AI and Enshittification: How to Use AI Without Making Technology Worse

AI can now help a business build faster, publish faster, automate faster, and scale faster. But what if faster is not the same as better?

In boardrooms across the globe, artificial intelligence is hailed as the ultimate efficiency catalyst. Executive teams are deploying Large Language Models (LLMs), launching autonomous AI agents, and cutting operational overhead at a record pace. Yet, beneath these metrics of raw speed lies an uncomfortable reality: unbridled output is rapidly eroding digital quality.

When automation is stripped of human judgment and architectural control, it creates systemic platform decay. This phenomenon known as AI enshittification is the single greatest strategic threat to long-term customer trust, brand equity, and platform survival.

Navigating this challenge requires a clear understanding of how forward-thinking leaders can leverage responsible AI development to build products that become inherently more valuable as they scale.

What Is Enshittification?

Coined by technology writer Cory Doctorow, enshittification (also termed platform decay) describes the systematic degradation of digital products and online services.

AI enshittification refers to the deterioration of digital products, services, or user experiences when AI is deployed primarily to maximize output volume, short-term monetization, or cost-cutting at the expense of user trust, functional utility, and long-term value creation.

Doctorow breaks this decay down into three sequential stages:

1. User Focus: A product launches with exceptional utility, offering genuine value to capture an active, loyal user base.
2. Platform & Business Focus: Once users are locked in by high switching costs, the platform shifts its priority toward monetizing those users to benefit business partners or advertisers.
3. Value Extraction: Finally, the business squeezes every fraction of profit out of both users and partners to satisfy short-term targets, leaving behind a cluttered, low-utility shell.

While this cycle historically defined social media platforms and digital marketplaces, artificial intelligence acts as a high-powered accelerant.

Why AI Could Accelerate the Problem (With Real-World Examples)

AI does not inherently ruin technology. However, it operates as a force multiplier: AI can scale bad business decisions just as efficiently as good ones.

When companies treat generative AI as a shortcut to mass execution rather than a tool for quality enhancement, they trigger platform decay across real operational channels:

• Mass Pollution in Search & Content: Google Search's launch of AI Overviews was designed to synthesize information. However, early deployments famously hallucinated absurdities like recommending non-toxic glue to keep cheese on pizza or advising people to eat rocks based on satirical posts. When companies use AI to flood the web with synthetic, SEO-farmed articles, search engines become polluted with low-value noise.
• The Automated Customer Support Wall: Major airlines and e-commerce platforms have aggressively replaced human support agents with under-trained LLM chatbots to cut payroll. The result? Customers trapped in automated loops, unable to resolve urgent booking errors or refund claims, shifting the burden of resolution friction entirely onto the consumer.
• Feature Bloat & Cognitive Load: Workplace communication tools like Slack and Zoom have pushed aggressive AI summarization tools and conversational pop-ups into their primary interfaces. Instead of simplifying work, forcing unprompted AI features onto users increases UI clutter and distracts from core functionality.
• Algorithmic Engagement Traps: Social platforms use predictive models optimized strictly for watch-time and click-through rates. This pushes sensationalist, AI-generated "slop", fake videos and synthetic outrage to the surface, alienating users who seek genuine connection.
• Codebase Bloat & Technical Debt: Engineering teams using auto-generating AI coding assistants without strict human architectural oversight often push thousands of lines of bloated, repetitive code into production, introducing hidden vulnerabilities and long-term maintenance overhead.

The Hidden Cost Driver: AI Token Limits, Context Decay, and System Overhead

A major factor accelerating product decay is a misunderstanding of how LLMs process data: token economics, context window limits, and cost trade-offs.

What Is an AI Token?

A token is the fundamental unit of data an LLM processes. In English text, 1 token is roughly equal to 0.75 words (or 4 characters).

• A 1,000-word document consumes ~1,333 input tokens.
• Every prompt sent to an API (input tokens) and every response generated (output tokens) incurs a direct financial cost.

Context Window Limits and the "Lost in the Middle" Effect

Every model has a strict context window limit, the maximum number of tokens it can read and hold in memory at one time (e.g., 8,000 tokens up to 1M+ tokens in modern models).

When businesses dump massive amounts of unstructured data into an LLM context window without architectural optimization, three technical issues occur:

1. Context Decay & Degradation: Research demonstrates that LLMs suffer from the "Lost in the Middle" problem. When context windows are stuffed with thousands of tokens, models accurately recall information at the very beginning and very end of the prompt, but miss critical facts buried in the middle.

1. Exponential API Costs: Naive implementations send the entire conversation history back to the model with every new user query. A chatbot conversation that starts at 500 tokens can quickly grow to 20,000 tokens per interaction, scaling API costs exponentially without adding any new value.
2. Latency Inefficiencies: Processing massive token payloads significantly slows down model inference time (Time-To-First-Token), forcing users to sit and wait for responses, directly ruining the user experience.

The Real Problem Isn't AI — It's What We Optimise For

The root cause of platform degradation is rarely the underlying neural network; it is the executive metric chosen to guide the system.

Evaluating an AI business strategy purely through short-term productivity indicators creates a false economy. Sustainable value creation requires shifting focus from vanity volume metrics to genuine outcome metrics.

How Businesses Can Use AI Without Creating “Enshittification”

Preventing digital decay requires disciplined leadership, clear governance, and a commitment to responsible AI development.

1. Start with the User Problem, Not the AI Technology

Technology decisions must follow operational needs. Never begin an initiative with the directive, "Where can we add generative AI?" Instead, map customer pain points, workflow latencies, or data bottlenecks and apply AI only where it represents the most effective solution.

