The key-man test14 minNvidia's founder resigned effective immediately after a week of scandal, leaving the company at the center of the AI economy without the man who built it. The real question now is whether Nvidia's dominance was the product of one person, or a machine that can run without him.Read now →

Developers Were Asked What's Wrong at Work. Most of Their 172 Questions Came Back About AI
When a community of developers was invited to ask anything about their lives, most questions weren't about sleep or dating. They were about AI killing the joy of coding, impossible deadlines and bosses sold on hype. Here's what the research says, and how to handle it without becoming the office naysayer.
Contents
- The questions nobody expected
- What developers are actually saying
- Three problems, not one
- Problem one: nobody knows what happens next
- Problem two: managing management
- What the data says about AI code
- Why "just review it" doesn't work
- Human plus AI isn't automatically better
- The script: how to raise concerns without getting hammered
- Problem three: you
- Why this matters
- Counterpoints
- FAQ
- The bottom line
The questions nobody expected
A large online community with lots of software developers was invited to send in questions for an open Q&A about life. The kind you'd expect: How do I build a relationship? How do I get in shape? Why can't I sleep?
That's not what came back.
There were 172 questions across 26 pages, and the majority were about one thing: AI. Not AI as a cool tool. AI as the thing that changed their jobs, their mood and their sense of whether they chose the right career.
The questions read like a snapshot of software work in late 2026. And the patterns in them point to problems that research is starting to confirm.
What developers are actually saying
Here are the themes, paraphrased from the most-upvoted posts.
"The joy is gone"
One of the most-liked posts, with more than 500 likes, came from a developer who said AI-assisted development had drained the joy from coding. The work, they wrote, is now about finishing tasks, not solving problems. The future looks monotonous, and they feel too deep in the industry to change course.
Another said they used to love thinking critically and solving hard puzzles. Now they mostly pass sentences from one bot to another, like working on a factory line.
"I practiced for eight years. He got the job"
One story stood out. A developer had coded since before high school and spent years practicing web development. During an internship, they avoided AI. A classmate went all in.
Senior colleagues loved the classmate, who fixed old problems in the codebase while an AI agent ran around the clock and he scrolled social media. After the internship, the company called him back. Not the eight-year veteran.
Today the veteran is unemployed. The other, with two years of experience, still works there, and by the poster's account can barely build a simple CRUD app without an LLM. The poster says they now can't enjoy coding even as a hobby.
"Three times the work in a third of the time"
Another theme: management has been sold on AI as the answer to everything. The expectation became three times the output in a third of the time. Code quality dropped. Serious production incidents followed. And the blame, developers say, lands on engineers, who are then told to use more AI to catch the problems next time.
The result: the job becomes reviewing AI-generated code instead of getting into flow and building things. "It's pretty boring, but it pays well at least," one wrote.
More themes from the pile
- A senior engineer at a big tech company who loves shipping code but now gets impossible timelines on every project.
- Bootcamp grads and non-traditional developers anxious about whether their path still exists.
- A paradox: giving non-engineers AI to vibe-code their own dashboards increased demands on engineering, because now everyone wants more.
- A sharp line on workplace agency: responsibility without authority is how many companies treat their human staff, while their technology gets authority without responsibility.
Three problems, not one
Sort the questions and three buckets appear:
- What's actually happening? Will AI take my job? Should I use it? Is this a bubble?
- How do I deal with management? Bosses chasing AI hype, budgets shifting from people to tokens, impossible expectations.
- How do I deal with myself? Despair, low motivation, burnout and, worst of all, boredom.
The honest response to the first bucket is humility. The second and third are where practical help exists.
Problem one: nobody knows what happens next
The core uncertainty is real. Consider the open questions developers keep raising:
- Is there a bubble? Many developers point to circular-looking deals in which chipmakers, cloud providers and AI labs invest in and buy from one another, and wonder how much demand is real.
- Are AI companies losing money? Several big labs have reported large operating losses while subsidizing usage. If prices rise to reflect true costs, some argue, hiring humans could become cheaper again for some work.
- Will it get cheaper and better instead? The pro-AI case is that costs fall fast and capability keeps rising, so today's losses are temporary.
There's a fun argument for the human side: the human brain runs on roughly 20 watts, fueled by snacks and vegetables, not a data center.
But the fair conclusion is that almost nobody knows how this shakes out. Is AI already good enough to do what managers think? Will it get there? Or will aggressive AI mandates blow up in some companies' faces? Anyone who claims certainty is guessing.
You can't control that. You can control the next two problems.
