I was a 10x engineer. Now I'm useless.

Summary — technological focus, experiments, and analysis

Thesis / problem

The presenter argues large language models (LLMs) have “oneshotted” their ability to code: AI provides an “easy button” that produces working software quickly but destroys the developer’s craft, motivation, and sense of ownership.

Key concerns:

• Inability to meaningfully review AI-generated code.
• Loss of emotional connection to products.
• Addiction to instant results and the convenience of AI.
• Uncertainty about engineers’ future employability and what it means to be an engineer.

Experiment / technical setup

Goal:

• Move a local app into production so users can download and sign up — defined as the “survival/validation” criterion.

Tools and permissions given to the LLM:

• ChatGPT / Codex (referred to as ChatGPT 5.4).
• AWS command-line access.
• GitHub API access (read pipelines, commit code).

Instruction:

• Perform the full deploy and wiring without human review.

Result

• The LLM completed an end-to-end deployment: the app (shape.work) became live with a functioning download and signup flow.
• Technically successful: pipelines, commits, and deployment were automated and working.

Technical and product-level analysis

• LLMs excel at automating tedious, repetitive, and integration-heavy tasks (CI/CD, infrastructure wiring, commits) much faster than human engineers.
• AI-generated code is often inscrutable and produced faster than humans can review; human-paced code review cannot keep up with AI output.
• The presenter likens “vibe coding” with LLMs to evolution/natural selection: produce many variants and keep whatever passes tests — acceptable if the only requirement is a working product.
• Practical trade-offs:
  • Pros: speed and rapid shipping.
  • Cons: reduced code understanding, maintainability issues, and loss of artisan connection to the product.
• Emotional/product-market consequence:
  • AI-produced products may feel soulless to their creators.
  • Hiring an LLM is not equivalent to hiring passionate teammates who can advocate for, sell, and support the product.

Broader implications and personal/industry concerns

• Psychological dependency: LLMs are compared to drugs — once used, they are hard to stop using.
• Industry split:
  • One camp embraces AI tools to stay competitive.
  • Another rejects them as brittle statistical machines.
• The presenter reports personal harm from adopting these tools and uncertainty about:
  • Interviewing and future employability.
  • The meaning of engineering when traditional craft is devalued.
• Even setting aside ethical concerns (i.e., accepting “vibe coding”), the creator feels disconnected and unwilling to continue building or selling that way.

Possible responses considered

• Cold-turkey: delete/unsubscribe from tools to regain skill and identity.
• Limit AI to mundane tasks and perform difficult work by hand — but convenience often pulls people back in.
• Relearn manual coding and accept the long process required to build products in the traditional way.

Concrete example / reference

• OpenClaw and Peter Steinberger are cited as an example of shipping large codebases without personally reading all code.

Sources / main speakers referenced

• Presenter / video author (unnamed in subtitles) — primary speaker and experimenter.
• Peter Steinberger — referenced regarding OpenClaw.
• ChatGPT / Codex (referred to as ChatGPT 5.4) — LLM used in the experiment.
• AWS CLI and GitHub API — tools given to the LLM for automating deployment.