2. Implement Smart Token Management and RAG Architecture

To prevent context decay and runaway API costs, avoid dumping raw documents into LLM prompts. Implement Retrieval-Augmented Generation (RAG): vectorize enterprise knowledge bases, retrieve only the most relevant 3–5 context chunks per query, and use semantic caching to instantly answer repeated queries without incurring new token charges.

3. Measure Outcomes, Not AI Usage

Avoid vanity metrics such as total automated queries or tokens generated. Track structural indicators instead: Has resolution time decreased? Has output accuracy improved? Are customer retention rates increasing?

4. Keep Humans Involved Where Judgment Matters

High-stakes decisions such as strategic planning, complex enterprise support, creative direction, and core system architecture require human context. Implementing robust human-in-the-loop workflows ensures AI handles mechanical execution while humans maintain quality standards.

5. Use AI to Improve Quality, Not Just Increase Volume

Rather than leveraging automation to blast generic sales emails or pump out superficial marketing copy, deploy AI to analyze deep datasets, catch edge-case software bugs, aggregate research, and personalize enterprise workflows.

AI Should Remove Friction, Not Add More Complexity

The best software innovations feel invisible. High-value AI implementations do not demand constant user interaction; they eliminate steps in the background.

Effective AI and user experience design focuses on streamlining work:

• Automating routine data extraction, transformation, and background entry without interrupting focus.
• Reducing the number of clicks required to surface business intelligence.
• Surfacing precise operational context to decision-makers at the exact moment of execution.

The Role of Human-AI Collaboration

Long-term success relies on structured human-AI collaboration. Artificial intelligence excels at rapid pattern recognition, processing unstructured datasets, and executing routine steps. However, human teams remain essential across four key pillars:

• Context: Navigating business nuances, organizational realities, and brand integrity.
• Ethics & Governance: Enforcing data privacy, regulatory compliance, and system safety.
• Strategy: Defining long-term direction that historical data cannot predict.
• Accountability: Taking ownership of outcomes, technical architecture, and customer trust.

Organizations that succeed use AI to augment human capability rather than treat human judgment as an operational expense to be eliminated.

How InheritX Approaches AI With Business Value in Mind

At InheritX Solutions , we engineer technology built to solve concrete operational problems. We reject the practice of treating artificial intelligence as a marketing gimmick or a mechanism for mass, low-quality output.

As a custom software engineering and AI solutions partner, InheritX works across machine learning development, autonomous AI agent integration, optimized RAG pipelines, and enterprise application development. We help organizations design, build, and deploy production-ready systems engineered for long-term scalability and cost efficiency.

Our engineering approach centers on:

• High-Impact Framing: Pinpointing precise operational workflows where AI yields clear, verifiable ROI.
• Production-Ready MLOps & Token Budgeting: Building secure, observable data pipelines with semantic caching and optimized context retrieval to minimize latency and token costs.
• Seamless Integration: Embedding intelligent automation directly into existing software stacks with minimal operational friction.
• Measurable Business Impact: Benchmarking every deployment against actual performance indicators, speed, accuracy, process efficiency, and user retention.

A Simple Framework for Better AI Adoption

Before deploying any new AI capability, evaluate the initiative against this five-point framework:

 1. PROBLEM  ──► What specific operational bottleneck are we solving?

  2. PURPOSE ──► Is AI the simplest, most efficient tool for the job?

  3. PEOPLE ──► Where are human review, judgment, and oversight anchored?

  4. PERFORMANCE ──► What clear quality, token budget, and ROI metrics define success?

  5. PROOF ──► Does production data confirm an improved user experience?

• Problem: Define the concrete operational bottleneck or user pain point before writing code.
• Purpose: Verify that AI offers a clear quality advantage over simpler programmatic options.
• People: Establish clear boundaries for human oversight, review, and accountability.
• Performance: Measure operational speed, accuracy, token efficiency, and customer retention—never raw generation volume.
• Proof: Continuously audit production systems to verify that the implementation maintains software integrity as it scales.

Frequently Asked Questions

What is AI enshittification?

AI enshittification refers to the progressive deterioration of digital products, platforms, or user experiences caused by deploying AI primarily to maximize output volume, cut short-term costs, or inflate vanity engagement metrics at the expense of genuine user value.

What are AI tokens and token limits?

A token is the fundamental chunk of text (approx. 0.75 words) that an LLM processes. Token limits define the context window, the maximum volume of input and output data a model can handle in a single query before forgetting earlier context or incurring excessive costs.

How do token limits impact user experience?

When applications exceed optimal token limits or stuff raw text into prompts, models suffer from "context decay" (forgetting key details), increased response latency, and higher error rates, leading to degraded software interactions.

How can businesses optimize token usage to lower costs?

Businesses can optimize token usage by implementing Retrieval-Augmented Generation (RAG) to fetch only relevant data, using semantic caching for repeated prompts, trimming prompt conversation history, and selecting smaller, fine-tuned models for specific tasks.

What is responsible AI adoption?

Responsible AI adoption is the deliberate engineering of artificial intelligence centered on systemic reliability, user privacy, transparent governance, optimized token economics, and long-term value creation.

How can businesses measure whether AI is actually creating value?

Value must be measured through business outcome metrics such as task completion rates, lower error rates, reduced resolution times, optimized cost-per-transaction, and higher customer retention rather than vanity metrics like total AI queries or content generated.

The ultimate objective of AI adoption is not to deploy more technology for its own sake. It is to build intelligent, scalable systems that remain reliable, cost-effective, and trusted as your business grows.