---
title: "How do you turn a year of personal writing into a talk with an agent?"
published: 2026-10-19T14:00:00+03:00
updated: 2026-10-23
lang: en
canonical: https://ofershap.github.io/en/posts/agent-talk-from-content-archive/
original: https://ofershap.github.io/posts/agent-talk-from-content-archive/
---

# How do you turn a year of personal writing into a talk with an agent?

Collect the raw material into one file, write a brief with audience, goal, and time, ask for an outline before slides, and check every claim against the source before approving a slide. The agent is excellent at synthesizing existing material, and the responsibility for accuracy stays with you.

## In depth

### Collect the material before asking for output

My agent got a file with a year of my LinkedIn posts. That's what allowed it to find a real line of thinking instead of a generic deck about AI. Without your own material, you'll get a talk anyone could give.

Example: one tidy file sorted by date, with a link to every source. Thirty full posts beat 200 summary lines.

### Write a brief of audience, goal, and time

"Prepare a talk for me" returns something generic. A brief that works answers three questions: who sits in the audience, what should they do differently after the talk, and how many minutes you have. Mine was small business owners, 45 minutes, and one goal: that they go home with a first task they can give an agent.

### Ask for an outline before slides

The cheapest time to fix a talk is before it's a deck. Ask for an outline first: an arc, a core message per part, and at least one demo. After the outline is approved, ask for slides and a prep guide.

A prep guide is worth no less than the slides: what to say on each slide, how much time per part, and where to stop for questions.

### Check every claim against the source

The agent phrases confidently even when it's half-accurate. Every number, date, or tool name on a slide needs backing: a link to the source in the raw material, or a current official document. An unsupported claim leaves the deck.

Example: if a slide says "companies adopting agents save 30% of development time" and there's no source, replace it with data you can show, or with a sentence about your own experience only. Anthropic's long-context prompting guidelines also recommend giving concrete examples and asking for a quote from the source before synthesis, which is exactly this check.

### Prepare a short emergency version

Half an hour before the conference, things break: the projector, the power, the network. Keep a ten-minute version: three points, one example, one message. If everything collapses, you still deliver value.

## From my experience

In my post about the talk built from WhatsApp I told how I gave an agent a file with a year of my posts and asked for a talk on turning an entire company AI-native. I got a 15-slide deck and a prep guide, including the demo. Here is the process that worked, and what you must check before getting on stage with it.

## Frequently asked questions

### How long does a process like this take?

The production itself is fast. Most of the time goes to checking against sources and rehearsing, and you can't shorten those without paying in quality.

### Can the agent deliver the talk too?

You can't delegate the responsibility. It's your talk, your claims, and the audience asks you.

### What if my posts aren't rich enough for a talk?

Start with the question the audience keeps asking you. A small archive with real experience is worth more than a big archive of generalities.

## About the author

Ofer Shapira builds and leads AI in engineering teams, and writes about agents, decisions, and workflows you can verify.

Last updated: 2026-10-23