Has AI Lowered the Quality of Everything? Separating Speed from Real Value

Does anyone else feel like AI has lowered the quality of everything?

A Redditor summed up a common mood: AI makes things faster, but the output feels cheaper, shallower and everywhere. Development speed has jumped, yet the web is drowning in auto-generated slop. Many of us still visit the same sites and feel our own capacity for deep learning is slipping.

“The time it takes to create things has dropped to zero, but the actual value of the output feels incredibly close to zero.”

You’re not imagining the flood. And you’re not alone. But there’s a useful way to separate speed from value – and to work out where AI is genuinely changing things, versus where it’s just adding noise.

If you want the original discussion, it’s here: Reddit thread.

Speed vs value: why faster output often feels worse

AI has made average-quality content almost free to produce. That pushes three effects:

• Supply shock: when the cost to create drops to near-zero, the internet fills with lookalike content. Distribution algorithms reward volume and recency, not depth.
• Compression to the mean: most generative models optimise for “plausible” text or images. Without a great brief, data or editing, the result is safe and samey.
• Weaker learning loops: if you delegate too early, your own mental models don’t form. You finish tasks faster but understand them less.

So yes, on the surface, quality can slide. But that’s not the whole story.

Why AI content looks cheap – and what fixes it

Low-effort in, low-value out

Most “bad AI” is just “bad prompting and no editing”. Models aren’t oracles. They are pattern machines. Real value tends to require:

• Proprietary context: your data, decisions, constraints, and tone – often via retrieval augmented generation (RAG), which lets a model reference your own sources.
• A tight brief and structure: clear audience, goal, angle, and constraints beat “write me a blog post”.
• Human taste and verification: editing for precision, citations, and narrative keeps it out of the generic middle.

Benchmarks vs reality

Headlines that say “X times better” usually mean “a few points on a benchmark”. That might matter a lot for coding or maths, but less for marketing copy. If you want a better read on day-to-day usefulness, community-run evaluations like LMSYS Chatbot Arena can be more grounded than lab scores.

Where AI is actually changing things (quietly)

Even if the public web feels worse, inside teams the picture is different. The big productivity wins tend to be unglamorous:

• Code scaffolding and refactors: starting a feature, porting between frameworks, writing tests, and explaining legacy code.
• Document grind: summarising meetings, drafting first-pass policies, and extracting structured data from PDFs.
• Customer support triage: routing tickets, proposing replies with cited policies, and flagging risky cases.
• Search inside the firewall: private RAG over wikis, contracts and research – with citations and access controls.

These don’t change the websites you visit. They cut cycle time and reduce the pain of repetitive work. That’s progress, even if it’s not cinematic.

Learning in an AI-saturated web: protect depth over dopamine

If you’ve noticed your understanding thinning out, put some friction back:

• Switch to “AI-off” study blocks: read the primary source first, then ask a model to quiz you or clarify your notes.
• Force chain-of-thought (yours): explain concepts aloud or in writing before you check with AI. Compare your steps to the model’s.
• Use models as tutors, not writers: ask for counter-arguments, misconceptions, or Socratic questions. Don’t let it finish the work you should struggle through.
• Collect sources: always ask for citations and sanity-check a random sample. If they’re missing, assume risk of hallucination (confident nonsense).

UK-specific realities: privacy, costs, and compliance

In the UK, there are a few guardrails worth keeping in mind.

• Data protection: UK GDPR applies to inputs and outputs. Check legal bases, retention, and cross-border transfers when using third-party AI. The ICO’s guidance on AI and data protection is a good start: ICO AI guidance.
• Market scrutiny: the Competition and Markets Authority is watching foundation models and vendor behaviour. See its updates here: CMA foundation models papers.
• Regulatory trajectory: the UK’s “pro-innovation” white paper favours sector regulators over a single AI law, which means context matters: UK AI regulation approach.
• Costs and vendors: token pricing and context windows change often – always check official pages: OpenAI pricing, Anthropic pricing, Google pricing.
• Environmental impact: energy and water use are part of the equation. I’ve unpacked the data-centre water cycle here: Does AI waste water? The cooling cycle explained.

“Veo killed video” – why we’re not cooked yet

Launch videos often oversell and cherry-pick. Google’s Veo demos were impressive, but turning a sizzle reel into a dependable tool is slow, especially for cinema-grade control and consistency. If you’re curious, see Google’s overview: Google Veo. The gap between demo and daily, shippable value is where most disappointment lives.

A quick checklist to tell signal from noise

Before you adopt a new model or workflow, ask:

• Does it cut a real bottleneck or error rate on my tasks, not just a benchmark?
• Can I plug in my own data (safely) and get citations for auditability?
• What’s the cost per successful task, including retries and human review?
• Is latency acceptable for my workflow? If it’s slow, will the team abandon it?
• What’s the failure mode, and how will I detect and recover from it?
• Do I maintain a human quality bar – who signs off, and on what evidence?

So, has AI lowered the quality of everything?

It has lowered the barrier to producing average content, which makes the public web noisier and learning feel shallower if you’re not intentional. But inside products and organisations, steady, practical gains are very real – even if they don’t change your homepage.

The opportunity now is to pair speed with originality and accountability. Put your data and judgement in the loop, demand citations, and measure outcomes. The internet may keep filling with three-second throwaways. Your work doesn’t have to.