The problem
He writes that when a big project lands, “Your best people stop doing client work and start chasing details.” He calls that “the real cost. Not the tools. The hours your team loses.”
He says the problem is context. Each person opens a new chat, explains the project from scratch and gets a slightly different answer. Then you get the wrong date, an old event name or a post that “makes a partner look small.” His view: “The fix is not a better prompt. It is giving the AI the same facts, every time, for everyone.”
The blueprint (as he describes it)
-
Write one skill with the event facts. The skill held the names, dates and venues, the partner order, how registration worked, tone rules for writing about partners, and ready-made invitation templates. Everyone on the team loaded it. He says anyone could then draft an invitation, a social post or a follow-up “in minutes.” He writes, “That is where most of the time came back.”
-
Start from the CRM you already have. HubSpot holds every contact SJ Innovation has built up over the years. That was the starting list.
-
Let the skill write each invite for one business. The skill added one true line about each business and matched each person to the one event that fit them:
- Tech and founders went to the Lovable workshop.
- Finance, legal and Chamber members went to the main day.
- Retail, hospitality and local services went to the small business session.
“If we knew nothing real about a business, it did not get an invite.”
-
Send from real people. Six NYC team members read each email and scheduled it from their own Gmail, in between their normal work. He says each email “came from a real person the recipient might know, not a marketing account.”
-
Keep score in GHL. All event registrations lived in GHL. The team cross-checked who they invited against who signed up and followed up only with the right people.
-
Connect an agent to the tool. They connected Claude to GHL. Registrations from the website landed straight in GHL. Claude prepared social posts and SMS reminders inside GHL.
-
Keep a person in the loop. “A person on our team approved every post and every SMS before it went out.”
-
Build the site fast. The event site, queensai.nyc, took two days to build in Lovable. After that they made small changes as needed.
-
Leave the room to people. The Chamber brought the community and the stage. Firstlight Cloud Xchange ran ticketing and kept the program together. He says the best thing they did in the room was sit next to people and help them one-on-one.
His suggested starting point for other teams is to pick one project in the next month and do five things:
- Write one skill with the facts.
- Share it with everyone on the project.
- Start from your existing CRM.
- Connect the one tool your team copies and pastes into most.
- Keep one rule: “a person says yes before anything goes out.”
Stated results
- About 277 people across three free events, October 6 to 8.
- Three people ran it part-time. “Nobody stopped their regular job.”
- Workshop: 52 of the 72 people who registered came.
- Event site built in two days.
- Time saved: not quantified.
- Invitations sent, reply rates and registrations from invitations: None stated.
Caveats
- Vendor and promotion: This is his own post on ManagedCoder, his training brand for agency owners. ManagedCoder offers a newsletter, paid live classes and free skills. The post promotes his free Event Command Center skill and mentions 34 other free skills.
- Lovable partnership: The site footer says ManagedCoder is “in partnership with Lovable.” The event site was built in Lovable and one event was a Lovable workshop.
- Team case: This is a team and agency case, not a solo operator. SJ Innovation was one of the event’s presenters.
- Attendance claim: Linking the workshop turnout to personal invitations is his opinion (“I think this is a big reason”). He also credits the hands-on format.
- Baseline claim: The “about 25%” free-event show-up figure is described as “the usual warning,” with no source given.
- Three things he says broke:
- On September 17 the site’s registration table looked empty and registration seemed broken. It was not, because registrations were going to GHL.
- Planning emails, pages and posts used five different names for the event.
- They planned for the wrong crowd size.
- Handles:
- X: @managedcoder (verified 2026-10-22: brand account linked from the managedcoder.com footer; X bio links managedcoder.com).
- Instagram @managedcoder, TikTok @managedcoder, Facebook facebook.com/managedcoder, LinkedIn /company/managedcoder and YouTube @managedcoder (verified 2026-10-22: brand accounts linked from the managedcoder.com footer).
- Personal LinkedIn: linkedin.com/in/sjinnovation (verified 2026-10-22: name and work match from search; not linked from his site).
- No personal X, Bluesky, Threads or Mastodon found.
Sources
- https://managedcoder.com/how-three-people-ran-a-277-person-ai-week (primary: first-person post by Shahed Islam, dated October 10, 2026)
Want agents like these working your distribution? See the Agent Stack Teardown. Get new teardowns weekly: subscribe (double opt-in).
Get one teardown a week.
Same format, a new operator every issue. Double opt-in.