Problem two: managing management
Picture the scene. An executive heard a podcast where an AI company CEO said agents can do everything. They went to a leadership retreat where everyone was "using AI." They asked a model to build a greeting-card app and it worked. Now they want engineering to fire half the team, raise token budgets and ship more.
Many developers chose this career partly to avoid exactly this: business-side group projects instead of satisfying technical work with fellow engineers.
The instinct is to push back: this isn't how it works. That's often correct, and it often fails. Managers excited about AI don't like hearing it. And as the old saying goes, the nail that sticks out gets hammered.
So the advice is closer to jiu-jitsu or aikido: don't block the incoming energy; receive it and redirect it.
Notice who's getting promoted right now: often the loudest AI enthusiast, generating huge amounts of code and token bills, with little thought for architecture. You don't have to become that person. But you do need to stop being the obstacle.
Step one: match their emotional energy
If leadership is excited, show excitement. "I love this AI adoption push. I'm on top of it." That's basic empathy, not surrender.
Step two: share concerns in a form they can accept
Then bring evidence, framed as helping the plan succeed rather than proving it wrong. That's where data comes in.
What the data says about AI code
The most useful evidence comes from DORA, Google Cloud's long-running research program on software delivery, which surveys thousands of technology professionals.
The 2024 finding
DORA's 2024 report found that a 25% increase in AI adoption was associated with an estimated 7.2% decrease in delivery stability and a 1.5% decrease in delivery throughput, even as individuals reported feeling more productive.
Two cautions if you cite it:
- It's about AI adoption, not lines of code. Don't claim "every 25% more code means 7% more bugs."
- Don't extrapolate linearly. Saying "doubling output will cut stability by 30%" goes beyond what the study shows.
The 2025 and 2026 updates
The picture has evolved:
- DORA's 2025 report found AI use nearly universal, about 90% of respondents, and now positively associated with throughput. But AI adoption was still associated with higher delivery instability. DORA's framing: AI is an amplifier. It magnifies strong engineering systems and dysfunctional ones alike.
- A 2026 DORA report on return on investment argued that strong foundations, such as solid platforms, automated testing and fast feedback loops, determine whether AI pays off, and that teams often see a dip before gains.
That's a gift for anyone talking to management. The message isn't "AI is bad." It's "AI pays off when we invest in the guardrails."
Reframe bugs as early warnings
One clever reframing: more AI code surfacing more bugs can be pitched as accelerating discovery. Problems that would have stayed hidden show up sooner, if you budget time to fix them.
Why "just review it" doesn't work
The common corporate answer: let AI write the code and have humans review it.
A new large-scale study suggests that's shakier than it sounds. Researchers at the University of Saskatchewan mined 330,343 multi-file pull requests, covering 756,814 file instances from 182 GitHub projects across five programming languages. They examined whether a file's position in a pull request relates to later bug fixes, and whether review comments align with where bugs actually were.
Findings:
- Files lower in the list had slightly higher latent bug rates, rising from 56.7% at position 1 to 61.5% at position 30. Most review tools list files alphabetically.
- Review attention dilutes as pull requests grow. Each extra modified file reduced the odds that a given file got any review comment by about 8.7%.
- Mid-sized pull requests, around ten files, had the lowest observed risk.
The authors' conclusion: an "attention-effectiveness gap." Visible review activity doesn't necessarily translate into defect prevention.
Many developers add that AI-generated code is harder to review, often lacking the reasoning, comments and documentation that make human code legible.
The CFO test
Here's a vivid way to explain it to an executive. Hand a CFO a five-page spreadsheet and ask them to find the mistakes. They probably can. Then pull out 300 more pages and say the same thing. That's what "AI writes, you review" often means. What happens to anyone's error-catching as volume explodes?
Use that analogy to explain, not to score points. Making the CFO feel foolish won't help you.
Human plus AI isn't automatically better
There's a deeper research point. A 2024 meta-analysis in Nature Human Behaviour by MIT researchers reviewed 106 experiments with 370 effect sizes. On average, human-AI combinations performed worse than the best of humans or AI alone, though they beat humans alone.
The details matter:
- Combinations lost ground on decision tasks and did better on content-creation tasks.
- When humans were better than the AI alone, combining helped. When the AI was better, combining hurt.
- The studies date from 2020 to mid-2023, before today's frontier models.
Translation for the office: "human in the loop" isn't magic. Pairing people and AI needs designing.
The script: how to raise concerns without getting hammered
Put it together and you get a practical playbook.
1. Lead with enthusiasm
"This is a great direction, and I want to help us do it well."
2. Present risks in a familiar business frame
A SWOT analysis (strengths, weaknesses, opportunities, threats) works because executives know it:
- Strengths: our team, our product, leadership's push to modernize.
