Digital identity in the AI era

When someone wants to know about you professionally, they increasingly do not visit your profile. They ask an assistant. That assistant produces a summary assembled from whatever it can find - an outdated bio, a conference listing, a company page from two pivots ago - and the person forms an impression from that.

This is already happening at scale and almost nobody has noticed, because the summarisation is invisible to the person being summarised. You never see the version of you that gets described.

Three eras of professional identity

• Documents. You wrote a CV. A human read it. Slow, but you controlled the text and knew who saw it.
• Profiles. You filled in a platform's fields. The platform controlled the ranking, the reach and the format, but the words were still yours.
• Summaries. A model compresses everything findable about you into a paragraph. You control neither the inputs, the format, nor the audience - and you do not know it happened.

Each era traded control for reach. The third trade is different in kind, because for the first time the representation is generated rather than retrieved. It can be confidently wrong.

The failure mode is quiet

If a search result about you is wrong, you can see it and correct it. If a model tells someone you still work at a company you left in 2023, you never find out, and neither does the person who decided not to contact you. There is no error message for a misrepresentation that costs you an opportunity.

An identity you cannot inspect is an identity you do not own.

What owning it requires

A representation you control has to satisfy four properties, and most things marketed as digital identity satisfy one or two:

1. Authored. The source material is what you wrote, not what was scraped about you.
2. Grounded. Answers come from that material with a traceable citation, so a wrong answer is a findable bug rather than a plausible invention.
3. Permissioned. You decide which audience sees which layer, and you can revoke it.
4. Observable. You can see what was asked and what was answered. Without this, the other three are unverifiable promises.

Observability is the one people skip and the one that matters most in practice. A system that answers on your behalf without telling you what it said is not representation, it is exposure.

The practical position

You cannot stop models from describing you; that ship sailed with the public web. What you can do is publish an authoritative, structured, machine-readable version of your professional self, so that the accurate account is the one that is easiest to find and cheapest to use.

That is the pragmatic case for an AI twin, stripped of any futurism. Not because representing yourself to machines is exciting, but because the alternative is being represented badly by default, and never finding out.