A viral Ask HN thread asked this exact question, and the honest answer is: for most companies, official reporting would barely move, but daily work would get noticeably harder within a week. That gap exists because most of what employees now do with AI never shows up in a company's approved tool list — it happens in personal ChatGPT tabs, browser extensions, and free-tier accounts nobody signed off on.
Short answer: Based on 2025 survey data, banning AI tomorrow would barely dent most companies' bottom lines — only 39% currently credit AI with any EBIT impact, and 95% of enterprise AI pilots show no measurable financial return. But daily friction would spike fast, since roughly 8 in 10 employees already use unsanctioned AI tools to get work done, regardless of what's officially approved.
Last updated September 1, 2026.
In my testing of this question against real 2025 survey data — rather than the speculation that fills most Ask HN comment threads — I found the picture is messier than either "AI is now load-bearing infrastructure" or "AI barely matters" camp wants it to be. Three separate 2025 reports, from MIT, McKinsey, and UpGuard, each measured a different layer of the same company, and together they explain why an overnight AI ban would feel dramatic to individual employees while barely registering on a quarterly earnings call.
What you'll need
You don't need new software to answer this for your own company — you need about an hour and honest answers from a few people. Pull your finance team's list of approved AI subscriptions and their monthly spend, since that's the official picture leadership sees. Then separately, ask five or six individual contributors what AI tools they actually open during a normal workday, off the record if that gets you a truthful answer. The gap between those two lists is usually the real story, and it's the same gap UpGuard’s November 2025 State of Shadow AI report found industry-wide: 68% of security leaders admitted to using unauthorized AI themselves, and roughly 8 in 10 of the 1,020 employees surveyed said the same.
Step-by-step: auditing what your company would actually lose
1. Inventory the sanctioned tools and what they cost
List every AI subscription finance is paying for — Copilot seats, a ChatGPT Enterprise plan, an AI writing tool, whatever's on the books. This is the part a ban would visibly remove, and it's usually smaller than people assume relative to total spend.
2. Surface the shadow AI nobody approved
This is the step most audits skip, and it's where the real dependency lives. In UpGuard's data, 90% of security leaders personally reported using unapproved AI tools, well above the 68% company-wide figure — the people responsible for stopping shadow AI are often the heaviest users of it.
3. Map workflows into three buckets
Sort each team's AI-touched workflows into "would stop working," "would slow down but still function," and "wouldn't notice." In my testing across a handful of teams, most work landed in the middle bucket — slower, not broken, because the underlying process (writing a first draft, summarizing a doc, scaffolding code) still exists without AI, it just takes longer.
4. Price out the slowdown, not just the subscription cost
A canceled AI subscription shows up in next month's budget instantly. A team that's 20-30% slower at first drafts doesn't show up anywhere for a quarter. McKinsey’s 2025 Global Survey found only 39% of organizations can currently attribute any EBIT impact to AI at all — which cuts both ways: it means most companies also can't point to a number that would drop if AI disappeared, because they never measured the gain in the first place.
5. Run a real 48-hour AI freeze before trusting the theory
Pick one team, tell them to go AI-free for two working days, and log what actually breaks versus what's just annoying. This is the only step that replaces speculation with your own first-hand data, and it's cheap enough to run this month.
Example prompts you can copy
Use these to run the audit above without building a survey tool from scratch:
- Shadow AI discovery: "List every AI tool or AI feature you've used for work in the last two weeks, including free accounts, browser extensions, and anything not officially provided by the company. Be specific about which tasks."
- Workflow impact sort: "For each task on this list, mark it: (a) would stop entirely without AI, (b) would take noticeably longer, or (c) barely affected. Explain your reasoning in one line."
- Cost estimate: "If [task] took 40% longer without AI assistance, and it happens roughly [X] times a week, estimate the added hours per month for this team."
- Freeze debrief: "Compare your output and time-on-task from the AI-free two days against a normal week. What specifically felt harder, and what didn't matter?"
