What tl;dv Gave Me
This week I connected my meeting recorder (tl;dv) to Claude and gave it one job: go through every call I'd had in the last month, around eighteen of them, and find every promise I'd made to someone that never made it into a single plan or task tracker. Not "summarize this call." A question that ran across every conversation at once, the kind you can't answer by opening any single recording, because the answer is spread across all of them, not inside any single one.
The agent pulled the full transcript of each meeting, read all eighteen in parallel, used my personal knowledge base to understand what each one was about, cross-referenced them against my own plans and task tracker, and handed back a prioritized list of commitments that lived only in the conversation and nowhere else. Most of it only confirmed what I already knew, that the thing was handled, on my side or my team's. A few weren't, and those few alone made it worth doing. Start to finish, a few minutes. I never opened the tool's interface, never scrubbed a timeline, never reread a transcript by hand.
I'll keep the actual list to myself, since it's about real people and live conversations. The method is the part worth sharing.
tl;dv Got There Early
In May I argued that the user interface is collapsing into two surfaces: a REST API that agents talk to, and a single window where people just say what they want. In June, Apollo became the first clear example I wrote about, exposing its B2B platform through an MCP connector and publishing the numbers that showed people preferred it.
tl;dv is the third one I'm writing about, and it was actually early. It shipped its MCP server in April 2025, built on the Model Context Protocol, more than a year before I wrote that forecast. I connect it to Claude once, and after that the agent can list my meetings, pull any transcript, read the summaries, and reason over all of it at once. Now I don't open tl;dv at all. Claude calls it in the background while I stay in the chat window I already use.
The Real Unlock Is Having Everything in One Place
The connector alone didn't answer my question. To tell whether a promise had been kept, the agent needed three things together: the transcripts, for what I said; my knowledge base, so it knew what each meeting was even about and who the people were; and my plans and task tracker, to check each promise against. No single source answers it. The answer only exists where all three overlap, and for me the place they overlap is Claude. A single integration is just one feed into it, and it gets more useful with every other feed that joins it.
Why It Matters
This matters for two reasons.
First, it answers questions the old interface never could. Finding one pattern across thirty hours of meetings used to take an afternoon, so I never did it. Now I just ask, and the answer comes in a minute. It doesn't only make old work faster. It lets me ask questions I could never ask before.
Second, it is easier than clicking. People pick the way that takes less effort, and saying what you want takes less effort than learning an interface. For a while this felt unusual. It is becoming normal, one tool at a time. The old screens don't go away, but they stop being the first place you go.
What It Means If You Build Products
The lesson is not "add a chatbot." tl;dv exposed its capability and its data through a protocol so any agent could drive it. Two questions for your own product: can people use it without your UI, and can they ask it things your UI never planned for? Once your product is reachable by an agent, people will ask questions you never built a screen for, and get answers anyway. For most products that is the most valuable thing you can offer.
P.S. Yes, this is me circling the same idea from one more angle, exactly as I said I would. The angle keeps being worth it.
Originally published: tl;dv Steps Into the Future of User Interfaces — Alex Rezvov's Blog
