trust is the reserve currency

bloomberg ran a story this spring calling arr "the least-trusted metric of the ai era." the trigger: one of the loudest ai startups of last year admitted its famous revenue number was inflated. and the market's reaction was the interesting part: a shrug. nobody was surprised, because everyone had already stopped believing numbers.

demos are generated. reviews are generated. portfolios are generated. faces on video calls are generated. work samples, benchmarks, case studies: generated. we have crossed, quietly, into an economy where anything presentable can be produced for nothing.

that breaks something deeper than content. it breaks proof.

how markets actually run

markets don't run on information. they run on costly signals. a degree meant years you couldn't fake. a portfolio meant work you must have done. a polished product meant a team and money behind it. a warm intro meant someone staked their name. the signal carried information precisely because faking it cost more than having the real thing.

generative ai sets the cost of faking nearly every signal to zero, all at once. the signals don't become lies; they become noise. an impressive demo now carries almost no information about whether the company behind it is real. a beautiful brand tells you nothing about whether the product works. that's new.

and when signals die, markets don't stop. they reprice the one thing that can't be generated: trust with history behind it. that's what a reserve currency is: the thing everything else gets priced against when nobody believes the local paper. trust is becoming that, and you can watch the repricing happen in real time.

the tell: machines got trust infrastructure first

here's the strongest evidence that trust is the binding constraint: the agent economy, the newest economy we have, is building trust infrastructure before it builds almost anything else.

the ietf is standardizing cryptographic identity so an ai agent's requests arrive signed and verifiable. visa shipped a trusted-agent framework. amex now sells purchase protection for registered agents: an insurer literally underwriting the behavior of software. the card networks are donating agent-payment protocols to standards bodies. some of the sharpest infrastructure teams on earth are all aimed at one question: how does a stranger trust a machine?

meanwhile the human economy is converging on the four signals that survive when everything else is fakeable:

• time. a track record laid down in public, before the claim needed to be true. you can generate a portfolio tonight. you cannot generate having shipped in public for five years.
• skin in the game. liability, warranties, staked reputation. don't tell me you're good; show me who pays when you're wrong. watch guarantees become the default trust interface everywhere.
• verification. cryptographic provenance, audits, signed identity, for people and companies as much as for agents.
• vouching. a human with a name staking it on another. referrals are eating resumes for exactly this reason: the resume got cheap to fake and the vouch didn't. (my read from what i see in tech hiring; i might be over-indexing on my corner of it.)

trust is distribution now

follow the money one step further and this connects to the thing i care most about. adobe's data this year shows ai-referred shoppers now convert better than search traffic, a complete inversion from a year earlier. people increasingly buy where a machine they trust points them, and trust is aggregating: into a few assistants, a few brands, a few names.

which means distribution and trust are collapsing into the same thing. reach without trust is spam, and reach is now free, so it's worthless. trust with even modest reach converts absurdly well. the entire game of getting a product into the world is becoming: accumulate trust, then point it.

and trust has brutal compounding mechanics. slow to build, instant to destroy, impossible to buy at market: it can only be earned at history-speed. that makes it the best moat arithmetic left in software. it's also why the backlash to big brands shipping obviously generated ads mattered more than it seemed. a brand is a trust battery charged over decades, and the first companies caught filling the battery with slop discovered the discharge is immediate.

every era says this

"trust was always the currency. every era says this." true. two things are new anyway.

first, coverage. there was always low-trust commerce that worked fine, buying batteries from a stranger, because the artifact verified itself: the batteries worked or they didn't, and the stakes were low. that floor is disappearing. when reviews, photos, storefronts, and sellers can all be synthetic, even battery-grade transactions need a trust rail underneath, which is exactly why the payment networks are racing to build one.

second, the hedge that keeps me honest: verification technology might catch up. if cryptographic provenance gets stamped on everything by default, proof becomes cheap again, and this essay ages badly. i'd bet against it, because forgery and verification are an arms race and the forger moves second. but i hold that one loosely.

what i'd do about it

for a builder, the strategy writes itself, and it's uncomfortable because it's slow. put your name on things. ship in public. keep a visible record of being right, and more importantly, of correcting yourself when wrong: one honest public correction buys more trust than ten wins. take liability when you can afford to. and guard the asset like it's the balance sheet, because it is. you can rebuild a product in a weekend now. nobody has ever rebuilt a reputation in one.

money, at bottom, is a technology for trusting strangers. we are rebuilding that technology for a world where strangers can be manufactured.

in that world, you sell the one thing that can't be faked twice.