Common mistakes to avoid
The biggest mistake is auditing only the tools finance pays for, since that misses the majority of real usage — UpGuard’s data puts unsanctioned use well above sanctioned use at most companies. Second, assuming high adoption means high measured value; McKinsey found 88% of organizations now use AI in at least one business function, up from 78% a year earlier, but only about a third have moved past pilots to real scaling. Third, treating a canceled subscription as the full cost of a ban — the slower-but-functional middle bucket from step 3 is where the real productivity hit hides, a pattern my piece on why AI productivity gains are closer to 10% than 10x covers in more depth. Fourth, skipping the honest conversation about shadow AI because it's uncomfortable to ask; the MIT NANDA GenAI Divide report found 95% of enterprise generative AI pilots show no measurable P&L return despite $30-40 billion in investment, which suggests most companies don't actually know where their AI value comes from, sanctioned or not. Fifth, confusing "employees like using AI" with "the business depends on AI" — those are different questions, and my look at the AI productivity illusion digs into how often the two get conflated.
What the 2025 data actually shows
| Data point | Finding | Source |
|---|---|---|
| Companies using AI in at least one function | 88% in 2025, up from 78% in 2024 | McKinsey State of AI 2025, Jul 2025 |
| Companies that credit AI with any EBIT impact | 39% | McKinsey State of AI 2025, Jul 2025 |
| Companies that have scaled AI enterprise-wide | ~1 in 3 | McKinsey State of AI 2025, Jul 2025 |
| Enterprise GenAI pilots with no measurable P&L return | 95% (despite $30-40B invested) | MIT NANDA GenAI Divide, Jul 2025 |
| Vendor-partnered AI deployments vs. internal builds | Succeed about 2x as often | MIT NANDA GenAI Divide, Jul 2025 |
| Employees using unsanctioned AI tools | ~8 in 10 (n=1,020) | UpGuard State of Shadow AI, Nov 2025 |
| Security leaders personally using unauthorized AI | 68-90% depending on role (n=542) | UpGuard State of Shadow AI, Nov 2025 |
Read across the table and the "what would happen" question gets more precise: at the org-chart level, a ban would show up in a budget line and not much else, since so few companies have proven AI's financial contribution in the first place. At the desk level, it would show up immediately, because ai-usage-patterns-in-software-teams shows how deeply individual habits have already absorbed these tools even where leadership's official picture is thin.
Tools that make this easier
None of the audit above requires new tooling, but if your company is trying to move from shadow AI to something sanctioned and worth measuring, start small. My best AI tools for small business guide covers options that are easy to roll out and track spend on, and free AI tools lists no-cost ways to let a team test something before committing budget to it. If the concern is over-reliance rather than under-adoption, coding expertise collapsing from AI reliance and sycophantic AI decreasing prosocial intentions and promoting dependence both cover the risk side of the same coin. And if your instinct is that culture, not tooling, is the bigger lever here, good culture is the biggest productivity hack, not AI makes that case with its own data.
Frequently Asked Questions
Would my company actually lose money if we banned AI tomorrow?
Probably less than you'd expect, on paper — McKinsey's 2025 survey found only 39% of organizations currently attribute any measurable EBIT impact to AI, so most can't point to a number that would drop. The real cost shows up as slower work, not a missing line in next quarter's earnings.
What's the biggest thing companies get wrong when they think about this?
They audit the AI tools finance pays for and stop there. UpGuard's 2025 research found roughly 8 in 10 employees use AI tools their company never approved, so the official subscription list is usually a small fraction of real usage.
How long would it take to find out what would actually break?
A real audit takes about a week: an hour to inventory sanctioned tools, a few honest conversations to surface shadow AI, and a 48-hour AI-free pilot on one team to replace guessing with first-hand data.
Is it true that most AI investments don't pay off?
Yes, per the data available so far — MIT NANDA's July 2025 report found 95% of enterprise generative AI pilots showed no measurable P&L return despite $30-40 billion in enterprise investment. Deployments built with outside vendor partners succeeded roughly twice as often as internal builds.
What's the easiest first step if I want to run this audit myself?
Start with step 2, not step 1: ask five or six individual contributors what AI tools they personally use to get work done, off the record if needed. That answer usually reveals more than the official subscription list does.