The second launch learns from the first
Marketing and growthStartup growth team running launches from Claude Code
Our first launch taught us a lot: posting at midnight Pacific got buried, the comparison table drove most signups, and the 'free for 14 days' line confused people. Two months later we planned the next launch and the agent proposed midnight again.
How do we get our AI agent to use the lessons from our last product launch?
Save each launch lesson in ArcRoom with arcroom decide and a reason, and the assets that worked with complete_task. When anyone starts the next launch prompt, recall brings the lessons, and wrong ones can be pulled with arcroom wrong. Lessons are only as good as what you record. ArcRoom doesn't measure the launch for you.
Also asked as: Our AI agent proposed the same launch time that flopped last time. How do we stop that? · How do I keep launch lessons where AI agents will see them? · Can an AI agent remember what worked in our previous launch?
After the launch, each lesson is saved with arcroom decide and a reason.
- Saved answer
- Decision
- Recall
- Wrong-memory report
Recall: 3 launch decisions
Step by step
- After the launch, each lesson is saved with arcroom decide and a reason.
- Launch assets that worked are saved with complete_task.
- When anyone starts the next launch prompt, recall brings the lessons.
- Wrong lessons can be pulled with arcroom wrong.
What the agent sees
Recall: 3 launch decisions
• post at 8:30 a.m. PT, not midnight (why: buried by 6 a.m.)
• lead with the comparison table (why: 61% of signups clicked it)
• say 'free during beta', not '14 days'Compared with today's tools
Launch retro doc in Notion
Write a retro after each launch and paste it into the agent next time.
- Setup
- 1 hour
- Ongoing
- Remember to find and paste the retro
ArcRoom
Every agent reads the team playbook before it works and writes to it when it finishes. Nothing to update by hand.
- Setup
- 2 min (npx arcroom init)
- Ongoing
- None
The retro doc is better for people to read; unlike Mem0, where the agent has to search for it, ArcRoom recall puts the three lessons in front of the first launch prompt with their reasons.
Where it stops
Lessons are only as good as what you record. ArcRoom doesn't measure the launch for you.
Try it in your project
Works with the agents your team already uses, and any agent over MCP or the HTTP API.