<- all tokdocs
2026.09.20@jenna_gardner_ai

The Research Behind This Argument Found 90% of Workers Already Use Personal AI Tools While Only 40% of Their Employers Pay for a Subscription

Career
Watch on TikTok

View on TikTok ->

The video's central split between individual-level AI and system-level AI is well supported by the enterprise research, but the research locates the individual-level problem somewhere other than where the video puts it. Jenna Gardner posted this 44-second clip on September 20, 2026. It is a single unbroken talking-head shot filmed from the driver's seat of a parked car, with a stop sign and a white pickup truck visible through the windshield and a hedge visible through the rear side window. One black-boxed title card sits at the top of the frame for the entire runtime and never changes: Mistakes to avoid in "AI Transformation". The only other on-screen text is TikTok's auto-generated word-by-word caption track at the bottom. There are no slides, no charts, no product demos, and no citations anywhere in the 22 sampled frames.

The two-track argument, stated precisely

Gardner makes four connected claims in 44 seconds. AI transformation has an individual-user track and a systems-and-workflow track. The two get conflated. Training people on ChatGPT, Claude, and Copilot does not address the system level, and systemic change is what makes AI benefits compound. Building workflow solutions without input from the teams affected produces something that does not get adopted, and she closes with the line "it's not truly a solution if it doesn't get adopted."

There is no number, no company example, no named framework, and no source in any of it. That is worth stating plainly because the format invites the assumption that a specialist is summarizing research. She is not citing anything here. The claims stand or fall on whether independent evidence supports them, so I went looking.

Where the evidence backs her: tool rollouts do not move the P&L

The MIT Project NANDA report The GenAI Divide: State of AI in Business 2025 (v0.1, July 2025, lead author Aditya Challapally) makes almost exactly her first point, in stronger language. Its executive summary reads: "Tools like ChatGPT and Copilot are widely adopted. Over 80 percent of organizations have explored or piloted them, and nearly 40 percent report deployment. But these tools primarily enhance individual productivity, not P&L performance."

That is her "training people on ChatGPT is not solving for the system level," restated with a sample behind it. The report is built on interviews with 52 organizations, survey responses from 153 senior leaders, and analysis of 300 public implementations.

BCG reaches the same conclusion from a different direction. Its January 15, 2025 report From Potential to Profit: Closing the AI Impact Gap, based on a survey of more than 1,800 executives, states the 10-20-70 rule directly: companies that succeed "dedicate 10% of their efforts to algorithms; 20% to data and technology; and 70% to people, processes, and cultural transformation." The same report puts it in one line: "Winning with AI is a sociological challenge as much as a technological one. The soft stuff -- reimagining workflows, upskilling talent, and driving organizational change -- turns out to be the hard stuff."

Her second point, that solutions built without input from affected teams do not get adopted, lands on the same 70%.

Where the evidence complicates her: the individual gap is access, not training

Gardner's framing of the individual track is "giving people trainings on ChatGPT and Claude and Copilot." The MIT data suggests the individual track has already solved itself without any training program.

The report's section on what it calls the shadow AI economy found that "while only 40% of companies say they purchased an official LLM subscription, workers from over 90% of the companies we surveyed reported regular use of personal AI tools for work tasks. In fact, almost every single person used an LLM in some form for their work." It goes further and says shadow AI "often delivers better ROI than formal initiatives."

So the individual-level deficit that shows up in the data is a licensing and sanctioning deficit rather than a skills deficit. Employees crossed the individual-adoption line on their own, on personal accounts, while procurement was still evaluating. A company that responds to this by commissioning a ChatGPT training series is addressing a gap that its staff closed months earlier on their own time.

BCG's AI at Work 2025 report, published June 26, 2025, from more than 10,600 respondents across 11 countries, adds the nuance that keeps training on the table. It found regular AI use among frontline employees "has stalled at 51%" while more than three-quarters of leaders and managers use generative AI several times a week, and that "only one-third of employees say that they have been properly trained." Training still matters at the frontline. Gardner's implied audience of knowledge workers who need a ChatGPT class is the group least likely to need one.

The 95% number this genre runs on is shakier than it looks

The MIT report is the source of the widely repeated claim that 95% of AI pilots fail. Gardner does not cite it, but this video sits inside the content genre that number built. Three things about it are worth recording.

