workflows active: 55 last morning round: Mon 07:00 last deploy: 2026-10-06
Forbes / Operations WORK / 03

The operating playbook and AI

Two of us run 10+ concurrent lifecycle programs because the standards, prompt libraries, QA checklists, and governance are written down. Custom GPTs do the work that can be trained.

Problem

We had ten-plus lifecycle programs and two people. Anything that lived in someone's head became a bottleneck. AI could take a lot of the repeatable work, but not without guardrails, because it'll produce confident garbage at scale if you let it.

What I built

Standards and QA gates

I wrote the standards down first: what a program is, what every email has to contain, and what counts as launch-ready. QA checklists gate everything before it ships, whether a person or a machine made it.

An email template proof with the Forbes banner and a merge tag greeting, followed by quarterly planning notes
A template proof from quarterly planning, merge tags included.

Prompt libraries, custom GPTs, and governance

We keep prompt libraries for the recurring work, versioned like anything else we ship. Governance spells out what AI is allowed to touch, what needs human review, and who owns the output. Custom GPTs in ChatGPT Enterprise are trained on those standards, and the trainable work runs through them now.

The operating playbook what runs on what 10+ concurrent programs run by a team of two Custom GPTs the trainable work runs through them; specialists keep the judgment calls Standards what a program is, what every email has to contain, what counts as launch-ready Prompt libraries the recurring work, versioned like anything else we ship QA checklists gate everything before it ships, human or machine made Governance what AI may touch, what needs human review, who owns the output yellow outline: the layer the trainable work runs through
The playbook, and what runs through it.

Two buckets and value reporting

Contractor work is split into two buckets, trainable execution and real expertise, and we only pay outside rates for the second. I also set up value reporting for the martech stack, so every tool has to show what it contributed.

The two-bucket model operating playbook / who does what One bucket contractor work, undifferentiated Two buckets trainable execution: custom gpts, one operator real expertise: outside rates before after
Contractor work before and after the split.

Stack

chatgpt enterprise (custom gpts) / claude code / n8n / smartsheet / monday.com

Outcome

Campaign work modeled at 149 to 206 specialist hours has been delivered in roughly 71 to 74 hours by one AI-assisted operator: about a 60% reduction, projected to reach about 79% once the lifecycle agents launch. The modeling is against specialist-hour benchmarks, and everything still passes the same QA gate before it ships.

Next: Stedd, the founder page