Amazon built a token leaderboard to push employees to use AI more. Employees gamed it, generating useless output just to rank higher and Amazon eventually pulled it down. Meta built something similar. Microsoft pushed engineers toward GitHub Copilot and engineers who preferred other tools used those instead, because no enforced policy said they could not. Meta and AT&T have reportedly started pulling back on AI use as costs keep climbing. These are not small companies stumbling through an early experiment. They have resources most organizations will never see and they are still getting this wrong because they went looking for results before they understood what they were working with.
That is the pattern. Companies see the end result they want and skip the work that gets them there, then act surprised when the costs come in hot and the outcomes do not follow. They want efficiency and productivity gains but will not sit down and actually understand the technology, understand their own business or build a strategy that connects the two. That gap between wanting the result and doing the work is exactly where AI sprawl lives.
The Short Answer
Companies that adopt AI without policies, training or a designated toolset are not executing a strategy. They are producing AI sprawl, the unmanaged accumulation of tools across departments without shared governance, consistent output or centralized cost control. The lack of foundation results in higher costs, weaker collaboration and a workforce with no real framework for the technology they were told would help them with their work.
This AI sprawl is caused by unmanaged accumulation of AI tools deployed across an organization by different teams without a shared inventory, central approval, governance structure or cost oversight.
What Is AI Sprawl?
AI sprawl happens when AI tool adoption outpaces the structure needed to govern it. One person is using Claude, another is using Gemini, a third is on ChatGPT and someone else is running Perplexity, with no central approval, no shared standard and no way for leadership to track what the whole picture is actually costing. Every department added what worked for them and the accumulation is what the bill reflects.
Business Insider named what employees are actually doing inside this environment: botsitting. That is giving AI the context, corrections and edits it needs before the output is usable. The assumption was that AI would save time but what it created in many cases is a new task inside the existing job. Workers say they save time individually but only 13% say those savings have significantly improved their company’s performance. The individual gains and the organizational returns are two very different things and most companies are only measuring one of them.
The AGI Assumption Is Costing Companies Real Money
A significant part of what is driving bad AI decisions is a misunderstanding of what the technology actually is right now. Many companies are making staffing, budget and strategy decisions as if artificial general intelligence already exists. It does not.
AGI would think on its own. It would develop its own strategies, learn what it does not know and make decisions without human direction. The AI that exists right now does not do that. It works from patterns. It needs human input, human iteration and human oversight at every stage to produce anything useful. It can handle automation. It can handle repetitive workflows. It is a powerful tool. It is not a replacement for human thinking and we are far from the point where it becomes one.
That distinction matters enormously because companies that confuse the two make decisions they cannot easily walk back. The most dangerous one is laying off staff on the assumption that AI can fill the gap. Here is what actually happens: the layoffs go through, the severance gets paid, the AI costs come in higher than projected and the company discovers what AI cannot do that those people were doing. Then hiring starts over. The company has paid severance, absorbed inflated AI costs and now they have to find, hire and bring new people up to speed for roles it just eliminated. They spent more money trying to save money because they did not understand the technology they were betting on.
Why Gamifying AI Made Everything Worse
Amazon did not just want employees to use AI; they wanted to measure it, rank it and reward it. The leaderboard turned AI usage into a competition and competitions produce what they always produce: people optimizing for the score, not the outcome. Employees generated output to rank higher and the output itself became secondary. Amazon pulled the leaderboard, but the episode shows exactly what happens when you push for usage before building a foundation for it. If you gamify it, you should not be surprised by what you get.
The Microsoft situation is a different version of the same problem. Engineers who preferred tools other than the one Microsoft was steering toward used those tools because nothing prevented them from doing so. That is not a failure of individual employees. That is what happens when leadership says use AI without specifying which AI, without building rules around that expectation and without giving employees a clear framework to follow. The absence of a policy is a decision. It is just not a strategic one.
Why AI Without Structure Weakens Collaboration
The outcome people least expect from AI sprawl is weaker collaboration. More tools should mean more speed and more output. What companies are actually finding is that more uncoordinated tools produce more disconnected results and collaboration cannot survive disconnected results.
At Travelport, chief transformation officer Lee Senderov described what this looks like at the individual level: one employee burned through 160 times the tokens of the next most prolific AI user over just four days. Two people using different tools can spend time and budget creating near-duplicate work independently when they would have previously worked together to produce one deliverable. That is not efficiency. That is individual productivity cannibalizing team output.
