Are large companies ready to let AI agents act without people?
In a TCS survey of 1,333 senior leaders, 10% said agents could execute decisions without human approval across several functions and 9% said agents securely orchestrated work across applications. The sample covers large organisations and reports perceptions, not audited deployments or productivity.
The short version
- What happened
- The company survey covers 1,333 senior leaders in 25 countries and 12 industries; 31% were CxOs, 14% unit or divisional heads and 55% senior vice presidents or vice presidents.
- Who it affects
- Workers should not read the 84% figure as evidence that AI has raised their own output, pay or job quality; the study did not measure those outcomes.
- What remains unproven
- The study was produced by TCS, which sells AI consulting and technology services, and was not independently peer reviewed.
Keep in mind
The study was produced by TCS, which sells AI consulting and technology services, and was not independently peer reviewed.
Respondents were senior leaders at organisations with at least $500m in annual revenue; workers, small employers and customers were not surveyed.
The public report does not disclose recruitment, response rate, questionnaire wording, fieldwork dates, weighting, missing responses or confidence intervals.
Productivity, autonomy, security and resilience are self-reported rather than verified against operational logs, incidents or financial outcomes.
Regional and industry sample sizes are disclosed, but subgroup findings are not accompanied by uncertainty estimates or full cross-tabulations.
Multi-source analysis · 2 linked sources. Primary analysed source date: 4 November 2026.
Keep following the evidence
Join The Impact Brief founding list for selected AI reports, their effect on people and the evidence limits. Read a sample first.
Read the full report · 10 min The evidence, methods, implications and limitations
The direct answer: most surveyed leaders describe assistance, not broad autonomous action
The strongest answer in TCS’s new enterprise study is caution. Ten per cent of the 1,333 senior leaders surveyed said AI agents could execute decisions without human approval across multiple functions or business areas. Nine per cent said their agents could securely orchestrate workflows across enterprise applications using governed tools and permissions. Those are not identical measures, but both put broad, controlled autonomy in a small minority of the large organisations represented.
That matters because the term ‘agent’ is often used for systems with very different authority. A tool that drafts an email for a person to approve is not equivalent to one that changes a customer account, approves a payment or moves data between protected systems. The survey suggests many organisations remain closer to assistance than delegation. It does not prove that 90% have no agents, or that more autonomy would necessarily be desirable. It shows that few respondents reported the combination of reach, decision authority and secure orchestration described in the questions TCS summarises.[1][2]
Who was surveyed—and who was not
The report gives a useful denominator and unusually detailed sample map. It covers 1,333 senior business and technology leaders across 25 countries and 12 industries. North America supplied 326 respondents, including 271 in the United States and 55 in Canada. Continental Europe supplied 416; the United Kingdom and Ireland 204; Asia-Pacific 173; Latin America 153; and Australia and New Zealand 61. Country counts are disclosed within those regions, so the ‘global’ label can be assessed rather than taken on trust.
The sample is deliberately senior: 31% were CxOs, 14% unit or divisional heads and 55% senior vice presidents or vice presidents. It is also restricted to large organisations. Twenty per cent worked for businesses with at least $20bn in annual revenue, 34% for those between $5bn and $20bn, 33% between $1bn and $5bn, and 13% between $500m and $1bn. Small and medium-sized employers, frontline workers, contractors, unions and customers were not the people answering. The results therefore describe leadership perceptions in large enterprises, not the experience of the workforce or economy as a whole.[1]
The distinction that matters
What the survey percentages do—and do not—show
Each figure reports a leadership response in a vendor survey of large organisations. None is a direct measurement of company-wide productivity or system safety.
84%
Rank productivity as the leading AI value
Perception or priority; no measured productivity denominator
10%
Report agents deciding without human approval across several areas
Broad autonomy reported by a minority
9%
Report secure orchestration across enterprise applications
Security capability was self-reported, not audited
7%
Report an organisation-wide provider-switching plan
Plan existence was reported; resilience was not tested
The industry denominator is broad, but it does not make every sector estimate precise
Manufacturing was the largest industry group with 203 respondents, followed by banks and financial services with 150, retail with 133 and technology with 120. Travel, hospitality and logistics contributed 112; communications, media and information services 110; consumer packaged goods 97; life sciences 91; healthcare 83; utilities 81; energy and resources 80; and insurance 73. The total is 1,333, which lets readers see how the overall finding was constructed.
