Amazon spent $200bn on AI infrastructure this year. It pushed every corporate employee to use AI tools daily. It tracked adoption through management dashboards. And the result? Worse code quality, longer development cycles, and a workforce that feels surveilled rather than supported.
A Guardian investigation published on 11 March 2026 spoke to more than half a dozen current and former Amazon corporate employees across software engineering, UX research, and data analysis roles. Their accounts paint a consistent picture: mandatory AI adoption without clear strategy is creating more work, not less.
This is not an Amazon-specific problem. It is a pattern emerging across organisations that confuse AI deployment with AI value.
The real story behind the numbers
Amazon’s AI push exists within a specific financial context that every business leader should understand before drawing conclusions.
| Metric | Figure | Context |
|---|---|---|
| AI infrastructure spend (2026) | $200bn | Plus $50bn OpenAI investment |
| Corporate layoffs (last 4 months) | 30,000 | ~10% of 350,000 corporate workforce |
| AI tool success rate (employee estimate) | ~33% | Lisa, supply chain engineer with 10+ years at Amazon |
| Revenue trend | Growing quarterly | Layoffs happening alongside record revenue |
Critical Context: Amazon is the second-largest employer in the United States. Its workplace practices influence standards across white-collar and blue-collar industries globally. What happens inside Amazon does not stay inside Amazon.
The gap between investment and outcomes here is striking. Amazon is not short on resources, talent, or technical infrastructure. If forced AI adoption is failing at this scale, it should prompt serious questions about the approach itself.
What’s actually happening inside Amazon
The Guardian’s sources describe a workplace where AI adoption has become performative rather than productive. Three patterns stand out.
Pattern 1: The slop cycle
Dina, a New York-based software developer, joined Amazon two years ago to write code. Now she mostly fixes what AI breaks. The internal tool Kiro “frequently hallucinates and generates flawed code,” she told the Guardian. Her workflow has become circular: use AI to generate code, discover it is broken, spend time debugging or reverting, start again manually.
A colleague in Denny’s team claimed an AI agent saved a week of developer effort on a feature. When Denny checked the actual code review, he found “dozens of comments from colleagues pointing out basic issues.” His assessment: the development cycle would likely take the same amount of time or longer.
Reality Check: When AI-generated code requires extensive human review to catch basic errors, you have not saved developer time. You have redistributed it from creation to correction, often with a net increase.
Pattern 2: Hackathon-driven tooling
Amazon typically runs quarterly hackathons. Sometime last year, these shifted primarily to generative AI hackathons, producing waves of internal developer productivity tools. Employees describe being shown “random tools” by managers and told to try them.
Denny called these tools “half-baked” and said they add to his workload because he has to evaluate each one. The tools come with surveys about the experience, consuming more time. This is not innovation. It is organisational busywork dressed up as transformation.
Strategic Insight: Internal AI hackathons can generate useful prototypes. But pushing unvetted hackathon outputs onto teams as mandatory tools without proper testing, training, or fitness-for-purpose assessment creates drag, not acceleration.
Pattern 3: Surveillance as strategy
Amazon’s approach to measuring AI adoption reveals a fundamentally misguided theory of change. The company shifted its internal feedback system (Amazon Connections) from questions about team functioning and job satisfaction to questions about AI usage frequency. Managers now have dashboards tracking individual AI tool adoption, with some targeting 80% weekly usage across their teams.
Promotion documents reportedly include a new question: “How did they leverage AI?” The Wall Street Journal reported that “managers do consider who is all-in on AI when it comes to promotions.”
Warning: When you measure AI adoption by frequency of use rather than quality of outcomes, you incentivise performative adoption. Employees will use tools to be seen using tools, regardless of whether those tools help.
The human cost organisations keep ignoring
The Amazon case exposes something deeper than a failed tool rollout. It reveals what happens when organisations treat AI adoption as a compliance exercise rather than a capability-building process.
Who bears the real cost?
| Stakeholder | Impact | Risk level |
|---|---|---|
| Junior engineers | Stunted learning curves from offloading work to AI; asked to train their own replacements | High |
| Senior engineers | Time consumed reviewing AI-generated slop instead of solving hard problems | Medium-High |
| Managers | Pressured to hit adoption metrics regardless of team productivity | Medium |
| Customers | Service outages linked to AI tool failures (13-hour incident in December 2025) | High |
| The organisation | Declining code quality, institutional knowledge loss, talent attrition | Critical |
Sarah, an early-career Amazon engineer, put it bluntly: “Part of my new job role, it feels like, is being asked to train the AI to essentially replace you.” She worries that offloading her work to AI is stunting her professional development. This is not resistance to change. It is a rational assessment of career risk.
Hidden Cost: When junior staff are asked to document their workflows so AI can replicate them, you are extracting institutional knowledge whilst simultaneously preventing those staff from developing deeper expertise. This is a knowledge management crisis disguised as an efficiency programme.
Nick Srnicek, author of Platform Capitalism and senior lecturer at King’s College London, identified the structural issue: “The rushed deployment of AI means an uncritical expansion of surveillance since these tools increasingly require detailed knowledge of personal workflows and data. To make them more capable means giving management greater insight and control over workers’ everyday activities.”
