Agent Lab Impact Report: 122 Working Agents in Seven Weeks

BNEDai Agent Lab participants built at least 122 working AI agents across seven weeks of live sessions, from a pool of 132 source-record registrations. Of those surveyed after their sessions, 90.5% reported moderate or extreme confidence to build their next agent independently, and 100% committed to teaching one other person what they learned. [1][2][3][4] These are first-cohort numbers from a program that launched June 29, 2026 and ran at least a dozen live sessions through August 14, 2026.

This page walks through the data: what happened, who was in the room, what they built, and what the numbers do and do not tell you.

How many people registered and how many built?

132 source-record registrations came through a combination of manual calendar invitations, the bnedai.com landing page, and Luma event listings. [1] Of those, at least 122 participants built working AI agents during live sessions, verified through the instructor's session records. [2]

Those are two different numbers measuring two different things. Registration is intent. A build is output. Some registrants did not attend, and completed agent types varied. The 122 figure reflects participants who left a session with a functioning agent, not those who merely logged in.

Separately, 224 or more Gumloop referral accounts were created through BNEDai's ambassador dashboard. [5] That number tracks platform sign-ups from BNEDai's referral link. It is distinct from registrations (some account holders never registered for a session) and distinct from builds (an account creation is not the same as a completed agent).

9 participants registered for two or more sessions without being prompted to return. [6]

Who was in the room?

The participant pool was roughly 95% women. [7] Registrants included senior managers, PhDs, EdDs, MBAs, MPAs, PMPs, and clinical leaders. Their employers spanned Fortune 500 companies, global energy firms, financial services, travel and hospitality, education, and municipal government. [7] Participants joined from multiple U.S. states and internationally from London and Salvador, Brazil. [7]

These profiles are generalized here. The full employer list and named participants appear on the Impact Report page only, per standing disclosure policy.

What did they build?

The most common first build was a Meeting Prep agent, which reads a calendar, pulls context on attendees, and produces a briefing document before the meeting starts. [8]

Other builds included:

- Scout agents for job search and executive search applications [8] - Multi-company meeting prep agents, where one participant who hosts across several companies built separate agents per company [8] - A Job Search Manager with Scout, Matcher, and Ghostwriter sub-agents working together [8]

Every build ran on the Gumloop no-code platform. Gumloop offers a 14-day trial, then paid plans. [9] The builds followed a single safety rule taught in every session: draft, never send. [10] Agents prepare output for human review. They do not act on their own.

Jacqueline V. Twillie's teaching frame for the relationship between builder and agent: "Agents are your employees, you are the manager." [10]

What did the session data show?

Two sessions have detailed attendance and engagement records:

June 29 session: 24 invited, 15 built live, 63% show rate. [11] This was the first session.

July 20 session: 28 registered, 21 attended live, 75% show rate, 53-minute average stay in a 73-minute session, 5.0 feedback across three ratings categories. [12]

A Threads post on July 8 about the Agent Lab reached 15,200 views and 344 replies. [13]

What did participants say?

### Survey data (21 respondents, Google Forms)

90.5% of surveyed participants reported moderate or extreme confidence to build their next agent independently. [3] Instruction clarity rated 4.95 out of 5. [14] 100% of surveyed participants committed to teaching one other person what they learned. [4]

These survey results come from 21 Google Forms respondents. That is the sample, not the full participant population. The 90.5% confidence figure describes those 21 people, not all 122 builders. No extrapolation to the full group is claimed.

### Luma feedback ratings (14 respondents)

13 of 14 Luma feedback ratings were five stars. One was four stars (Sean Kosofsky, August 3 session). [15]

Named reviews from the Luma export:

- Kristine Frost, 5 out of 5: "Wow...what a powerful session!" [15] - Alexandria Sarmiento, 5 out of 5: "working every day to make my life easier" [15] - Susan Quinn, 5 out of 5, August 10: "Super great and not as scary as I thought it was going to be" [15]

What barriers did participants encounter?

Three barriers surfaced across the seven-week run:

1. 01

   Managed-device restrictions.

   One company in the Atlanta metro area had corporate device policies that blocked the Gumloop platform entirely. The participant could not access the build environment from her work machine. [16]

1. 02

   Credit and billing questions.

   Some participants had questions about Gumloop's billing model and how credits work on the platform. [16]

1. 03

   Trial expiration concerns.

