Adobe Is Metering Creativity Like Gas — Without the Ledger to Prove It

Forty-seven generative credits. That is how many a freelance art director burned last week during a single product shoot — and I watched every one of them leave her account. Screen shared, Firefly panel open, a client nodding on the far side of the call. Background stripped. Sky replaced. A model's hand regenerated three times because the fingers kept melting into the sleeve. Four minutes, forty-seven credits, and not a single line of that consumption written anywhere a third party could verify. That number is what I carried into Adobe's earnings, not the headline revenue, not the seat count. The credits. Because a company that begins metering a creative act by the unit is quietly abandoning the subscription logic that built it — and replacing that logic with something crypto has been arguing about for a decade: pay-per-use resource accounting with no public ledger behind it. The protocol is cold; the evangelist is warm. Adobe has built the cold part and forgotten the warm part, the part that makes accounting trusted. Let me walk back. Firefly, Adobe's generative engine, sits inside Photoshop, Illustrator, Premiere, and a browser sandbox. Base access is free; sustained generation runs on credits, replenished monthly with a plan and sold in packs once you exhaust them. On paper it is elegant. In practice it is the most consequential pricing experiment in creative software since the shift to SaaS in 2013 — run by a company that has never had to defend a metered unit to the public. In early 2017, auditing ERC-20 implementations at a hackathon in Austin, I learned to distrust a number until I could trace where it came from. Gas made sense to me immediately, not as a fee but as a language: every unit of computation named a cost, and every cost was legible to anyone running a node. When I look at generative credits, I see the same primitive with the transparency surgically removed. There is a meter. There is no ledger. Chasing the frontier where code meets belief usually means finding the place where those two things agree; here they do not. The metering itself is not the problem. Inference costs money — silicon, power, cooling, the amortized training run. Anyone who has priced a GPU-hour knows the bill arrives whether or not the customer is delighted. And here is where Adobe's calculus gets interesting: charging per generation is a hedge against the very economics that make AI expensive. A flat subscription asks the vendor to absorb variable cost inside fixed revenue. A credit system pushes variable cost onto the user, capped only by what the user is willing to spend. From a margin-defense standpoint it is the obvious move. I have watched at least three DeFi protocols attempt the same maneuver through dynamic fee curves and get crucified for it — not because the mechanism failed, but because nobody could read it. That is the asymmetry. Adobe's users cannot see why fifty credits produced one image and one credit produced another. They cannot see the price of a regeneration against the price of a first pass. They cannot see whether the credits they are burning subsidize someone else's heavier workloads. In a permissionless system that opacity is a feature request. In a closed one it is a trust deficit that compounds every billing cycle. So the earnings number I actually care about is not revenue. It is the ratio nobody puts on the slide: credits consumed versus credits purchased beyond plan. If the base allotment is generous enough that most users never top up, Adobe has shipped a cost center disguised as a value-add and asked shareholders to applaud. If the top-up rate is high, the company has discovered a second revenue engine and the S-1 logic of the last decade needs revision. The truth almost certainly sits between, and the disclosure will almost certainly be thin. Two months of auditing nothing but public contracts taught me the interesting number is usually the one the filing does not require. Then there is the conversion problem, and it is worse than the pricing problem. Free Firefly lowered the barrier to generation to roughly zero — exactly what you want when building habit. But habit that never converts is a liability wearing a growth chart. The classic SaaS funnel assumes the free user lacks a capability the paid tier unlocks. Generative credits invert that: the free user has the capability and lacks the budget. You are not selling access; you are selling headroom. Selling headroom is a different business. It has more in common with cloud compute than with creative suites. And it has more in common with protocols than Adobe would like to admit. Here is the thing I keep circling. The creative industry does not need Adobe to open-source its weights to trust the system. It needs provenance. It needs to know which pixels were generated, which were captured, and who signed off on the transformation. Adobe already ships the scaffolding: Content Credentials, its implementation of the C2PA standard, a cryptographic manifest that travels with a file and records its edit history. That is, functionally, an attestation layer — a signed, tamper-evident chain of custody for