What successful AI adoption actually requires
The Amazon experience provides a clear counter-example. Here is what works instead, based on the patterns that fail at scale.
Priority actions by organisational maturity
Early stage (exploring AI tools):
- Let employees identify where AI helps their specific workflows
- Measure outcomes, not adoption rates
- Provide structured training before mandating tool use
Growth stage (deploying AI across teams):
- Establish quality gates for AI-generated outputs before they enter production
- Create feedback loops where teams can reject tools that do not fit their use cases
- Separate experimentation from production workflows
Mature stage (AI embedded in operations):
- Monitor productivity holistically, not just AI usage metrics
- Invest in human skills that complement AI capabilities
- Build institutional processes for AI tool evaluation and retirement
Success Factor: Ifeoma Ajunwa, founding director of the AI and Future of Work Program at Emory University, told the Guardian that forcing employees to adopt tools “usually backfires.” Her assessment: “Generally, employees are in a better position [than management] to determine what tools can aid productivity.”
The framework that actually works
- Problem-first, not tool-first. Lisa, the supply chain engineer, framed it perfectly: “You don’t look at the problem and go, ‘How do I use this hammer I have?’ You look at it and go, ‘Is this a problem for a hammer or something else?’”
- Quality gates before scale. AI-generated code should pass the same review standards as human-written code. If it cannot, it should not ship.
- Training before mandates. Amazon employees reported seeking AI training on their own, with internal sessions focused on speed rather than quality. One trainer advised employees to “ask the AI to check its own work” - advice that fundamentally misunderstands how these tools fail.
- Measure what matters. Track development cycle time, code quality metrics, customer-facing incidents, and employee satisfaction. Not how often someone opened an AI tool.
Four non-obvious challenges in mandatory AI adoption
1. The productivity measurement trap
When Amazon reports that “the vast majority of our teams” find value in AI tools, it is measuring self-reported sentiment from employees who know their AI usage is being tracked and their promotions may depend on enthusiasm. This is not productivity data. It is compliance data.
Strategic Reality: Self-reported AI productivity gains in environments where AI adoption is tied to career progression are unreliable. You need independent measurement of actual output quality and cycle times.
2. The institutional knowledge drain
Junior employees who offload work to AI miss the learning-by-doing phase that builds deep expertise. Sarah’s concern about her stunted learning curve is not just personal - it represents a systemic risk. Organisations that skip the human learning phase will find themselves dependent on AI tools without the internal expertise to evaluate, correct, or replace them.
3. The outage multiplier
Amazon experienced at least two outages connected to internal AI tools, including a 13-hour customer-facing interruption in December 2025 after engineers “allowed its AI tool to make certain changes.” As AI-generated code proliferates without adequate review, the surface area for these failures grows. Each outage undermines customer trust and internal confidence simultaneously.
4. The morale-productivity spiral
Mandatory AI adoption combined with layoffs creates a toxic dynamic. Employees are told AI will make them more productive, then watch colleagues get laid off whilst the company posts record revenue. Maria, a former product manager laid off in January 2026, described the underlying logic: “If you say you automated away two hours of someone’s job, you need to convert that into savings on that job title. That’s the unspoken math of what they’re doing.”
Competitive Reality: Over 1,000 Amazon workers signed a petition raising concerns about the “aggressive rollout” of AI tools. When your best employees are organising against your transformation strategy, the strategy has failed - regardless of what adoption dashboards show.
The core lesson for UK organisations
Amazon’s experience is not a story about AI failing. The tools themselves are secondary. This is a story about what happens when organisations pursue adoption metrics instead of adoption outcomes.
Three factors that determine whether AI adoption creates or destroys value:
- Employee agency in tool selection. Workers who choose to use AI for tasks where it genuinely helps will outperform workers who are forced to use AI for everything. Every credible study on technology adoption confirms this.
- Quality standards that do not bend for speed. Amazon’s CEO urged employees to “get more done with scrappier teams.” When speed becomes the primary value, quality controls erode. AI makes this worse, not better, because it can generate plausible-looking output at volume.
- Honest measurement that separates activity from impact. Dashboard metrics showing 80% weekly AI tool usage tell you nothing about whether your organisation is moving faster, producing better work, or serving customers more effectively.
What to do next
- Audit your current AI adoption approach. Are you measuring usage or outcomes? If your primary metric is how many people use AI tools, you are on the Amazon path.
- Talk to your teams. Not through surveys attached to AI tools, but through genuine conversations about where AI helps and where it creates friction.
- Establish quality gates. Before any AI-generated output reaches customers, it should meet the same standards as human-produced work.
- Protect learning pathways. If junior staff are using AI to skip foundational skill development, you are building a workforce that cannot function without tools it cannot evaluate.
The organisations that will benefit most from AI are not the ones that adopt fastest. They are the ones that adopt most thoughtfully.
Source: Varsha Bansal, “Amazon is determined to use AI for everything - even when it slows down work,” The Guardian, 11 March 2026.
Get in touch to discuss your approach.