AI agents in the human workplace
B · Useful, limited Scenario essay from Deloitte's Center for Integrated Research and Human Capital practice that recycles survey figures from other Deloitte studies and anonymous executive interviews, with no new dataset of its own.
Summary based on the full text·Original: about 10 minutes to read
In short
David Mallon, Brad Kreit and Natasha Buckley of Deloitte (published 10 September 2026) build four scenarios for how AI agents could sit inside organizations by 2030. The scenarios cross two uncertainties: whether firms treat agents as labor or as infrastructure, and whether the gains spread broadly or concentrate among a few. The conclusion is that four priorities hold in every scenario: governance, strategy, workforce planning and trust. The authors argue that spending on these adjacencies matters as much as spending on the technology.
Key numbers
- 93% / 7%share of organizational AI investment going to technology infrastructure versus work and people issues (Deloitte study)
- 3.8xhow much more likely workers with high trust in their organization are to use agentic technology (Deloitte TrustID)
- 64% vs 5%respondents who call decision-making very important to AI success versus those reporting great progress on it
- ~70%respondents naming an inability to trust and govern agents as a major barrier to value
As reported by the source.
English·Web page·opens deloitte.com
What it reports
- Four futures: "colleague machine" (labor, shared gains), "output engine" (infrastructure, shared gains), "hollowed middle" (infrastructure, few winners), "first-mover future" (labor, few winners). Each comes with signals to watch, for example junior roles disappearing faster than senior roles are redesigned.
- Cites a Gartner forecast that the average Fortune 500 company may run 150,000 AI agents by 2028.
- Two-thirds of senior managers expect multi-agent AI to significantly change how work is done within two to three years; about a quarter say their processes or workforce are ready.
- Nearly three-quarters of organizations in Deloitte's 2026 State of AI in the Enterprise survey plan at least moderate use of agentic AI within two years.
- 93% of organizational AI investment goes to technology infrastructure, 7% to work and people issues.
- Nearly 70% name the inability to trust and govern agents as a major barrier; 64% say decision-making is very important to AI success, 5% report great progress.
- 53% of respondents in a Harvard Business Review study admitted sending colleagues AI-generated work that was "unhelpful, low effort, or low quality" because they were overloaded.
- In an informal Deloitte webinar poll of 1,700 people, 47% named judgment as the most important behavior in the age of AI; high-trust workers are 3.8 times as likely to use agentic tools.
How they got it
Scenario analysis by Deloitte's Center for Integrated Research and human capital practice, drawing on earlier Deloitte surveys, a horizon-scanning process, and executive interviews conducted in May 2026. The scenarios were tested with industry executives. The authors state that the scenarios are not forecasts.
Limits
- Almost every number comes from other Deloitte publications or third parties; sample sizes and methods for most of them are not given in this piece.
- The 1,700-person webinar poll is informal and self-selected.
- Deloitte sells the governance, workforce and change services that the recommendations point to.
Why it matters for companies
The labor-versus-infrastructure question is a real decision most companies have not taken explicitly, and it changes who owns agents, how they are measured and who is accountable when they fail. The practical advice is concrete: map actual workflows, place human checkpoints where judgment or exceptions occur rather than at a final sign-off, and track error rates alongside speed. The warning about junior roles is relevant to any firm that relies on entry-level work to build future experts. The 93/7 investment split is a quick test a manager can run against their own AI budget.
Summary based on the full text. Read the original for the full argument and data.Read the original on deloitte.com