DEV Community

Shivani Srimanthula
Shivani Srimanthula

Posted on

I Built an AI Agent That Never Forgets a Promise

You walk into a meeting. The client asks about the thing you said you'd send last week. You have no idea what they mean.

Every salesperson, founder and account manager has been there. I built an agent so I'd never be there again.

What it does

The Meeting Prep Agent is built on Hindsight, a memory system for AI agents. It gives every contact their own memory bank. After each meeting, I log what happened: the notes, the promises made, and the follow-ups still open. Before the next meeting, I ask for a brief and get everything I need in one place.

How Hindsight powers it

Memory is the product here, not a feature:

  • One memory bank per contact. Each relationship builds its own history, so nothing gets mixed up.
  • retain() after every meeting. Notes, promises and follow-ups are saved as separate memories.
  • reflect() before the next one. Hindsight reasons over everything on file and writes the brief.

Take Hindsight away and there's nothing to brief from.

The test that convinced me

I ran a three-step story against live Hindsight with a fictional client:

  1. Before any meetings: the brief is empty. No history exists.
  2. Meeting 1: the client needs a security whitepaper by January 17, and I promise it.
  3. Meeting 2: I never sent it. The client is frustrated and mentions a competitor who already delivered.

Then I asked for a brief before Meeting 3. It flagged the missed deadline, tied it to the competitor threat, and told me to confirm no exclusivity had been granted. I didn't re-enter any of it. That's the moment memory changed what the agent told me.

The honest version

This is a memory-powered agent that you trigger. You log a meeting and ask for a brief. Today, promises and follow-ups are entered as separate fields, and it doesn't check whether you kept a promise unless you mark it resolved.

What's next

  • An LLM that reads raw meeting notes and extracts promises itself.
  • Reminders before calendar events.
  • A single chat where you say "just met Maya, she wants pricing by Friday" and the agent decides what to do.

Code: github.com/shivanisrimanthulacom-cloud/meeting-prep-agent-hindsight

Top comments (1)