Tech AI News - 테카이
환호하는 관객 앞의 로봇 뒤에서 조용히 전선을 고치고 있는 직원

Why AI adoption fails — Responsibility without authority, invisible work

· 5 min read

This post was translated from the Korean original by AI.한국어 원문 읽기 →

Multiple studies published in 2026 reach a similar conclusion: the roadblock to AI adoption isn’t technology, it’s the organization. Two failure mechanisms keep recurring: “responsibility assigned without authority granted,” and “no one recognizes the work that makes AI usable.”


At a glance

Study/reportSample / timingFocusKey findings
MIT Sloan Management Review columnField case analysis (released 2026.4)Psychological barriers on the front linesFear, perception of extra workload, and lack of perceived value drive resistance
Prosci State of AI Adoption 2026Survey of enterprise AI adoption leadsChange management maturity35% of managers feel prepared for the change manager role; people factors account for 56–64% of adoption challenges
Glean Work AI Institute, Work AI Index 20266,000 digital workers in the US/UK/AU (surveyed 2025.12–2026.1)Invisible work6.4 hours/week spent on “botsitting”; 87% use AI, but only 13% perceive organizational performance gains
BCG AI at Work 202611,749 people in 14 countries (released 2026.6)Strategy, training, leadershipClear strategy lifts outcomes by 25 percentage points, while tool improvements add only 5 points; 36% say training is sufficient
Large-scale randomized experiment in HBRCorporate experimental groupsResponsibility designFraming AI agents as “employees” weakens individual accountability, increases unnecessary escalation, and degrades review quality

Background — Why do the same failures repeat?

Performance measurements of generative AI pilots have already yielded pessimistic figures multiple times. According to Fortune (August 2025) reporting on the MIT Media Lab NANDA project, about 95% of enterprise generative AI pilots failed to deliver measurable revenue or cost outcomes. Reuters (June 2025), reporting on a Gartner outlook, projected that over 40% of agentic AI projects would be canceled by the end of 2027.

Yet usage surged over the same period. In Glean’s Work AI Index 2026, 87% of surveyed digital workers said they use AI for work, but only 13% said they noticed clear organizational performance improvement. People factors explain this gap where adoption is complete but results don’t follow.

Failure mechanism 1 — Assigning responsibility without authority

The US Chamber of Commerce notes that the value of a role is determined by three questions: who has execution authority, who is accountable for outcomes, and who shares in the gains. If you hand people AI tools but keep approval flows and accountability boundaries unchanged, employees end up “knowing more” rather than “being able to do more.”

A large-scale randomized experiment featured in HBR quantifies this issue. When AI agents were given titles and positions to be treated like “employees,” individuals’ sense of accountability weakened, unnecessary escalation to higher-ups increased, and review quality declined. Willingness to adopt did not increase. The research team concludes that instead of positioning AI as a person, organizations should redesign workflows, roles, and governance to clarify human accountability.

BCG’s 2026 study (11,749 people in 14 countries) shows a similar pattern. In more than half of the surveyed companies, leadership set direction while ownership of results shifted to individuals or teams, and about two-thirds of leaders said they had significantly reduced procedural approval and review steps. Shifting responsibility without transferring authority tends to result in either burnout or blame avoidance.

Failure mechanism 2 — No one acknowledges the invisible work

Glean’s Work AI Index 2026 introduces the term “botsitting.” It includes supplying missing context, validating and editing outputs, debugging errors, and shuttling between disconnected AI tools. Respondents said AI saved them about 11 hours per week, but they spent 6.4 of those hours back on botsitting. Moreover, 69% admitted to sending out AI outputs without sufficient validation—likely a consequence of skipping botsitting.

In the BCG study, 47% of regular AI users said they spend more time managing and directing AI than on the work itself, and 41% reported increased cognitive load. None of this shows up in official job descriptions or evaluation metrics. From the organization’s perspective, productivity seems up, but in reality, invisible work is filling the gap.

Analysis: The two mechanisms share the same root. If an organization treats AI adoption as “handing out tools,” it skips two tasks: redesigning authority and re-recognizing labor. You end up with people who are accountable for outcomes but lack the authority to change processes, alongside unrecognized labor propping up results.

What organizations that succeed do differently

According to Prosci’s State of AI Adoption 2026, people factors account for 56–64% of AI adoption challenges, yet only 35% of organizations say their managers are adequately prepared to serve as change managers. The same survey found that lack of training accounts for 38% of adoption challenges.

BCG argues that strategic clarity makes a far bigger difference than better tools. Organizations with a clear strategy saw outcomes improve by 25 percentage points, while tool improvements alone added 5 points. Yet only 36% felt training was sufficient, and roughly one-third of frontline respondents felt leadership’s AI communications were clear. Just 28% felt words and actions were aligned.

The MIT Sloan Management Review column (original, 2026.4) uses late-adopting industries like construction and mining to suggest viewing frontline skepticism not as resistance but as a rational response. It proved easier to gain acceptance by demonstrating results with existing metrics—such as win rate or reduced customer complaints—rather than inventing new ones, and by layering AI onto tools already in use rather than introducing entirely new systems.

Key success factors

  1. Reallocate decision rights — When shifting responsibility, also transfer the approval and modification authority needed to influence outcomes. Responsibility without authority erodes execution intent.
  2. Formalize invisible work — Reflect validation, editing, and context-supplementing tasks in job descriptions and evaluation metrics. Unrecognized labor is unsustainable.
  3. Bridge strategy and execution — Close the gap between what leaders say and what the front line experiences as execution, with managers as the linchpin.

Limits and open questions

These studies differ in sample composition (country, industry, level) and in how they define “people factors” and “hidden work.” They show correlations rather than proving that specific actions cause outcome lifts, so treat the figures as directional.

Applying this to our organization

Large enterprises may prioritize building managers’ change management capabilities (redesigning approval authority, incorporating metrics), while SMBs should check workload distribution to prevent invisible work from concentrating on a few people. Startups, with fewer approval steps, face less of an authority problem, but should beware of ambiguous roles that let invisible work quietly become “just someone’s job.”

References

  • #ai
  • #ai adoption
  • #change management
  • #leadership
  • #organizational design
  • #hidden work
  • #human factors