A brief history of extractive software

There are two kinds of software. One demands that you feed it—your time, your typing, your discipline—before it gives anything back. The other reads the work you’re already doing, and gives back without asking. Call them extractive and ambient. Once you see the split, you see it everywhere. And you notice that every time we’ve been offered the choice, we’ve chosen ambient.

The tell

Extractive software runs on a toll. Its value depends on humans continuously feeding it structured input that exists nowhere else: the sales rep retyping a conversation as six fields and a dropdown, the engineer updating a ticket to describe work that already happened, the consultant reconstructing her week for the timesheet. The data is secondhand testimony about work that happened somewhere else, transcribed by a witness who would rather be doing anything else.

Ambient software reads the primary source. It derives its value from the signals the work already produces—the exhaust—without changing how the work gets done.

Here’s a quick diagnostic for any tool you use: when you and the software disagree about whose time matters, who wins?

We mean “extractive” in both senses, by the way. Economists use the word for institutions designed to pull value out of people without giving much back—which is a fair description of a system that asks your team for hundreds of hours of data entry so that someone else can read a dashboard. Extractive software mines its users. Ambient software mines the work itself.

We’ve been here before

For most of history, keeping a record was a separate act from doing the thing. Someone had to transcribe reality into the system, and the transcriber was always the weakest link. Then, over and over, the transcription fell away:

  • Nineteenth-century science ran on assistants logging instrument readings at fixed hours—until self-registering thermographs and barographs drew the curve themselves, all night, without complaint.
  • James Ritty’s 1879 cash register—patented as “Ritty’s Incorruptible Cashier”—made the record a byproduct of ringing up the sale. Bookkeeping stopped being a thing bartenders did and became a thing that happened.
  • Inventory went from hand counts to barcodes at the point of sale to RFID. The data became exhaust of the transaction itself.
  • Toll booths became E-ZPass became open-road tolling. The road stopped interrupting you in order to measure you.
  • Doctors replaced patient glucose diaries with continuous monitors, because self-reported data was always part fiction. Any doctor will tell you the diary said what the patient wished were true.

Notice what happened to accuracy each time: it went up. Removing the human transcriber didn’t degrade the record. The transcriber was what degraded the record.

Then software forgot

Here’s the strange part. Software—the technology that should have finished this job—spent its first fifty years reversing it.

The reason was mundane: databases could only eat structure. A computer couldn’t read a conversation, so someone had to retype the conversation as fields. And so business software deputized entire professions as transcribers of their own work. The CRM is the canonical case: a shadow copy of reality, maintained by hand, forever out of date.

Jonathan Grudin diagnosed the failure back in 1988: collaborative systems fail when the people doing the work are not the people getting the benefit. The rep types; the VP gets the dashboard. That misalignment is structural, which is why thirty years of CRM adoption initiatives, mandatory fields, and end-of-quarter data-hygiene emails never fixed it. Extractive software doesn’t fail because people are lazy. It fails because it’s a tax levied on one group to benefit another, and people are rational about taxes.

Where ambient already won

Wherever a domain’s signals happened to be born structured—clicks, transactions, GPS pings—ambient software showed up and quietly ended the argument:

  • Yahoo paid humans to catalog the web into a directory. Google read the link structure the web had already produced. The extractive approach wasn’t just worse; it was incapable of keeping up with reality.
  • Quicken asked you to keep a ledger. Mint read the transactions that already existed.
  • Bug report forms deputized users as QA. Sentry and Crashlytics let the crash report itself.
  • Traffic data came from road sensors and radio helicopters. Waze read the phones already in the cars.
  • Nielsen mailed families paper diaries. The set-top box just watched.
  • Sales reps typed call notes from memory. Gong recorded the call.

You already prefer ambient software. You chose it everywhere you were given the choice. Extraction survives only where the signal is locked in prose—conversations, threads, documents—where no machine could read the primary source.

The constraint just expired

That constraint was real for fifty years, and it’s the only honest defense of extractive software: extraction was never anyone’s preference. It was a workaround for machines that couldn’t read.

Large language models ended it. Turning unstructured work into structured records is now cheap, which means the largest trove of work data in existence—email, where deals are actually negotiated, commitments actually made, and relationships actually maintained—just became machine-readable in place. The contacts, the organizations, the tasks, the history: it’s all in there, in the primary source, waiting to be read rather than retyped.

When the constraint that justified a toll disappears, the toll doesn’t survive on nostalgia.

What ambient is not

Three honest objections, because this frame is too easy to abuse:

Ambient is not surveillance. Bossware is ambient capture with the beneficiary inverted: observing workers for someone else’s benefit. Grudin’s test cuts both ways—if the person being observed isn’t the primary beneficiary, “ambient” is just extraction of a darker kind. Any ambient system worth trusting needs hard privacy boundaries: private stays private, and sharing is deliberate.

Some structure is judgment, not fact. A deal stage is an opinion. An ambient system can propose the judgment, but it should accept human intent as enrichment—never demand it as a toll.

The toll forced reflection. Typing the call note made you think about the call. This is the best argument for extraction, and we have some sympathy for it. But the CRM graveyard of empty fields suggests most people never paid the toll anyway. A ritual nobody performs isn’t reflection; it’s guilt.

Where we obviously land

You can guess where this goes. We think the CRM is the last great extractive holdout, and that email is where its data lived all along. Carom reads your team’s email in place and reconstructs what the CRM always wanted—contacts, organizations, threads, files, tasks—shared across your team, with private things kept private, and with nothing to enter or maintain. Value at minute one, not after months of discipline.

The best way to judge an ambient tool is to watch it work: create a demo account, or reach out if we can help.