Shorter Code Writing, More “Tooling” Work

As AI development environments improved, the toolkit expanded (e.g., cloud IDEs, cloud coding assistants, and related workflows).

  • The speaker says these tools often work well.
  • However, they can add many extra steps, such as:
    • Tool calls
    • Checks
    • Follow-up prompts
  • Result: the apparent time savings can shrink and may even backfire.

Deadlines and Expectations Are Accelerating

The speaker contrasts earlier consulting/project approaches with today’s expectations:

  • Then: Projects might take 6–9 months to:
    • understand requirements
    • coordinate work
    • internalize domain knowledge
  • Now: Teams are often expected to deliver a prototype within days or a week.

He believes AI-assisted requirements writing contributes to this compression, which can increase:

  • confusion
  • rework

Lost Context and Meaning in the Requirements Pipeline

A central complaint is that AI can filter and “process” requirements multiple times along the chain:

  • Customer → Business analyst → Developer

The speaker argues this leads to:

  • incorrect assumptions
  • invented details
  • “shifted meaning”

Net effect: reduced clarity and harder implementation of the actual intent.

Rewriting Feedback Can Distort Evaluation

He warns about using AI to rewrite feedback or generate performance evaluations:

  • AI may subtly reframe what someone meant.
  • That can distort the basis for assessment.

Humans Still Must Do the Judgment Work

Even when AI writes code, developers still must:

  • decide what to build
  • integrate solutions into real corporate systems
  • test properly
  • validate against the real problem
  • manage stakeholder coordination

He emphasizes that “real solutions” involve process and decisions beyond code generation.

Psychological and Cultural Effects

The speaker claims AI increases cognitive load and supports a culture of nonstop activity:

  • Cognitive load: more information to interpret
  • Pressure and guilt: anxiety when not producing, likened to a race where everyone else is “moving”
  • Output that gets discarded: he claims many developers generate large amounts that ultimately get thrown away (he mentions “95% ends up in the trash”)

Recommendation: Don’t Quit—Use AI Selectively

He explicitly advises against resignation just because AI exists.

Instead, his stance is to adapt by using AI carefully, not for everything—for example:

  • don’t rely on AI to correct or rewrite human-made requirements/text without verification

Illustrative Side Complaints

He also shares anecdotes including:

  • AI-generated snippets appearing during device work
  • dissatisfaction with AI-generated LinkedIn text
  • frustrations caused by AI outputs containing hidden/undesirable content or formatting issues (including concerns about “hidden signatures” or hidden behavior)

Presenters / Contributors

  • Onur (the video’s speaker/creator; referenced as “Onur”)

Original video