HN Debrief

Jobs and Software Is Fucked

  • AI
  • Jobs
  • Software
  • Gaming
  • Economics

The post is a frustrated first-person account from a game developer who says software hiring is now broken from both ends. Getting interviews is harder, online coding screens are flooded with AI-assisted cheating, and the pressure to adopt AI at work feels like giving up on craft, testing, and the human parts of creative industries that are already under layoff pressure. The piece is not a measured labor-market analysis. It is a burnout document from someone who sees fewer jobs, more noise, and less dignity in the work.

Treat this market as structurally different from the 2020-2022 boom. Rely less on cold applications, show concrete work, and if you are hiring, assume your funnel is filtering out good people long before humans see them.

Discussion mood

Mostly grim and weary. People broadly agreed the software job market is unusually bad, with frustration aimed at broken hiring funnels, fake or overfiltered job postings, and managers using AI as either a replacement fantasy or a convenient justification for doing less hiring.

Key insights

  1. 01

    Rate hikes explain more than AI

    The stronger explanation for the hiring collapse is macroeconomics, not a sudden technical replacement story. Cheap capital during the pandemic created too many software jobs and pulled too many people into the field. When rates rose in 2022, overhiring reversed fast. AI then showed up at exactly the right moment to become the public narrative for a contraction that was already underway.

    Do not plan your career or headcount as if this is only an AI shock. Watch interest rates, startup funding, and enterprise budgets, because those may move hiring faster than model quality does.

      Attribution:
    • commandlinefan #1
    • an0malous #1
    • FloorEgg #1
  2. 02

    Automated screening is blocking real candidates

    The hiring funnel itself is now a material part of the problem. One commenter involved in hiring said resume screening and AI-based judgments produced zero interviewable outside applicants for an SRE and DevOps role, forcing the team to rely on referrals. Another argued recruiters and HR teams have incentives to make screening look complicated and valuable, while fake or already-filled postings add even more dead traffic. That changes the story from "too many applicants" to "bad filters are destroying market matching."

    If you are looking for work, put more effort into referrals and direct intros than resume spray. If you are hiring, audit your screening stack with real examples and measure how many viable candidates it silently rejects.

      Attribution:
    • 8b16380d #1
    • lenerdenator #1
    • ponector #1
  3. 03

    Code quality may lose to acceptable slop

    A hard-nosed business view cut through the craft debate. Many companies do not care about elegant or durable code if the software appears to work long enough to hit the near-term goal. That is why claims that "AI still needs humans to ensure production quality" are less reassuring than they sound. The market may reward the person who ships mediocre AI-assisted output faster than the person who ships cleaner systems more slowly, at least until the costs land on someone else later.

    Tie engineering quality to uptime, incident rates, security, and business risk in ways leaders can see. If you cannot make the cost of slop legible, speed will beat quality in budgeting and staffing decisions.

      Attribution:
    • unknownfuture #1
    • gafferongames #1
    • echelon #1
  4. 04

    Games are a poor fit for code generation

    Game development came up as a concrete case where the replacement story runs ahead of reality. Commenters described Unreal, Unity, Godot, visual scripting, reflection-heavy systems, giant switch statements, and years of engine-specific edge cases that make code look normal while behaving nothing like textbook examples. In that environment, LLMs help more with tutorials, boilerplate, and some debugging than with real feature ownership. Management can still cut staff over AI hype, but the tooling itself is much less capable than generic demos suggest.

    If you work in a messy domain, document why your stack resists automation and show it with examples. If you are evaluating AI rollouts, separate demo-friendly code generation from ownership of systems with hidden state and legacy constraints.

      Attribution:
    • YuechenLi #1
    • yallpendantools #1
    • gafferongames #1
  5. 05

    Demand still exists in narrow technical niches

    The market is not uniformly frozen. People in silicon design, verification, firmware, and infrastructure tied to the AI hardware build-out said hiring remains strong for candidates with domain-specific experience such as PCIe, Ethernet, and DDR. AI-native software roles also still hire, but they increasingly ask for practical experience with coding agents, context handling, and real workflows rather than generic machine learning credentials.

    If you can move toward bottleneck specialties, do it deliberately. The broad middle of generalist software is crowded, while hardware-adjacent and infrastructure-heavy niches still have pricing power.

      Attribution:
    • fl4regun #1
    • IshKebab #1
    • vanuatu #1
  6. 06

    Trades look attractive because the ladder is clearer

    The most striking career-switch story was not mainly about loving hands-on work. It was about predictability. The move from failed software entry attempts into diesel mechanics offered paid training, guaranteed raises, lower applicant competition, and a path to stable income that felt more mortgage-compatible than repeated software applications. That argument resonated because it focused on career structure, not anti-tech romance.

    When comparing careers, look beyond top-end salary. Evaluate training costs, hiring friction, income stability, and whether the path from novice to competent worker is actually visible.

Against the grain

  1. 01

    Refusing AI can become pure self-harm

    The blunt counterargument is that employers will adopt a tool that is cheaper and faster if it is even "good enough". From that angle, refusing LLMs is not protecting standards. It is mostly opting out while everyone else raises output with the same tools. The sensible position here is not full trust or full rejection. It is using LLMs aggressively while still reviewing the result.

    Build competence with AI-assisted workflows even if you dislike the surrounding hype. The market may not reward principled refusal, but it can still reward people who know where the tool helps and where it fails.

      Attribution:
    • atleastoptimal #1
    • teaearlgraycold #1
  2. 02

    Creative panic understates the role of taste

    One pushback against the apocalypse framing was that generated images, writing, and code still need expert judgment to be useful. The scarce thing is not the act of producing another draft. It is the ability to tell whether the draft is wrong, ugly, incoherent, or dangerous. That does not save total job counts, but it does weaken the claim that expertise itself has suddenly stopped mattering.

    If your role depends on judgment, make that judgment visible as a deliverable. Show review criteria, failure cases, and edits that non-experts would miss so your value is not mistaken for mere output volume.

      Attribution:
    • VonGuard #1
    • drdaeman #1
  3. 03

    The only viable move is adaptation

    A smaller but firm camp rejected the despair outright. Their view was that the technology is already out, the frontier labs could disappear tomorrow, and the techniques would still remain. Waiting for a reversal or trying to preserve older workflows is not a strategy. The only useful response is to learn the new tools and use them to make the surrounding mess slightly less bad.

    Do not base your planning on an AI rollback. Assume the tools stay, then decide what to learn, what to avoid, and how to keep your leverage as the workflow changes.

      Attribution:
    • casey2 #1
    • robmn #1
    • micromacrofoot #1

In plain english

AI ↩
Artificial intelligence, here mainly referring to tools that generate code or text.
ChatGPT ↩
A conversational AI product from OpenAI built on large language models.
DDR ↩
Double Data Rate memory, a common type of computer memory used in servers, desktops, and other hardware.
DevOps ↩
A set of practices that combines software development and IT operations to ship and run software more effectively.
Firmware ↩
Low-level software stored close to hardware that controls how devices operate.
Godot ↩
An open source game engine used to build 2D and 3D games.
PCIe ↩
Peripheral Component Interconnect Express, a high-speed hardware interface used to connect components such as graphics cards and storage devices.
SRE ↩
Site Reliability Engineering, a software and operations discipline focused on keeping systems reliable, scalable, and measurable.

Reference links

Career advice and hiring

Macro and labor market signals

AI in creative industries

AI tools in trades and physical work

Books and film references

  • Heat 2
    Mentioned because the original post quoted the famous line from Heat and commenters discussed the film's meaning.