Matt Swulinski, the growth mind behind Superhuman and Whisperflow, has a blunt take for founders: your current marketing team is probably obsolete. He argues that in an AI-driven world, most marketing roles will be eaten by AI agents, leaving a tiny fraction of truly essential humans. His acid test for new hires reveals a stark reality: less than 1% of candidates grasp this shift.

Key Takeaways

  • Matt Swulinski argues most current marketing professionals lack the "systems thinking" required for an AI-native era and will be replaced by automated agents.
  • A systems thinker can meticulously break down their job, map interdependencies, and use AI to automate 80% of lower-leverage tasks, freeing up human bandwidth for strategy.
  • Swulinski tests for this by asking candidates about their personal AI workflows, an approach that reveals deep automation skills versus superficial AI familiarity.
  • He predicts a future where companies operate with only 20% human strategic oversight and 80% execution by AI agents, giving early adopters a massive competitive edge.
  • To build your own AI-native capabilities, founders should adopt structured AI workflows, like Swulinski's 'Session End' skill, to compound learnings and retain memory.

The Swulinski's 'Session End' AI Skill for Compounding Work

This framework is designed to ensure every AI interaction builds on previous knowledge, creating a constantly improving system.

  • Objective: To analyze, distill, and store all work and learnings from an AI terminal or cloud code session into a knowledge base, enabling compounding improvements and memory retention.
  • Tool Integration: Connect your AI (e.g., Claude Code) to an Obsidian vault (or similar node-based system) to serve as a daily log.
  • Skill Trigger: Invoke the 'session end' skill when a work session is almost done or needs to be concluded.
  • Analysis and Distillation: The skill analyzes the current session, distilling what was worked on, the problem frame, what was accomplished, and any outstanding items.
  • Knowledge Transfer: It moves the distilled information into the Obsidian vault, creating a chronological log with decisions, open tasks, learnings, and essential notes.
  • Interrelationship Mapping: The agent can then analyze this growing web of knowledge, identifying connections between past and present work (e.g., working on similar creative types for Meta ads across different sessions), which humans would typically forget.

When This Works (and When It Doesn't)

This skill ensures that every AI work session is compounding, building a personal 'treasure trove' of documented work and memory. It helps to overcome the limitation of AI memory and provides a structured way to revisit past work and understand how different efforts contribute to overall progress, acting as a valuable resource for anyone building, tweaking, and refining processes. Swulinski's insight here is gold for technical marketing, product growth, or any role heavy on process iteration.

However, this framework works best for tasks with clear inputs, outputs, and iterative cycles. It falters when the "work" is highly abstract, subjective creative ideation, or requires nuanced human intuition that AI cannot yet mimic. If your daily tasks don't involve extensive AI terminal use or cloud code, integrating an Obsidian vault might feel like overkill. This is for the builder who lives in their AI console, not the one who uses it for occasional brainstorming.

What to Do With This

Tomorrow, pick one repetitive task in your startup that you currently do with AI. Perhaps you use ChatGPT to draft weekly social media posts or Claude to outline blog articles. Before you close that AI session, implement Swulinski's 'Session End' skill. For example, instruct your AI: "End session. Analyze what we worked on today (e.g., 'drafted 5 Instagram posts for product launch'), what problem this solved, key learnings (e.g., 'longer posts perform better for this audience'), and what's next. Summarize this for my personal knowledge base." Then, manually transfer this summary to a simple text file or Notion page. This simple habit starts building your personal 'memory layer' for AI work, revealing patterns and accelerating your learning in ways traditional human memory simply can't match.