The case for coaching humans with AI — sourced.

A 2026 survey roundup (Wrivio, Airia) puts it at roughly 65% of US employees using AI tools their employer never approved — often on personal accounts. Other 2026 surveys put shadow-AI users around half of all employees. Ban-first responses tend to move the use onto phones, where it is invisible to every tool a company administers.

Cyberhaven data-exfiltration research (via 2026 roundups) found about 11% of the data employees paste into ChatGPT-style tools is sensitive or confidential — source code, customer data, and internal documents are the most common types. Roundups also report roughly 27% of employees admit entering confidential company data into public AI tools (Airia, Second Talent).

The 2026 evidence says standalone training largely doesn't: Study.com's State of AI Jobs and Skills 2026 found 35% of employees have had no AI training at all, and 85% can't apply the training they did get to daily work. Formal AI upskilling was reported by only 26% of organizations in 2026, down from 35% a year earlier (Metaintro) even as tool spending rose. Coaching delivered inside the tool, at the moment of use, is the documented alternative that survives contact with the workday.

A Stanford Social Media Lab / BetterUp Labs study published in Harvard Business Review (Sept 2025) coined workslop: AI output that looks polished but lacks substance, shifting the burden of understanding back to the reader. 40% of 1,150 US full-time employees reported receiving it in the past month; researchers estimated about $9M/year in lost productivity for a 10,000-person company. It is, in our terms, a coaching failure — nobody taught the sender context, verification, or task–AI fit.

2026 industry analyses (Redress Compliance, Balanced+, Stackmatix) put active Microsoft 365 Copilot use around a third of purchased seats, naming lack of hands-on, in-tool training as the most common cause. At list prices ($21/user/month Copilot Business), a company paying for 100 seats gets real work out of roughly 35.

Bluevine's 2026 Small Business AI Trends report: 74% of small business owners use or are testing AI — and data security is the top barrier (33%, up 10 points year over year). A separate 2026 owner survey (via Under30CEO) found only 36% of owners whose staff use public chatbots have written rules for them.

Metaintro's 2026 survey summary reports roughly 44% of workers believe AI does more harm than good in their workplace — and survey roundups through 2026 show monitoring software deepens distrust when it is covert and one-way. The consistent signal: people accept being coached, resist being watched. That is the design line CoaPro follows — the rollout message in the product is literally “this isn't here to catch you.”

In the US, several states have electronic-monitoring notice statutes — Connecticut, Delaware, New York, and Colorado among them (New York requires written notice at hire plus a conspicuous posting). The EU AI Act treats AI used in employment as high-risk under Annex III, with deployer obligations live since August 2026. CoaPro builds the notice + attestation step into onboarding and explains the landscape in plain language on our monitoring-law page — that page explains, it does not give legal advice.

No — and our position is more specific than reassurance: AI doesn't replace humans; it makes humans so good with AI that using it without them doesn't make sense. The human provides context the model doesn't have, owns decisions that need judgment, and verifies before anything reaches a client. The research backs the economics: workslop (HBR 2025) costs more than it saves precisely because the human was out of the loop. We coach every employee — not just early adopters — because the company's output multiplier comes from the whole team's skill.

Every stat above is quoted with its named source, and the full link list is below, matching our claims discipline. These are third-party findings — none are presented as CoaPro's own data. Where a figure comes from a secondary roundup, we say so.