G2 2026 AI Agents Insight Report: The Distrust Tax
66% run agents in production while 45% lose more than a quarter of expected gains to oversight and rework: the Distrust Tax. Survey of 1,019 B2B decision makers.
G2 · 2026
22 of 22 items · newest filed first
66% run agents in production while 45% lose more than a quarter of expected gains to oversight and rework: the Distrust Tax. Survey of 1,019 B2B decision makers.
G2 · 2026
74% of IT leaders expect to be held accountable for agents they do not fully control; 62% say most agents in use were never formally approved by IT. Survey of 1,001 US IT decision makers.
Workato · May 2026
In a controlled experiment, workers paid a female presenting AI agent 10.25 percent less than an identical male presenting one, so agent persona design (name, voice, avatar) is not decoration: it changes how the agent's work is valued.
Proceedings of NordiCHI '26 (ACM) · Oct 2026
In firms that adopt generative AI, the junior share of the workforce falls mainly because senior hiring grows, not because junior jobs are cut.
SSRN · 20 Sept 2026
34% of GRC leaders already authorise agents to modify controls or policies, while only 9% have redesigned workflows around agents.
Optro · 2026
Only about 34% of agentic projects reach production; data fragmentation is cited by 55%, and production leaders rate far stronger on semantic knowledge. Briefing attached: Knowledge Before Agents.
MIT Technology Review Insights · 2026
A field guide turning 37 AI transformation failure patterns into say, change, build responses, with a triage tool.
Field guide · 25 Sept 2026
Enterprises measure operational impact from AI, not the business outcomes they say they want: 59% expect staff productivity gains, 50% cost reduction, 46% profit growth.
ISG · Sept 2026
Build ROI at workflow level with baselines; do not use output volume as a proxy for value. Brief paired with the ISG report.
Section · 2026
Nine field tested techniques: a skill is a harness extension that must work across models, harnesses, and prompting styles. Pairs with The Case Against Skills.
X (article) · 14 Jul 2026
SWE-Skills-Bench tested 49 public skills: 39 did not change performance, 3 made it worse, token use rose up to 451%. Closes with a keep, retest, retire audit.
Every / Context Window · 16 Jul 2026
Peer reviewed protocol: prompts as an active process for bridging new AI output with existing knowledge, aimed at managers and teams.
Business Horizons 67 (2024) · 2024
Technology acceptance extended with subjective norms, compatibility, facilitating conditions, and trust, plus an assessment tool. An OCM instrument.
Business Horizons 67 (2024) · 2024
Delegation versus pairing: only about half of knowledge work is allocation work; the rest needs an agent working alongside, from a shared operating manual.
Every / Context Window · 5 May 2026
A Goal as a scoped completion contract with auditable success checks reframes agent work from bigger prompts to persistent objectives.
OpenAI Cookbook · 1 Jun 2026
The first large scale payroll evidence that generative AI is cutting entry level hiring in exposed jobs, which changes the workforce pipeline every AI adoption plan depends on.
Stanford Digital Economy Lab · 26 Aug 2025
Org charts freeze authority while agents unbundle roles into tasks. Map the work instead: trigger to steps to outcome, each step labelled human, agent, or hybrid.
Slide deck · 2025
31,000 workers in 31 countries: the Frontier Firm model, an 82% of leaders calling 2025 pivotal, 81% expecting agents integrated within 12 to 18 months. Vendor framing; still the reference vocabulary in Copilot client work.
Microsoft · 2025
Whisper based tools in 40 health systems: hallucinations found in 8 of 10 transcripts in one review, nearly 40% of flagged hallucinations judged harmful. The governance lesson has not expired.
AP News · 25 Nov 2024
1,005 US knowledge workers aged 22 to 39: 93% of Gen Z respondents used two or more AI tools weekly and three in four had recommended tools to peers. Brief paired with newer generational data.
Entrepreneur · 25 Nov 2024
Durable institutional questions: an AI advisory group, a values statement, procurement language preserving the right to turn vendor AI features off, and review of vendor training data terms.
1EdTech · 2024
The baseline snapshot of how higher education first organised around AI: responsibilities handed out almost everywhere, formal roles almost nowhere.
EDUCAUSE · 8 Dec 2023