The Missing Format.

For as long as I have been a communications engineer, I have noticed that we live on islands. The voice engineers go to voice shows, the messaging people go to messaging shows, video to video, mobile to mobile. Even the unified-communications folks, who you would think would be the exception, somehow only go to unified-communications shows. I do not blame anyone for it. Each of these silos is a life's work. But it came with a price. We became masters of our own vocabularies, like authors who spend a whole career considering single words and never once put them together into a sentence. And it is the sentences that carry the meaning. It is sentences, gathered up into conversations, that hold the most irreplaceable, sensitive, and powerful data a business has. Each kind of engineering contributes to the conversation. None of them owns it.

Here is what I eventually could not stop seeing. Almost everything a business makes ends up with a file. Images got JPEG. Documents got PDF. Spreadsheets got XLSX. Even your contacts got one, the humble vCard, quietly carrying a name and a number between applications that were never designed to talk to each other. That is the whole job of a file format: take something worth keeping, agree on a shape for it, and from that day forward it can be saved, sent, verified, and used by software nobody had to clear it with first.

Conversations never got one.

Sit with that for a minute, because it is stranger than it sounds. Think about what is actually inside a conversation. The decision. The objection nobody wrote down. The thing a customer mentions at minute four that never makes it into the ticket. I have come to think of this as dark operational data, and it is probably the best record a company has of what its customers actually want. At most companies, that record is not a thing at all. It is a pile.

It scatters the second the call ends. The recording lands in one system, the transcript in another. A notetaker writes a summary. A rep types a half-remembered version into the CRM. There is a sentiment score in an analytics tool, a compliance copy in an archive, a follow-up rotting in a queue somewhere. One conversation becomes a dozen pieces, every piece in a different system under different rules, with no thread tying any of it back to the thing that actually happened.

And here is the part that should bother you: you do not even own most of those pieces. You rent them. They live inside somebody else's product, in their format, on their terms, and the day you switch vendors most of your history stays behind. The conversation that built the customer relationship belongs, in practice, to whatever software happened to record it that day.

We have all put up with this for years, mostly because it never hurt enough to fix. You patched it with integrations and exports and a healthy amount of hope. I did too. That era is ending.

What changed is AI

Or more precisely, what AI does when you hand it something with holes in it. It does not stop and tell you the recording was only half there, or that the transcript dropped a minute. It fills in. It guesses, smooths the seam over, and hands you back something clean and confident with no mark anywhere to tell you how much it made up. “Customer satisfied. Case resolved. No further action needed.” Maybe. You cannot actually tell, because what the model read was never the whole conversation and was never traceable to what was really said.

Now add agents. Software is starting to hold conversations on behalf of the business, with customers and with other software, and we still have no agreed way to record what an agent said, under what authority, on whose behalf. Worse, you can no longer assume a recording is real, or unedited, or made by who it claims. At the IETF we like to say that mass surveillance is an attack; I would add that an untraceable recording is one too. Conversations now need what software supply chains had to learn after a few very expensive breaches: provenance. A record of where a thing came from and what was done to it along the way.

So the missing format went from mildly annoying to expensive. That is why this is a now problem and not a someday one.

The fix is the one engineers always reach for

Give it a file. Make the conversation one object. Put the people in it, the actual back-and-forth, the analysis that got run, the files that went around it, and the consent that is supposed to govern all of it, in one place. Make it portable, so it travels instead of being held hostage. Make it verifiable, so a person, a regulator, or a model can confirm it is real and that nobody quietly touched it. And let it carry its own rules, so when a customer says stop using my data, every copy knows, not just the one system that first hit record.

vConone conversation, one object

Parties Dialog Recording / transcript Analysis Attachments

consent

lawful basis · purpose limits · expiry · verifiable proof

That object exists. It is called a vCon, short for virtualized conversation, and it is moving through the IETF, the same organization behind the protocols the whole internet runs on. The open part matters more than it might sound. Most “new categories” are one vendor inventing a word so they can lock you inside it, and you can usually smell that a mile away. This is the other thing. The format belongs to everybody, which I would argue is the only way a standard for something as personal as a human conversation was ever going to earn anyone's trust.

PDF let a document be useful outside the application that made it. A vCon does the same for a conversation, with one difference: the next thing to open it probably will not be a person. It will be an agent.

Seventeen thousand calls, in one city, in one month

Let me make this concrete. We recently ran a trial with The Frontline Group and 211 of Kansas City. If you are unfamiliar with 211, it is a citizen helpline funded by the United Way. If you find yourself at risk of homelessness, or your power has been shut off, you call 211 and they tell you where to get help. Which means people are calling on the worst day of their lives, or at least a very bad one, and in one month, in one city, there were seventeen thousand of those calls.

No staff on earth can listen to them all, so for years those conversations effectively vanished the moment they ended. In the trial we carried each call as a vCon, and that month of calls stopped being a pile of recordings and became something the organization could actually work with. And here is the part that mattered as much as the listening: the protection. These are vulnerable people, calling about evictions and shutoffs and worse, and you cannot ask them to trade their privacy for help. Because each vCon carries its own consent and its own rules, and because you can derive a reduced copy that holds only what a given reader needs to see, the United Way and its partner charities could understand the ground truth of what people were asking for without ever exposing who was asking. The caller's identity stayed protected; the pattern of need did not.

The people deciding where help goes next could finally prioritize based on what every caller said, not the small fraction a human had time to review. These are people who were literally never heard from before. Now some of them are, and hopefully some lives are better for it. Sitting with seventeen thousand calls from people having their worst day was a dark day for me too. It also ended any doubt I had about why this format needs to exist.

None of this is theoretical

vCon is a chartered IETF working group with published drafts and sessions at the last several meetings, and the telecom and contact-center crowd is paying close attention. It is in production, too, from a BPO pushing hundreds of thousands of conversations a month through it to a financial institution generating millions of vCons a day. There is a community growing around it as well. Jeff Pulver, the fellow who gave us Voice on the Net back in the nineties, founded the vCon Foundation, whose membership now spans telecom, contact-center, and conversational-AI companies, and its meetups keep popping up: Boston, New York, Dallas, Atlanta, plus a monthly virtual meeting that draws people from all over the world. The people closest to the wire picked it up first, which is how these things always go. Everybody else catches up later.

It is worth remembering how this movie ends. Email had no standard, then it did, and a whole industry grew on the one that won. Same with the web. Same with instant messaging, where formats knocked around for years before the standard landed and the market took off around whoever was already building on it. The opening is always that quiet stretch before the standard goes official, while the words for the category are still up for grabs. That is about where we are right now.

The companies that win this decade will not be the ones sitting on the most conversation data. Everyone will have more of that than they can use. The winners will be the ones who can actually do something with it: across all their tools, consent still attached, in a form you would trust a model to read, and theirs to keep instead of rent. Which really just means treating a conversation like what it always was, something worth holding onto in one piece.

It needed a file. It was the last thing without one. Now it has one.