How I use AI agents to prepare briefings before every meeting so I always know the context, history, and next steps.
I used to spend 15 to 30 minutes before every meeting pulling up context. Who is this person? What did we discuss last time? What are their current priorities? Are there any open items I promised to follow up on?
Multiply that by 8 to 12 meetings per week, and you are looking at 3 to 5 hours of prep time. Not thinking time. Not strategy time. Just pulling information from different places and reading through it.
Now an agent does this for me. Every meeting starts with a one-page briefing that covers everything I need to know. Total prep time on my end: two minutes to scan the briefing.
The briefing is a single markdown file with five sections. Each section answers a specific question I would otherwise need to answer manually.
Context summary. Who am I meeting with, what organization are they from, what is their role. For recurring contacts, this includes a brief history: when we first connected, what the relationship is about, any personal details I should remember (they mentioned a product launch last time, their team just went through layoffs, they are expanding into a new market).
Last interaction. A summary of our most recent conversation. What was discussed, what decisions were made, what was left open. If the last interaction was an email thread rather than a meeting, it summarizes the thread.
Open items. Anything I committed to doing that has not been completed. This is the section that prevents the most embarrassing moments. Nothing damages trust faster than forgetting you promised to send something.
Their recent activity. If this person is active on LinkedIn or has recent press mentions, the briefing includes a two-sentence summary. This is not for stalking. It is for starting the conversation with awareness. “I saw you launched the new product last week, how is the reception?” is a better opener than “so, what have you been up to?”
Suggested agenda. Based on the context and open items, the agent proposes three to five talking points. I do not always follow them, but having a suggested structure means I never sit down wondering what to cover.
The exact architecture, memory layers, and delegation patterns I use to run 50 agents across two businesses.
Get the AI Agent Blueprint →The automation triggers 30 minutes before any meeting on my calendar. Here is what happens in sequence.
The calendar agent reads the upcoming meeting details: title, attendees, time, any notes or links in the calendar event. It extracts the attendee names and email addresses.
The memory search agent queries my contact database and conversation history for each attendee. It pulls the most recent interactions, any tagged relationships (client, prospect, partner, team member), and any open tasks associated with that contact.
The research agent does a quick external scan. It checks LinkedIn for recent posts and job changes. It checks news for company mentions. This step adds about 20 seconds but has saved me from awkward conversations multiple times. You do not want to pitch a partnership to someone whose company just announced major layoffs. Context matters.
The briefing agent compiles everything into the standard five-section format and saves it to a shared folder. I get a notification on Telegram with a link to the file.
Total elapsed time from trigger to delivered briefing: about 90 seconds. The bottleneck is the external research step. Everything else is database queries and text generation.
The hardest part was not the AI. It was the calendar integration.
My meeting schedule lives in Google Calendar. The agent needs to read upcoming events, extract attendee information, and trigger at the right time. I use a cron job that checks the calendar every 15 minutes and fires the briefing workflow for any meeting starting in the next 30 to 45 minutes.
The timing buffer matters. Too early and the briefing might miss last-minute calendar additions. Too late and I do not have time to read it before the meeting starts. The 30-minute window works for my schedule because I rarely add meetings with less than 30 minutes notice.
For meetings added at the last minute, I can trigger the briefing manually with a single command. The agent generates the same output, just without the scheduling automation.
Start with the output format, not the input. I designed the briefing template first, then figured out where to get each piece of information. This prevented scope creep. Every data source in the pipeline exists because it fills a specific section of the briefing. No extra data gets collected just because it is available.
The contact database is the foundation. The quality of the briefings depends entirely on the quality of my contact records. If I do not log conversations and decisions after meetings, the briefings for future meetings will be thin. This created a virtuous cycle: the briefing system motivated me to keep better records because I could see the immediate benefit in future briefings.
External research needs guardrails. The first version of the research step was too thorough. It would pull ten LinkedIn posts, three news articles, and a company summary. The briefing ballooned to three pages, which defeated the purpose. Now the research is capped: one recent LinkedIn post (if any), one news mention (if any), one sentence on any job or company changes. Concise beats thorough for meeting prep.
Not every meeting needs the same depth. A weekly standup with my team does not need external research or a suggested agenda. A first call with a new prospect needs maximum depth. I set up three briefing tiers: light (just context and open items), standard (full five sections), and deep (five sections plus company research and competitive notes). The tier is determined by a tag on the calendar event. Default is standard.
Before this system, I spent roughly 4 hours per week on meeting prep. After: about 30 minutes (scanning briefings plus occasional manual notes).
That is 3.5 hours per week, or about 14 hours per month. At a conservative value of $100/hour for my time, that is $1,400/month in recovered capacity. The system cost nothing beyond the infrastructure I already run for other agents.
But the time savings are not the real value. The real value is quality. My meetings are better. I remember what people told me. I follow up on commitments. I reference things they care about. I come in with a plan instead of figuring out the agenda in the first five minutes.
One client told me directly: “You are the only consultant who always remembers where we left off.” I did not tell them about the agent. The result is the same either way.
You do not need 50 agents to do this. A simple version works with three components.
A calendar reader that extracts meeting details. Most calendar APIs make this straightforward. Google Calendar, Outlook, and CalDAV all support programmatic access.
A contact/notes database. This can be as simple as a spreadsheet or as sophisticated as a CRM. The key requirement is that it is searchable by contact name or email.
A text generation step that formats the available data into a consistent briefing template. Any language model can do this. The template is the important part, not the model.
Start with the light tier: just context and open items. That alone will save you time and prevent the most common meeting failures. Add the research and agenda suggestions once the foundation works.
The newsletter covers practical automation patterns like this one, with implementation details you can apply the same week.
Every meeting is an opportunity to build trust or lose it. Walking in cold, forgetting context, missing follow-ups: these are not character flaws. They are systems failures. And systems failures have systems solutions.
Subscribe to The Signal
One issue per week. The biggest AI build of the week, the top five videos worth your time, and an operator's read on what actually matters. For builders who ship, not dreamers who scroll.
No spam. Unsubscribe anytime. Just The Signal.