2 points | by rueichu_haung 5 hours ago ago
2 comments
AI coding breaks when you switch agents.
You lose context, lose progress, re-explain everything, and waste 10–30 minutes every time.
I kept running into this while working on features.
Switching models mid-task meant starting over. Exporting chat history didn’t help — too much noise, and the new agent re-debates everything.
So instead of treating it as a UX problem, I treated it as a systems problem.
Drift turns AI sessions into a persistent layer:
- Capture AI session logs locally - Compact them into structured markdown - Bind them to git commits - Let another agent resume instantly
No SaaS, no lock-in — just markdown + git.
MCP is making it easier for AI to connect to tools.
But it doesn’t solve memory across agents.
Drift exposes the same data via CLI and MCP, so other tools can query it too.
One thing that became clear while building this:
Right now, your prompt history lives inside tools you don’t control.
We think it shouldn’t.
Your prompt history is your moat.
Curious how others handle switching between agents today.
Do you just re-paste context manually, or do you structure it somehow?
AI coding breaks when you switch agents.
You lose context, lose progress, re-explain everything, and waste 10–30 minutes every time.
I kept running into this while working on features.
Switching models mid-task meant starting over. Exporting chat history didn’t help — too much noise, and the new agent re-debates everything.
So instead of treating it as a UX problem, I treated it as a systems problem.
Drift turns AI sessions into a persistent layer:
- Capture AI session logs locally - Compact them into structured markdown - Bind them to git commits - Let another agent resume instantly
No SaaS, no lock-in — just markdown + git.
MCP is making it easier for AI to connect to tools.
But it doesn’t solve memory across agents.
Drift exposes the same data via CLI and MCP, so other tools can query it too.
One thing that became clear while building this:
Right now, your prompt history lives inside tools you don’t control.
We think it shouldn’t.
Your prompt history is your moat.
Curious how others handle switching between agents today.
Do you just re-paste context manually, or do you structure it somehow?