Writing
Papers on building agentic systems for marketing.
Long-form, versioned writing. Every paper is a Markdown file in a public GitHub repo, so the history, the sources and the discussion stay in the open.
Source
edwinlov3tt/edwin-sitemain · builds on every push
Latest commitPersonal site: AstroPaper + Claude Design UI, writing vault, members gateb252058 · Oct 13, 2026
008Eval gates: shipping agent output without babysitting itMost teams review every agent output by hand, which caps throughput at human speed. This paper proposes eval gates: rubric-scored checkpoints that decide whether an artifact advances, revises, or escalates — and shows how to calibrate them against human judgement.Draft v0.4EvalsAgentsSep 30 2026 · 1 min007Context budgets: what to give an agent, and what to withholdMore context is not better context. A framework for budgeting what an agent sees per step — brand rules, prior campaigns, audience data — and the failure modes that appear when budgets are ignored.AgentsContextAug 12 2026 · 1 min006The agent loop: a reference architecture for marketing automationMarketing teams meet LLMs through a chat window and rarely leave it. This paper describes the agent loop — a bounded cycle of plan, act, evaluate and revise over typed artifacts — and how to implement it with Claude Code so a team can trust, inspect and change it.AgentsClaude CodeJun 02 2026 · 3 min005A reproducible environment for agent projectsAgent projects fail in boring ways: a drifted dependency, a missing CLI, a secret in the wrong place. A devcontainer-based template that makes every project run identically on day one.WorkflowMay 14 2026 · 1 min004Marketing teams should learn to orchestrate agentsAn argument that the durable marketing skill of the next few years is not prompting, but designing and supervising systems of agents — and what that means for hiring.MarketingMay 03 2026 · 1 min003From prompt to pipeline: automating campaign briefsTurning a messy kickoff thread into a structured brief schema, a calendar and first drafts — and why the schema mattered more than the model.AutomationApr 11 2026 · 1 min002Notes on Claude Code subagentsField notes on splitting work across subagents: when it helps, when it multiplies confusion, and the conventions that kept mine predictable.Claude CodeDec 08 2025 · 1 min001Why I stopped writing mega-promptsThe short essay that started this repo: one prompt that does everything is a system you cannot debug.AgentsOct 21 2025 · 1 min
Series
Agentic Marketing Systems
A five-part reference on building marketing automation that a team can trust.
2 / 5
- 01The agent looplive
- 02Context budgetslive
- 03Eval gatesdraft
- 04Escalation pathsplanned
- 05Operating the systemplanned
Series
Environments
Setups that make agent projects reproducible and boring — in the good way.
1 / 2
- 01A reproducible environmentlive
- 02Secrets & sandboxesplanned