Missing Persistent Memory in Cowork
Head of Product Marketing at SaaS startup on automating product marketing with Claude Cowork
This is the line between an AI teammate and a stateless tool. In practice, the marketer is rebuilding the same brain every time, by reattaching brand docs, product facts, source lists, and formatting rules before Cowork can do useful work. That adds setup labor, raises the chance of missing a key source like CRM or Gong context, and keeps repeatable workflows in the hands of one power user instead of the broader go to market team.
-
The current workaround is a manual context file plus connectors and skills. Each session needs an MD file with role, brand, voice, and goals, then linked sources like Drive, NotebookLM, newsletters, and internal docs. Even weekly workflows still depend on that setup being rebuilt and maintained.
-
The missing memory shows up most when the work needs company specific judgment. The marketer can trust Cowork for internal Slack updates, but still checks external content, competitive analyses, and churn reporting because product facts, messaging rules, and source priority change over time and are easy to lose between sessions.
-
Templates are the product answer because they turn agent building from writing logic into filling in blanks. The interview describes the ideal flow as a prebuilt sequence of steps, source slots, and review rules. Anthropic already supports integrations like Google Drive and Canva, so the bottleneck is not access to tools, it is persistent orchestration across them.
The next step for agent products is a shared working memory layer that keeps source of truth docs current, remembers preferences across runs, and ships with opinionated templates for common jobs like competitor tracking, weekly updates, and outreach prep. Once that exists, AI work moves from expert configured automations to software a sales, CS, or marketing team can actually run themselves.