Last week I had two calls that gave me completely opposite takes on the same topic.
On Monday, someone told me AI clones are having a real moment. They've been building digital versions of themselves for content and sales outreach, and the results are working. The market's not as resistant as people expected.
On Thursday, someone else said authenticity is what's going to matter most over the next few years. AI-generated content is getting easier to detect and audiences can feel when something isn't real. The people building a genuine presence are going to win.
Both conversations made sense in the moment. Both people were smart and specific. I walked away from each call nodding.
I never connected them.
An AI agent caught it.
The Weekly Synthesizer
I built an agent I call the weekly synthesizer. It runs every Friday and reads every transcript I generated in the last seven days — calls recorded through Lindy, Granola, Otter, and a few other tools.
Most weeks that's somewhere between 10 and 20 conversations. A lot of transcript data. The synthesizer pulls all of it and generates a single document with three things:
Recurring themes. What topics kept coming up across multiple conversations this week? If three different clients all mentioned the same pain point, that's a signal worth noticing — even if none of those calls were directly related.
Outstanding action items. What did I say I'd do across my calls this week, and haven't done yet? A promise on Monday, a commitment on Wednesday, a follow-up I said I'd send on Friday — all of them in one place rather than buried in separate call notes.
Contradictions. Where did I hear opposing signals on the same topic from different people?
The AI clone vs. authenticity flag landed in that third category. The agent surfaced it with a note: these two conversations contain contradictory signals on the same topic. Worth investigating.
I would not have caught that on my own. Those were separate meetings on separate days with separate people. My brain had filed them as separate conversations. There was no reason I'd ever put them side by side.
Why Most Transcripts Die
I had a conversation with Evan Baehr a few months ago where he made an observation I keep thinking about. He realized that 95% of his Granola transcripts were just sitting there, never turned into documents. They were way richer than email — actual full conversation records — but he wasn't using them. The information existed but wasn't accessible.
The same thing happens with most people's meeting notes. The auto-generated summary gets created, maybe you glance at it once, and then it lives in some folder you never open again. The information technically exists. But it doesn't compound. It doesn't connect to anything else.
The weekly synthesizer is the layer that makes transcripts useful instead of just archived. Not by summarizing each meeting individually — you can do that easily with a basic Lindy agent. But by reading all of them together and finding what's invisible when you look at them one at a time.
What Gets Surfaced
The contradiction example is the one that surprised me most, but the recurring themes function is probably more consistently useful.
One week the synthesizer noted that four different conversations touched on the same topic: people feeling overwhelmed by the number of AI tools available and not knowing where to start. Those were conversations with a client, a podcast guest, someone I met at a networking event, and a person who emailed me a question. They had nothing to do with each other. But the agent read all four and flagged the pattern.
That became the topic of a workshop I built the following week. The signal was already there — I just needed something to surface it across conversations I'd stopped thinking of as related.
Action items are the other big one. I'm a fairly reliable note-taker during calls, but the notes live in different places. The synthesizer pulls all of them into one list, sorted by what I've committed to doing and what's still open. It's not a replacement for a task manager, but it's a good weekly audit of whether I'm following through on what I say I'll do.
How to Build One
The basic version of this agent doesn't require a complex setup. You need:
A consistent transcript source. Pick the tool you use most — Granola, Otter, or your AI meeting notetaker of choice — and make sure every call uploads to a folder your agent can access. The most common mistake is having transcripts scattered across five different apps. Pick one or two and route everything there.
A weekly trigger. Schedule the agent to run Friday afternoon or whenever your week naturally closes out. Give it access to the transcript folder.
A clear synthesis prompt. The prompt matters more than the tool. Tell the agent explicitly what you want it to look for: themes, action items, and contradictions. Without that structure, you get a summary of summaries, which isn't useful.
An output you'll actually read. Mine goes to email. A Google Doc also works. Wherever your weekly review happens, put the output there.
The more consistent your transcript pipeline, the more useful the synthesizer gets. In weeks where I have fewer calls, the output is less interesting. In busy weeks, it earns its keep.
The Broader Point
Most AI tools give you a better version of something you already had — faster email, cleaner notes, quicker research. The weekly synthesizer does something different. It creates a capability you literally didn't have before: the ability to read all your conversations at once and find patterns across them.
No human can do that in real time. You're in one meeting, then another, then another. The connections between them only exist in hindsight, and most of us don't have the bandwidth to make them manually.
The agent does. And it turns out there's a lot worth noticing in those connections.
Thanh Pham is the founder of Asian Efficiency and an AI consultant based in Austin, TX. If you want to build your own meeting intelligence system, the 4-Day AI Sprint walks through the fundamentals.