Axios HQ surveyed more than 1,200 executives and employees in 2026 and found that missed deadlines from poor communication doubled in one year. AI increased the volume of communication inside organizations and decreased the clarity of it at the same time. Only 26% of employees say their company has clear, enforced AI use policies, which means most people are guessing. They are making individual calls about which tool to use, what to put into it and how to handle the output and those individual calls are stacking up across every department into something leadership will eventually have to explain.
What AI Sprawl Actually Costs
The real cost of AI sprawl is not found in subscription fees. It lives in the time lost to fragmentation, the security risks that come from ungoverned data moving through tools nobody vetted, the training burden that falls on teams using five different platforms and the money going toward redundant software that nobody is coordinating.
The companies experiencing this are major ones. They are the majority and the costs show up the same way every time: rising spend with no clear return, teams moving in different directions and leadership eventually calling it a technology problem when the technology was never actually the issue.
What a Real AI Strategy Actually Starts With
The first question is not which AI tool to use. The first question is what you are actually trying to accomplish and where your company is genuinely losing ground. Where is the inefficiency? Where are the costs that do not need to be there? Where is the work that is slow or inconsistent or burning through resources without matching returns?
Those questions sound basic and most companies skip them entirely. They reach for the tool before they understand the problem, which is exactly how you end up with high token bills, duplicate work and a workforce confused about what they are supposed to be doing with the AI they were told to use. Many organizations that are already successful do not need a revolution. They need to find where they are leaking and apply AI to those specific places. The rest of the operation does not need to change.
Once you understand the pain points, here is what the foundation looks like:
- Designate the approved tools. Not every AI platform serves the same function and not every platform is appropriate for every type of company data. Leadership makes this decision, vets it against security and data governance requirements and communicates it clearly across the full organization. Not one department at a time.
- Build standard operating procedures for how AI integrates into existing workflows. Which tasks does AI support? What does the review process look like? Who carries final responsibility for the output? Those questions cannot be left for each employee to answer independently.
- Structure company information for AI use. Decide what institutional knowledge, internal documents and data should be available to the AI systems your teams use and how that content is organized and kept current. The output will only be as useful as the information behind it.
- Train employees to use approved tools with structure. High-performing organizations invest in both AI tools and employee skills training at the same time. Handing employees a tool without showing them how to use it is not a strategy but setting yourself up for AI sprawl.
- Monitor usage and cost on a consistent basis. If leadership is not tracking which tools are in use, by whom and at what expense, the budget will keep climbing without any return. What happened at Amazon was not surprising. It was the predictable result of incentivizing usage without managing it.
This Has to Start with Leadership
The framework has to come from the top; this is how AI adoption actually works inside an organization and every company that has pulled back a leaderboard, reversed a policy or started rehiring people it laid off because of AI has already proven it.
When leadership does not set the structure, employees fill the gap with individual decisions because the job needs to be done. Those individual decisions at scale produce fragmentation, not strategy and the organizations that are pulling ahead are not using more AI tools. They are using AI with intention, oversight and more clarity behind every decision, because someone at the top decided what the direction was before anyone opened a platform.
If your company has announced an AI initiative without establishing policies, designating approved tools, launching training or building a cost monitoring system, you have not started an AI strategy. You are just following a trend which will result in more losses than wins. Those companies are getting the best results and are ahead because they did the work that others declined to do.
Efficiency Has a Foundation
Leadership that actually understands its business does not adopt AI the way most companies have. It does not announce first and figure out the details later. It identifies where the operation is losing ground, decides where AI can close that gap and builds the rules before anyone opens a platform.
The leaderboards came down. The policies got reversed. The people got laid off and then rehired. None of that had to happen. It happened because the result was more appealing than the work and that is a choice companies keep making until the bill gets too high to ignore. The bottom line is most companies have AI tools. Very few have done the work to understand them.
Frequently Asked Questions
What is the difference between AI and AGI and why does it matter for business decisions?
The AI available right now works from patterns and needs human input, iteration and oversight at every stage. AGI would think on its own, develop its own strategies and operate without human direction. We are not at AGI. Companies making staffing and budget decisions as if we are will spend more money correcting those decisions than they would have spent doing the research first.
How do I know if my company has AI sprawl?
If different teams are using different AI platforms without central approval, if there are no written policies about what company information can be entered into AI systems or if leadership cannot clearly account for what tools are in use and at what cost, your company has AI sprawl.
Why does an AI strategy have to start with the business before the tools?
Because the tool cannot tell you what problem it is solving. If you do not know where your company is losing time, money or efficiency, you have no basis for evaluating whether any AI tool is actually helping. The strategy has to start with the business problem. Everything else follows from that.
I am an executive communications strategist with experience in government, media and corporate organizations. I write about AI, the workforce and what responsible communication looks like when technology moves faster than people are ready for.