First, the arithmetic. The report's funnel for task-specific enterprise GenAI tools is 60% of organizations evaluated them, 20% reached pilot, and 5% reached production. That is a one-in-three evaluate-to-pilot rate and a one-in-four pilot-to-production rate. Five percent reaching production out of the 60% that evaluated is an 8.3% success rate among organizations that actually tried, not a 5% success rate. The executive summary's separate headline, "95% of organizations are getting zero return," describes a different population from the 5% production figure, and popular coverage routinely merges the two into one statistic.

Second, the report attaches its own caveat, which almost never travels with the number: "Research Limitations: These figures are directionally accurate based on individual interviews rather than official company reporting. Sample sizes vary by category, and success definitions may differ across organizations."

Third, the primary link is gone. The original URL, https://nanda.media.mit.edu/ai_report_2025.pdf, now returns an HTTP 302 redirect to the NANDA group's overview page, and that overview page makes no mention of the report, the GenAI Divide, or the 95% figure. I verified the redirect directly. The copy I read and quote from here is a third-party mirror whose PDF metadata matches the original, created July 13, 2025 by Aditya Challapally.

For scale comparison, BCG's January 2025 survey found roughly one-quarter of executives reporting significant value from AI. MIT's July 2025 number is 5%. Those two figures are five times apart and both circulate as though they measure the same thing.

What the video leaves out

The word "compound" carries the whole argument and never gets defined. Gardner says systemic solutions are required "for the AI benefits to actually compound," which asserts a mechanism without naming it. Compounding across what, on what timescale, measured how, all go unaddressed.

The video also gives no method. It identifies a conflation and a failure mode without offering a way to tell which track a given initiative belongs to, how to sequence the two, or what team input actually looks like in practice. For a 44-second clip that is a fair constraint rather than a flaw, and the framing is genuinely useful as a diagnostic. It is not an implementation plan, and the confident delivery may read as one.

One more omission is the money. Every source I checked frames this as a spending problem. MIT puts enterprise GenAI investment at $30 to $40 billion against that near-zero measured return. The video treats the two tracks as a conceptual error rather than a budget allocation question, which is where the decision actually gets made.

Key Takeaways

  • The video's core claim, that individual-level tool training and system-level workflow change are distinct problems and that the first does not produce the second, is directly supported by MIT NANDA's finding that ChatGPT and Copilot "primarily enhance individual productivity, not P&L performance."
  • BCG's 10-20-70 rule, published January 2025 from a survey of 1,800-plus executives, allocates 70% of AI effort to people, processes, and cultural change, which is the strongest quantified backing for Gardner's adoption argument.
  • The research locates the individual-level gap differently than the video does. MIT found 90% of surveyed companies' workers already use personal AI tools regularly while only 40% of those companies bought a subscription, making it an access problem more than a training problem.
  • The 95% failure figure underpinning this content genre has a live arithmetic ambiguity, carries an explicit "directionally accurate" caveat from its own authors, and its original MIT-hosted PDF URL now redirects to a group page that does not mention the report.
  • Whisper transcribed the audio accurately with one substantive error. It rendered "Claude" as "Cloud" in the phrase "trainings on ChatGPT and Cloud, Copilot," which changes a named product into a generic infrastructure term. The transcript runs the full 43.6 seconds of a 44-second video and does not cut off. It captures none of the on-screen content, so the title card Mistakes to avoid in "AI Transformation" and the car setting are invisible to anyone reading the transcript alone.
  • Unverified: Gardner's professional role. Search results consistently list a Jenna Gardner as Director of Applied AI, Media at DEPT, which matches the account's #applied hashtag, but LinkedIn and the TikTok profile page both blocked automated retrieval, so I could not confirm the identity from a page I actually loaded. Also unverified: McKinsey's widely cited finding that workflow redesign correlates most strongly with EBIT impact from AI. Every mckinsey.com request from this environment timed out, so I am not citing or leaning on it. The MIT report's per-question sample sizes are not broken out in the document, so the 90% and 40% shadow AI figures cannot be checked against a stated denominator.

Resources

Published September 20, 2026. Writeup generated from a favorited TikTok.