   Participants who started on the 14-day trial asked about what happens when it ends and whether their agents would stop running. [16]

These barriers point to the friction that sits between a live build and a sustained practice. The agent works. The question is whether the workplace lets the agent keep working.

What does this report not cover?

This report documents outputs and participant feedback from the first seven weeks. It does not measure:

- Whether agents are still running weeks or months after the session - Revenue impact or productivity gains from the agents built - Long-term retention of the skills taught - How many of the 100% who committed to teach someone followed through

Those are different questions requiring different instruments. If and when that data exists, it will get its own evidence page. For a closer look at how the 90.5% confidence figure was measured, including the survey instrument and response distribution, see 90.5% Build Confidence: What We Measured and How. The underlying numbers on this page were first published in the Impact Report No. 1 press release.

If you want to understand what an AI agent is before reading the build data, start there. If you want to build one yourself, the step-by-step build guide walks through a single session. For the operating rules taught alongside every build, see AI agent operating rules. To register for a live session, visit the Agent Lab.

Methodology

BNEDai Agent Lab ran at least a dozen live, one-hour sessions between June 29 and August 14, 2026. Registrations were tracked through manual calendar invitations, the bnedai.com landing page, and Luma event listings. The instructor's session records documented which participants completed working agent builds.

Post-session surveys were administered through Google Forms. 21 participants completed the survey. Questions covered build confidence (rated on a scale from no confidence to extreme confidence), instruction clarity (rated on a 1-to-5 scale), and commitment to teach another person. Luma's built-in feedback system captured an additional 14 ratings with optional written reviews.

The 132 registration figure is a source-record count combining all three intake channels. The 122 build figure comes from the instructor's session records, attested August 15, 2026. The 224+ Gumloop referral account figure comes from BNEDai's ambassador dashboard on the Gumloop platform and measures account creation, not session attendance or agent completion.

No third-party audit was conducted. All data is self-reported by BNEDai and its participants.

Source citation table

| # | Claim | Source | Date | Link | |---|---|---|---|---| | 1 | 132 source-record registrations | Manual calendar + bnedai.com landing page + Luma | June 29 to Aug 14, 2026 | Impact Report | | 2 | At least 122 participants built working agents | Instructor's session records, attested | Aug 15, 2026 | Impact Report | | 3 | 90.5% reported moderate or extreme confidence to build independently | Google Forms survey, 21 respondents | Aug 2026 | Impact Report | | 4 | 100% committed to teaching one other person | Google Forms survey, 21 respondents | Aug 2026 | Impact Report | | 5 | 224+ Gumloop referral account creations | Ambassador dashboard | Aug 2026 | Impact Report | | 6 | 9 repeat registrants (2+ sessions) | Registration records | June-Aug 2026 | Impact Report | | 7 | ~95% women; senior managers, PhDs, EdDs, MBAs, MPAs, PMPs, clinical leaders; Fortune 500, global energy, financial services, travel/hospitality, education, municipal government; multiple US states + London + Salvador Brazil | Registration records and session records | June-Aug 2026 | Impact Report | | 8 | Builds: Meeting Prep agents, Scout agents, multi-company meeting prep, Job Search Manager with sub-agents | Instructor's session records | June-Aug 2026 | Impact Report | | 9 | Gumloop: 14-day trial then paid plans | Gumloop platform terms | Current | Gumloop | | 10 | "Draft, never send" safety rule; "Agents are your employees, you are the manager" | Session transcripts | June-Aug 2026 | Impact Report | | 11 | June 29: 24 invited, 15 built live, 63% show rate | Session records | June 29, 2026 | Impact Report | | 12 | July 20: 28 registered, 21 attended, 75% show rate, 53-min avg stay in 73-min session, 5.0 feedback | Session records and analytics | July 20, 2026 | Impact Report | | 13 | Threads post: 15,200 views, 344 replies | Threads post analytics | July 8, 2026 | Impact Report | | 14 | 4.95/5 instruction clarity rating | Google Forms survey, 21 respondents | Aug 2026 | Impact Report | | 15 | Luma feedback: 13/14 five-star, 1 four-star; named reviews from Frost, Sarmiento, Quinn | Luma feedback export, 14 respondents | June-Aug 2026 | Impact Report | | 16 | Barriers: managed-device restrictions (Atlanta metro), credit/billing questions, trial expiration concerns | Session records and participant questions | June-Aug 2026 | Impact Report |