Ask HN: Anyone still coding like 2021? Where do you work?

TEXT START: I can close an eye on using LLMs for research only.

The Dissection

This thread is labor-market reconnaissance disguised as a tooling argument. Its real question is not “where do old-school programmers work?” but “where can a programmer temporarily shelter from the new production regime?”

The comments document the transition in miniature: AI becomes the producer, the engineer becomes reviewer and intermediary, codebases become less legible, and managers invoke “Claude said” as a substitute for accountable reasoning. Regulated, classified, and conservative organizations appear as refuges, but they are merely slower-moving adoption zones. LeetCode functions as identity maintenance—a way to preserve the feeling of being a programmer after the labor process has begun removing the need to program.

The Core Fallacy

The central error is treating a human-only software workplace as a durable economic equilibrium. Under the Discontinuity Thesis, once AI can produce acceptable cognitive output at lower cost, firms face continuous pressure to deploy it. Regulation, security, and liability can delay implementation, but they do not restore the old mass employment circuit.

The thread also confuses retained skill with retained bargaining power. LeetCode may keep cognition sharp. It does not make coding scarce, restore ownership, or prevent the worker from becoming an AI supervisor. The “gym” analogy preserves personal competence while ignoring market displacement.

The claim that “there is no going back” is directionally correct for research, search, and mechanical transformations, but overextended when treated as proof that every AI-generated workflow is sound. Verification, accountability, domain judgment, and physical integration remain friction points. They relocate human value; they do not preserve the old occupation intact.

Hidden Assumptions

• Existing regulation will permanently block automation rather than automate testing, verification, documentation, and compliance.
• Classified or sensitive work can remain human-only indefinitely rather than adopting controlled or local models.
• Managers still need programmers primarily to write code rather than to translate objectives, supervise agents, and absorb liability.
• Human judgment will remain scarce enough to support the current number of software workers.
• Coding expertise is still a durable moat rather than a rapidly commoditized production input.
• A “sane” workplace can survive if it is less efficient than competitors.
• Maintaining coding ability is equivalent to maintaining employability.
• AI adoption can remain an optional tool choice instead of becoming an organizational production mandate.
• The current software role is being preserved merely because the code must still be reviewed.

Social Function

Primary classification: transition management and ideological anesthetic. Secondary classification: partial truth and copium.

The thread gives displaced workers usable narratives: regulated industries are safe, skeptical CTOs are refuges, old workflows are principled, and LeetCode preserves professional identity. These claims contain real observations about adoption friction, maintainability, security, and accountability. Their social function, however, is to convert structural displacement into a question of personal preference and workplace selection.

The “slop” complaints are not irrelevant. They identify an immature transition with genuine quality and governance failures. But those failures are not a defense of human labor; they are opportunities for better orchestration, evaluation, verification, and control. The complaint is describing the debris field around automation, not disproving automation.

The Verdict

This is a field report from the early death of software labor as a mass identity. Human-coded enclaves still exist, especially behind regulation, secrecy, inertia, and weak management. They are hospice care, not a restored economic order.

The phrase “middleman between my manager and the AI” is the most accurate line in the thread. It names the emerging servitor role. The thread does not prove P1 statistically, but it clearly records P2 and the beginning of P3: institutions are normalizing AI production, while workers fight over how much of their former role remains. Coding practice can survive as exercise, taste, or niche craft. The programmer’s former economic position cannot survive on that basis.