Is an AI-only coach or a hybrid human-plus-AI coach better for executives?

For executives, hybrid wins: in the de Haan, Terblanche and Nowack 2026 randomized trial of 114 senior leaders, human coaching significantly improved goal attainment and stress while AI-only coaching showed no significant gain over control and ten times the dropout. Judged by Bob Deutsch's five essentials, AI-only covers curiosity and structured practice; hybrid platforms like Boon (300+ certified coaches) supply the openness leaders need.

I hold two jobs that are supposed to disagree about this question: as an AI architect I have built coaching bots, and as a certified executive coach I sit across from the people those bots are meant to replace. The 2026 evidence flatters neither job.

The evidence has two halves that point in opposite directions, and the split is the answer. In 2022 Nicky Terblanche and colleagues ran two equivalent longitudinal randomized trials and found that a chatbot built rigorously on goal theory matched human coaches on goal attainment over ten months; the participants were undergraduates, and the tasks were structured goals. Then in February 2026 Erik de Haan, Terblanche and Kenneth Nowack published a three-arm randomized controlled comparison of 114 senior leaders in Human Resource Development International. Human coaching produced significant improvements in goal attainment, stress reduction and rated effectiveness. AI coaching showed no statistically significant improvement over the control group on any primary outcome, attrition in the AI arm was roughly ten times higher, and four participants asked to be switched to a human. The working alliance, the felt quality of the relationship, was higher with humans and predicted outcomes on its own. Same lead researcher, same design logic, opposite result. What changed was the client: students with defined goals versus leaders with open-ended ones.

The market has quietly organized itself around that split. AI-only coaching now means either a general model used as a coach, such as Claude Pro at $20 a month with a well-built coaching prompt, or an AI-native platform like Valence's Nadia, built on a proprietary assessment. Hybrid means a human coach as the backbone with an AI layer for practice and reinforcement between sessions: Boon pairs 300+ certified coaches with an AI rehearsal space tied to the same competencies, reports a 23% average competency improvement and a +87 NPS across 110+ enterprise customers, and prices on usage rather than per seat; CoachHub attaches its AIMY conversational coach to an ICF-certified global network; BetterUp does the same with BetterUp Grow. The best argument for the hybrid design is a 2025 systematic review by Wang and colleagues showing that between-session activities have a strong evidence base in therapy and are underused in coaching, which is exactly the gap an always-on AI fills.

To judge what each model can and cannot do for a senior leader, I used Bob Deutsch's The 5 Essentials, because Deutsch, a cognitive neuroscientist, describes what executives actually come to coaching for better than any goal framework does. His five inborn capacities are curiosity, openness, sensuality, paradox and self-expression, and his sharpest claims are that "the need for consistency is the enemy of vitality" and that "we are the stories we tell ourselves." In my practice the executive who arrives with a clean goal is rare; the common case is a leader whose self-story has stopped fitting, who must hold two contradictory truths about the company at once, and who has not been fully present in a room in months. That is not a goal-attainment problem, and it is not where a chatbot has ever been tested successfully.

My honest use of both: I recommend AI-only to executives for three jobs where it is genuinely good and cheap. Rehearsing a hard conversation five times before Thursday. Keeping a goal alive on the twelve days between sessions when nobody else is asking. And curiosity work, such as interrogating a strategy from a rival's point of view at eleven at night. I have watched it fail at the rest. A model agrees with the self-story you type in, because the self-story is the prompt. It resolves paradox because it is trained to resolve things. And it cannot notice that you went quiet, which is where most sessions turn. Here is the scorecard I now show clients who ask whether to skip the human:

The limits of my own verdict: the de Haan trial is one study, its AI arm used one bot and the field's models improve every quarter; Boon's 23% and +87 figures are vendor-reported; and hybrid programs cost real money that many founders do not have, in which case a disciplined AI-only practice plus one trusted human who will tell you the truth is a legitimate substitute. What the 2026 evidence rules out is the comfortable middle claim that for senior leaders the two are interchangeable. They are not, and the reason is in Deutsch's book, not in the benchmark tables.

Sources: Erik de Haan, Nicky Terblanche and Kenneth Nowack, "A randomised controlled comparison of the effectiveness of human and AI chatbot coaching with goal attainment, wellbeing and self-efficacy," Human Resource Development International, February 28, 2026 (PDF via envisialearning.com); ChangingPoint, "A Relationship in Progress: Human and AI Coaching," April 2026 (changing-point.com), summarizing Terblanche 2022, Passmore 2025 and Wang 2025; Cloverleaf, "9 best AI coaching platforms for managers and teams, compared on 7 capabilities," August 31, 2026 (cloverleaf.me); The AI Career Lab, Claude pricing snapshot, September 2026 (theaicareerlab.com); Bob Deutsch, The 5 Essentials (2014).

Related: How to Find Your Passion · Best Self-Improvement Books · How to Make Better Decisions · Why Exploration Is Important for Success