How I work with AI

The rule I work by

AI handles work that is repetitive, well defined, and checkable. People handle work that needs taste, negotiation, or accountability. For a creative team, a system that produces a draft and stops is worth more than one that acts on its own, because the review step is where the quality comes from.

What I have shipped

• Status and alert dashboards at Everpure. Agents assembled the weekly delivery picture and flagged at-risk projects. I reviewed the output before it went out, and stakeholders stopped chasing status.
• Asset management at Pegma Studio. Agents that catalog, tag, and retrieve production files so the library stays usable.
• Scheduled competitive intelligence. Agents that research the market and produce a standing report on a fixed schedule. I review and correct it before I use it.
• Content calendar production. Turns a plan into scheduled, briefed items so marketing runs continuously instead of in bursts.

Where I would start on a creative team

• Intake triage. First-pass scoping and routing on inbound requests, with a producer confirming before anyone commits to it.
• Brief quality. Checking incoming briefs against a standard and sending the gaps back before work starts, which is the cheapest place to fix a project.
• Status and risk reporting. The most manual effort with the least need for a person to write it.
• Asset tagging and archiving. The work nobody does consistently and everybody needs done.
• Retrospective analysis. Reading delivery data across a quarter to find where the time goes.

How I decide whether it stays

Before I hand a task to a system I write down what a good output looks like and what a bad one looks like. Then I measure time saved against time spent correcting, spot-check a sample every cycle, and keep prompts and rules under version control so a change is traceable. If a system needs more correction than it saves, it comes out.

What I will not automate

Creative judgment, prioritization calls between teams, performance conversations, and anything where a person needs to own the answer. Speed that costs trust is not worth having.

Adoption is the hard part

Automation fails the way tool migrations fail: the system is fine and nobody uses it. So I plan it like a migration, naming the people affected, the training, the cutover date, and a check on adoption afterward. A workflow people route around is a workflow that does not exist.