- Weaknesses: gaps in testing and review capacity.
- Opportunities: real productivity gains from AI.
- Threats: instability, security vulnerabilities, outages on key products, legal exposure, and evidence that code review alone won't catch it all.
You're adding the bad to the good, not declaring doom.
3. Serve "protective truths," not hard truths
In a therapist's office, hard truths heal. In a company, people often shoot the messenger. So frame risks as protecting the company's goals: "There's a non-zero chance of serious security issues. Here's what the data on AI-assisted debugging suggests. I'd love to work with leadership on a plan that hits our financial targets safely."
4. Admit the data is moving
Most hard numbers are a year or two old and models have changed. The instability effect could be smaller now, or larger. Saying "we don't know, so let's measure" makes you sound rigorous, not negative.
5. Document everything
Send a short email summarizing your concerns and the evidence. If things break and blame comes your way, you have a record.
6. Skip "I told you so"
When problems hit, say "we should recalibrate," not "you ignored me." If someone comes at you hard, you can firmly point to the data you raised. Play nice by default, and stand your ground when you must.
You have more social skill than you think
The stereotype of the socially clueless developer is outdated. Most developers are perfectly capable socially. They simply haven't trained this skill the way HR and management have.
In tabletop-game terms, your charisma stat may be 15. You just lack proficiency in persuasion. That's learnable.
Problem three: you
The last bucket may matter most.
When everything around you is uncertain, you can't control whether AI takes your job or whether the bubble pops. What you can control is your own state of mind.
Uncertain times are when you most need:
- Deep motivation reserves, so a bad week doesn't become a bad year.
- Resilience, because there will be setbacks.
- Courage over despair.
- Clear perception, so fear doesn't distort how you see your options.
If you're depressed on top of an AI-driven shakeup, every decision gets harder. Taking care of your mental health is not a luxury here. It's strategy.
Practically, that can mean sleep, exercise, therapy or coaching, talking to peers in the same boat, and finding meaning outside work while the job is in flux. And it can mean deciding, deliberately, how you'll use AI: which parts of the craft you'll keep doing yourself because you love them, and which parts you're happy to hand off.
Why this matters
- Burnout is a business risk. Teams pushed to ship three times faster with weaker review create the incidents that cost companies money and reputation.
- The evidence backs developers' instincts. DORA's stability findings and the code-review study suggest "more AI code, same review process" is risky.
- Promotion signals are skewed. If companies reward raw output and token spend, they may lose the engineers who think about architecture.
- Careers are being decided now. The eight-year coder who lost out shows how quickly hiring norms can shift. Being skilled isn't enough; being visible and fluent with AI matters too.
Counterpoints
- AI does help many developers. DORA 2025 found AI now linked to higher throughput, and many engineers report real gains.
- Some resistance is nostalgia. Every major tooling shift, from compilers to frameworks, changed what "real coding" meant.
- Correlation isn't causation. DORA's figures are associations from surveys, and the code-review study measured modest effects.
- Social tactics have limits. At a toxic company, no script fixes bad leadership. Sometimes the right move is to leave.
- These are self-selected voices. People who send questions to an open Q&A skew toward those who are struggling.
FAQ
Is AI making code less stable?
DORA's 2024 report linked higher AI adoption to lower delivery stability, and its 2025 report found that link persisted even as throughput improved. Strong testing and platform practices appear to reduce the risk.
Does code review catch AI bugs?
Not reliably at scale. A study of 330,343 pull requests found review attention dilutes as pull requests grow, and visible review activity doesn't necessarily prevent defects.
Should I refuse to use AI at work?
Probably not. The safer path is to adopt it visibly, raise risks with data, propose guardrails and document your concerns.
Is "human plus AI" always better?
No. A meta-analysis of 106 experiments found combinations on average performed worse than the best of humans or AI alone, with gains mainly in creative tasks.
The bottom line
Developers aren't imagining it. The job is changing fast, often in ways that make it less satisfying, more pressured and, according to the data, riskier for the software itself.
But fighting management head-on rarely works. The developers who come out ahead will likely do three things: stay honest about uncertainty, translate their concerns into the language executives already speak, and protect their own mental health while the ground shifts.
Match the energy. Bring the data. Write it down. And don't let a hype cycle convince you that the skills you spent years building are worthless. They may be exactly what keeps the AI-written code from falling over.
Sources
- Google Cloud: Announcing the 2024 DORA report
- DORA: State of AI-assisted Software Development 2025
- InfoQ: New DORA report claims strong engineering foundations drive AI ROI
- arXiv: Does Order Matter? File ordering and code review effectiveness
- Nature Human Behaviour: When combinations of humans and AI are useful





