AI Coding & Agents

AI agents can extend a solo builder's operating capacity only when context, decisions and write authority are visible. Begin with one known task, route shared and campaign-specific knowledge into a workspace, then bound each run with a concrete deliverable. Multi-asset campaigns add tickets for dependencies, checks and human decisions, while joined outcome data closes the next planning loop. Promote deterministic repetition into scripts, but keep consequential public writes behind approval and tested code.

Start here

1. Building a Marketing Agent Workspace — Give one known workflow durable context, work and result locations.
2. Operating AI Agents with Bounded Briefs — Make each research or production run finite and reviewable.
3. Orchestrating Agent Campaigns with Tickets — Expose dependencies and parallel work across a campaign.
4. Guarding Agent Content Publishing with Server-Side Approval — Separate agent judgment from public write authority.

Pages

Team design and operation

• Staffing a Solo Business with AI Agents — Billy Howell's newsletter case; chief of staff first, then reviewed and measured specialists
• Operating AI Agents with Bounded Briefs — Billy Howell; bounded candidate sets, one workspace and five-line status reports
• Promoting Agent Workflows into Automation — Billy Howell; prove judgment work first, then move stable routines into scripts

Marketing campaign systems

• Building a Marketing Agent Workspace — Shann³; separate shared knowledge, vertical context, active work, outputs and reviewed learnings
• Closing Campaign Feedback Loops with Data — Shann³; join channel, analytics, CRM and experiment records for the next brief
• Orchestrating Agent Campaigns with Tickets — Shann³; encode dependencies, checks, outputs and human-review states before parallel work

Permission and publishing boundaries

• Guarding Agent Content Publishing with Server-Side Approval — harsehaj's Browserbase blog engine; read and draft in the agent, validate and write in approval-only server code

Tools

• Archify — Agent Skill that turns system descriptions into interactive architecture, workflow, sequence, data-flow or lifecycle diagrams
• Claude Code — Agent used for scheduled research, SEO extraction and a visually checked web build
• Cursor — AI code editor used for an app MVP and a local markdown skill graph
• Rork — AI app builder used to generate a mobile MVP from references and prompts
• Shipper — AI app builder briefed with a researched user journey and visual references
• Fluent Korean — Output-style plugin for natural Korean responses from Claude Code

Gaps

• Cost and reliability comparisons between persistent agent rosters and discrete scheduled jobs
• Measured comparisons of folder-based campaign workspaces, databases and single-conversation workflows
• Attribution completeness, freshness failures and decision quality in agent-read marketing data layers
• Ticket overhead, parallel throughput and cross-asset consistency across multi-agent campaigns
• Incident rates, reviewer time and rollback performance for approval-gated agent writes
• Permission boundaries and audit logs for agents connected to email and private company documents
• Measured stopping rules for open-ended research when the candidate set is unknown
• Output-quality, export, ownership and App Store acceptance comparisons for AI app builders