Those cells are large enough to show that the study is not simply a technology-company poll, but many subgroup estimates would still carry substantial uncertainty. TCS does not publish the questionnaire, item-level denominators, confidence intervals or a table showing whether every respondent answered every question. It also does not disclose how organisations or individuals were recruited, the response rate, fieldwork dates, weighting or whether one person represented each employer. Without those details, a difference between regions or sectors may reflect sampling and question interpretation as well as a real difference in adoption.[1]
‘Productivity is the top benefit’ is a ranking, not a productivity measurement
TCS says 84% of executives ranked productivity improvement as the leading source of value from AI investments. That is evidence about what leaders believe or choose as a priority. It is not a time-and-motion study, a randomised comparison or a financial audit. The report does not give hours saved, output per worker, error rates, service quality, employee workload or the distribution of any gain. It also says 10% reported revenue uplift, but the public material supplies no amount, baseline, attribution method or comparison group.
For workers, this distinction is central. A manager may experience productivity as faster drafting while an employee experiences more tasks, tighter monitoring or additional checking. A system can shorten one stage and create review work elsewhere. Reliable evidence would track the full process: time, quality, rework, safety, staffing, pay and customer outcomes before and after deployment, ideally against a credible comparator. The survey does not do that. Its 84% figure should not be translated into ‘84% of companies became more productive’ or used to infer a particular economic gain.[1][2]
Leaders want model choice, but few report a company-wide switching plan
Ninety per cent of respondents said model independence mattered to AI investment decisions, while only 7% reported an organisation-wide provider-switching continuity plan. Thirty-four per cent rated sovereignty requirements involving data location, model choice and compliance as a high or very high organisational risk; TCS reports 43% for Europe. The gap is informative: many leaders value flexibility, but comparatively few say their whole organisation can switch providers if a model, service, price or regulatory condition changes.
Switching is not just selecting another model from a menu. Applications can depend on a provider’s prompts, safety filters, identity system, retrieval tools, data formats, evaluation thresholds and commercial terms. A continuity plan has to identify critical workflows, acceptable fallback performance, data-transfer constraints, permissions, testing and the people authorised to invoke it. The survey does not test whether reported plans work under disruption. Still, the 90%-versus-7% contrast gives procurement teams a practical question: is independence a stated preference, or a rehearsed operational capability?[1][2]
Workforce uncertainty has increased, while skills systems lag
The share of leaders saying it was too early to determine AI’s effect on jobs rose from 4% in TCS’s 2024 study to 19% in 2026. That comparison suggests experience has not produced a simple employment forecast. It may reflect faster changes in tools, uncertainty about which tasks will be automated, or the difference between pilot activity and durable operating-model change. Because the report does not publish matched organisations, question wording or a statistical test, the two percentages should be treated as separate survey snapshots rather than a panel tracking the same employers.
The report also says only 28% of respondents maintain an enterprise framework or evolving skills taxonomy. That is more concrete than a generic promise to reskill: an organisation needs to know which tasks are changing, which capabilities are required and whether training transfers into safer or better work. For employees, the absence of that structure can mean that expectations change before roles, evaluation or learning support do. For employers, it can leave deployment dependent on a few specialists and informal knowledge. The study does not measure training access, completion, proficiency, career progression or whether affected workers helped design the change.[1][2]
Knowledge and cost constraints complicate agent deployment
Forty-four per cent of respondents said important business knowledge remained embedded in employees or fragmented documents. Thirty-nine per cent were still assessing whether AI initiatives generated enough value relative to token and compute costs, and 46% reported targeted use of open-source AI for selected cases. Together, those findings describe a deployment problem that is organisational as much as technical. An agent cannot reliably follow a process that has no agreed record, and extracting tacit knowledge can expose sensitive information or remove context that experienced staff use to handle exceptions.
Documenting work can improve continuity, but it should not become an exercise in copying every employee interaction into a model. Organisations need permission boundaries, retention rules, source provenance, quality checks and a route for workers to correct inaccurate procedures. Cost assessment also needs the whole denominator: model use, integration, evaluation, human review, incidents, training and maintenance. TCS does not provide project-level cost or benefit data. The survey indicates that economic scrutiny is active; it does not establish which architectures or vendors deliver better value.[1][2]
Evidence check, commercial interest and what would change the assessment
The full report is a current, primary company study with transparent totals for geography, industry, company revenue and respondent seniority. Those denominators make it stronger than an announcement built around an unexplained percentage. It remains vendor-produced research. TCS sells AI, consulting and implementation services and presents a ‘Human+AI’ operating model as part of its interpretation. The report authors and editorial team are TCS personnel. The public report does not disclose an independent research partner, external peer review, survey instrument, recruitment method, response rate, weighting, fieldwork dates or uncertainty intervals.
