AI Frontier Ledger · The Signal · October 4, 2026
By Joe Moore · Research submitted September 28–30, with September 26 background.
AI can collect more answers. Leaders still have to earn the truth.
For agency owners, consultants, and founders, that distinction matters. You can automate a client update, summarize a team meeting, or add an AI assistant to the work. But a faster response does not tell you whether someone feels safe enough to say what is wrong.
AI workplace trust: the signal behind the score
A September 30 paper by Thilo Tamme and colleagues studied AI-assisted employee listening inside one global consulting firm. Across 44 first-session interviews, favorable ratings sometimes appeared alongside substantive concerns disclosed later: 20–41% of sessions, depending on the threshold. Participants also qualified or softened what they shared.
The paper is available on arXiv and lists acceptance at HICSS-60. This is a small field study in one organization, not a workforce-wide estimate. Read the paper and its study details.
My reading: a dashboard can look comfortable while the conversation underneath it is not. For a local agency, the useful question is whether a client or teammate can challenge the system—and whether a person will act on that challenge.
Why this belongs in Tuesday’s room
Jesse Pudles’s upcoming session is Keeping Culture Human When AI Joins the Workforce. Its announced focus is trust, engagement, and culture as AI tools and agents enter the work.
This research supplies a question for that conversation: what makes people willing to tell the truth when AI is part of the listening?
The connection is our editorial framing. It is not a claim that Jesse participated in, endorsed, or has already discussed this study.
Two more signals worth bringing to the table
Useful to whom—and at which stage?
A September 28 arXiv preprint by Sepideh Tajarmakan and Khashayar Hojjati Emami surveyed 443 designers in 43 countries. Reported AI use varied across the design workflow; 79.7% used it in at least one phase. Adoption did not mean use across every phase. Read the designer survey.
For an agency, map the actual handoff: discovery, first draft, review, delivery. Decide where AI is useful and who owns the final decision. One adoption percentage cannot describe that whole chain.
More prompts can be the wrong target
Earlier context, submitted September 26: Jie Gong and colleagues studied an internal AI query requirement at a medical-device company. Their preprint reports repeated or off-task queries under the original target; lowering the target reduced much of that activity. The authors estimate higher monthly sales among sales employees after the change. Read the incentive study.
That is one company’s experience. The question for your own business is practical: are you measuring customer outcomes, or rewarding people for touching the tool?
Clarity before code. Culture before the scoreboard.
Before adding another agent, put three questions in front of the people doing the work:
What customer problem should this improve?
What can the team question, override, or stop?
Who owns the decision when the output is wrong?
Your words build worlds. That includes the instructions you give a machine—and the permission you give a person to disagree.
Tuesday, October 6: two conversations, two registrations
AI Mastermind · Noon–1 PM Central. Join Joe Moore, Katrena Drake, and the live room. Bring the AI news, your questions, and the work in front of you.
Register for the AI Mastermind →
AI Mastermind AfterShow · 1 PM Central. Jesse Pudles joins Katrena Drake and Joe Moore for Keeping Culture Human When AI Joins the Workforce, part of the Create Your Own Gravity podcast.
Register for the AfterShow →
Each series has its own one-time registration and email list. Registering for the Mastermind does not register you for the AfterShow.
FAQs
Does this research prove AI makes teams less honest?
No. The employee-listening study examines how ratings and disclosures differed in one organization. It raises questions about trust and the listening process; it does not establish a universal effect of AI.
Is Jesse’s AfterShow the same as the AI Mastermind?
No. The participatory Mastermind begins at noon Central. The separate AfterShow guest conversation begins at 1 PM Central, with its own registration.
Can small agencies use these questions?
Yes. Apply them to one workflow: who reviews the output, who can challenge it, and what customer outcome you measure. The research findings themselves should not be treated as small-agency performance benchmarks.
