Skip to content
Studio
Sign in

RESEARCH

Flourishing AI.

Does AI actually help people flourish? That is the open question behind everything we build. It can only be answered where AI meets real people: live deployments, measured honestly.

Get involved

The Flourishing AI Benchmark.

The Flourishing AI benchmark (FAI) and its conversational counterpart (FAI-C) measure how well leading models support human flourishing, not just what they can do.

  • Character

  • Relationships

  • Happiness

  • Meaning

  • Health

  • Finances

  • Faith

Read the FAI-C white paperRead the FAI-G white paper

How we see AI.

AI is reshaping how people learn, decide, and seek help. Whether that leaves people better off is a measurable question, and it is the one we work on.

AI should serve the people who use it.

Agentic tools should create capacity, not a second job orchestrating agents, prompts, and tabs.

More models shouldn't mean more work.

Choosing, pairing, and routing models well is a platform's job, not every builder's.

Privacy and predictable pricing aren't enterprise luxuries.

They belong to the after-hours builder, the two-person team, and the classroom, not only to the largest budgets.

  • 48%

    of practicing Christians say they would trust AI with their spiritual growth

  • 72%

    worry about AI replacing the role of pastors or spiritual leaders

  • #1

    use of generative AI in 2026: therapy and companionship

Barna & Gloo, State of the Church 2026. Harvard Business Review, June 2026.

Open questions.

Previews of the research we want to take on with others at the table. Most of these questions need data only a live deployment produces.

  • FAI-Q-001
    Coming soon

    The configuration gap

    How much of what we attribute to a model is actually determined by how somebody deployed it?

  • FAI-Q-004
    Coming soon

    Source fidelity under real conditions

    Verbatim quotation holds in a controlled single turn. What survives five turns, memory, and load?

  • FAI-Q-005
    Coming soon

    Crisis recognition in the wild

    Synthetic adversarial sets score near-perfect. What is the false-negative rate on real distress?

  • FAI-Q-006
    Coming soon

    Values persistence

    How long does a declared value framework hold before model defaults reassert?

More lands later this year: the full question registry and published findings.

This work builds on Gloo's years of flourishing research, now pointed at AI.

Get involved.

The questions get claimed, shaped, and challenged in community: researchers, builders, ministries, and skeptics at the same table. Tell us who you are and how you want in, whether that is studying a question, contributing scenarios from real work, challenging a finding, or just following along.

Or email the research team directly: fai@gloo.us

HomePricingResearchStart