Confidence would rise if TCS released the questionnaire, anonymised response tables, recruitment and weighting procedures, missing-response counts and confidence intervals, with employer and respondent identifiers protected. Independent surveys of workers and operational audits could test whether reported autonomy, security and productivity correspond to system logs and outcomes. Longitudinal evidence should track the same organisations through deployment and compare task quality, review burden, incidents, wages and job transitions. Until then, the defensible conclusion is narrow: among surveyed leaders at large organisations, broad autonomous action and secure cross-application orchestration were uncommon, even as most ranked productivity as AI’s leading value. That is consequential current research, but not an urgent breaking event.[1][2]
Watch the background, then check the evidence
Illustrative photo · not a video frame
UNESCO · 11 min 27 sec · Background explainer
UNESCO Talks: Interview with 2024 Nobel laureate Professor Daron Acemoglu
YouTube loads only when you choose; playback starts when you press play.
Economist Daron Acemoglu discusses how choices about AI affect workers, inequality and society. Pair the interview with our comparison of task exposure, hiring and skills evidence.
Evidence check: An expert interview from 2024, offering a perspective on policy and work rather than a measured forecast of job losses.
Published 2024-10-14. Source checked 14 October 2026. Check captions on the publisher’s video page.
Read the AI and jobs guidePhoto: Dylan Gillis / Unsplash · Licence
Continue the story
Read next
Continue exploring ai agents: compare the finding and its evidence limits.
OECD interviews show how early AI agents are being kept within human checkpoints
Practitioners in 25 organisations across 11 countries describe agents moving into real workflows, with autonomy limited around consequential actions. The interview study is a window into practice, not a global adoption rate.
Read the next reportExplore the full source and evidence record
What this means for people
- Workers should not read the 84% figure as evidence that AI has raised their own output, pay or job quality; the study did not measure those outcomes.
- Broad agent autonomy was reported by a minority, leaving human approval, escalation and accountability central to most deployments represented.
- Only 28% reported an enterprise skills framework or evolving taxonomy, suggesting workforce systems may lag behind changing tools and tasks.
Global context
The sample spans 25 countries but is concentrated in large organisations and six regional groupings. Continental Europe contributes 416 responses, North America 326, the UK and Ireland 204, Asia-Pacific 173, Latin America 153, and Australia and New Zealand 61. Differences in regulation, infrastructure, labour institutions and enterprise size mean the overall result should not be projected to smaller employers or countries outside the sample.
What the evidence does not yet show
- The study was produced by TCS, which sells AI consulting and technology services, and was not independently peer reviewed.
- Respondents were senior leaders at organisations with at least $500m in annual revenue; workers, small employers and customers were not surveyed.
- The public report does not disclose recruitment, response rate, questionnaire wording, fieldwork dates, weighting, missing responses or confidence intervals.
- Productivity, autonomy, security and resilience are self-reported rather than verified against operational logs, incidents or financial outcomes.
- Regional and industry sample sizes are disclosed, but subgroup findings are not accompanied by uncertainty estimates or full cross-tabulations.
What to watch next
- Publication of the questionnaire, recruitment method, response rate, fieldwork dates and weighting
- Independent audits comparing leadership reports with system logs, security tests and measured work outcomes
- Worker-level evidence on review burden, job quality, training access, pay and task change
- Tests of whether provider-switching plans work during a real or simulated outage
- Longitudinal results showing whether agent autonomy expands without increasing errors, incidents or unaccountable decisions
Living evidence record
Impact record IAI-1RTYBW4
Evidence stage
Observed
Confidence
Supported
Reporting basis
Multi-source analysis
Independent or research support
Not yet
Record status
Monitoring
Last checked
4 November 2026
Source trail
2 direct sources across 2 source types.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Evidence trail
Sources used for this report
Links checked 4 November 2026
Research question, topics and regions
How senior leaders in large organisations describe enterprise AI value, autonomy, resilience and workforce readiness
Topics: AI agents · Workplace automation · Human oversight · Enterprise governance · AI skills
Regions: Global · Europe · North America · Asia-Pacific · Latin America
This report is labelled multi-source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Reader commentary
What does this mean where you are?
Tell us what you have seen at work, in your community or in everyday life. A question is welcome too. No account is needed; we review notes before publishing them.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.