Plugins Should Start With One Weekly Job

The short version: plugins should start with one weekly job. A long list of role-specific capabilities sounds impressive, but the operational win usually comes from making one recurring workflow finish cleanly enough that the team trusts it next Monday too.

That is the useful lesson behind the recent plugin momentum. When a plugin bundles skills, connectors, slash commands, and sub-agents, the temptation is to think in departments: sales plugin, marketing plugin, legal plugin. I think that frame is too broad for the first deployment. Departments are not workflows. They contain dozens of jobs with different stakes, data shapes, and failure costs. The system becomes much more useful when the team starts with one bounded weekly deliverable instead.

For example, a founder review packet is a job. A weekly outbound exception sweep is a job. A deliverability health memo is a job. Those workflows are narrow enough to define inputs, owners, and success conditions, but important enough to matter if they get faster. That is exactly the mindset behind The B2B Founder's AI Stack Should Start With One Workflow. The stack becomes practical once it wraps a real cadence, not a speculative persona.

There is a second advantage. A one-job plugin exposes trust gaps immediately. You learn which connectors are over-permissioned, which data is missing, and which parts of the workflow still need a trust index. That is much healthier than rolling out a general assistant that can theoretically do everything while nobody can name the one task it now completes better than before. The AI Operator story only works if the automation behaves like an owned operating lane, not a vague promise of intelligence.

My rule is simple: scope the first plugin around a weekly deliverable with a name, an owner, and a scoreboard. If it wins that job consistently, expand. If it cannot, do not hide the failure inside a broader role narrative.