media. Which is either the smartest thing Adobe has done this decade or the most under-milked asset in its portfolio, depending on how the company funds it. Content Credentials is the one place where Adobe's strategy touches the same substrate I spend my days on: a hash-linked record of authorship and modification. It is exactly the primitive you would design if you wanted to audit algorithmic output at scale. And it is currently marketed as a badge, a small icon in the corner of a preview window, rather than as infrastructure. I spent six months of my life during the 2022 winter mapping how separated consensus and execution layers prevent congestion. That exercise taught me the hardest part of any modular system is not the modules — it is the settlement layer everyone agrees to reference. C2PA wants to be that layer for media. Bitmap provenance, signed and chained, is a settlement problem in disguise. If Adobe treats it as a filter instead of a foundation, a startup will. The competition knows this. Midjourney out-generates Firefly on aesthetic range; Canva out-packages it on ease of use; the Stable Diffusion ecosystem out-develops it on extensibility. Adobe's edge is not the model. It is that the industry's files — PSD, AI, PDF — are already Adobe's dialect, and a designer deep in a pipeline does not switch dialects for a marginal quality gain. That lock is real and I do not dismiss it. But locks built on format compatibility are vulnerable to a world where the format stops mattering because generation happens in the cloud and only the flattened output travels. When the artifact is a prompt plus a manifest, the dialect is metadata, and metadata is the cheapest thing in computing to standardize. There is a second ledger question hiding underneath the consumer one, and it will matter more by 2027 than any individual plan tier. Enterprise buyers — ad agencies, media conglomerates, government communications offices — do not just want generation. They want an immutable record of what was generated, under whose credentials, against which model version, and whether the output touched restricted training data. That is a compliance artifact, and it is exactly the shape of a verifiable credential. I have spent the last two years connecting autonomous agents to decentralized identity rails precisely because this demand was inevitable: an agent that generates content on a brand's behalf needs a signed identity and a scoped mandate, or the brand has no way to prove authorship and no way to revoke it. Adobe has the pieces — Content Credentials plus its experience-cloud identity graph — to assemble this. Whether it does, or whether it sells models by the gallon and lets the provenance layer rot, is the whole question. The other number I will be hunting for is quieter. Inference cost is not a line item Adobe is obligated to break out, but it sits underneath gross margin, and it is hostile to the credit model in a way the credit model was designed to hide. Training costs amortize; inference costs recur with every token and every diffusion step. If premium plans bundle a monthly allotment generous enough to be a genuine value proposition, Adobe is effectively short volatility on its own inference bill — exposed to every efficiency gain and every slowdown in the GPU supply chain. When the top-up rate is high, that exposure is hedged by revenue. When it is low, the company is paying for enthusiasm out of margin. The line to watch is not credits sold; it is cost per credit delivered, and the gap between the two. This is the contrarian read on the earnings, and it runs against most of what the bull case is selling. The prevailing narrative says AI lifts average revenue per user, deepens stickiness, and defends the moat. I think that narrative is manufactured, the same way "liquidity fragmentation" was manufactured to justify a generation of new DEXs that solved nothing users asked for. Fragmentation was never the problem; the problem was that nobody wanted to seed a book against an unknown counterparty. It is the same pattern as a protocol launching with unsustainable emissions to fake usage: the metric looks like growth while the substrate is thin. Metering a generation without attesting to it is charging for gas on a chain with no explorer. You can do it, people will pay, and they will never fully trust it — and the moment someone offers them a receipt they can verify, the meter alone stops being a moat and starts being a complaint. Constructive pessimism says the risk is not that Adobe's AI fails. The risk is that it succeeds commercially while failing at transparency, and drags an entire industry's expectations about accountability down with it. The bet I would want Adobe to make is boring and expensive: publish the provenance graph, sign it, make it portable across competitors, and let the credits be argued over in public. That is the difference between a fee and a protocol. A fee is paid. A protocol is believed. Curiosity is the only leverage in DeFi Summer, and it is the only leverage here. The numbers drop this week. Watch the top-up rate, watch the margin commentary, watch for a single sentence about whether Content Credentials is treated as product or infrastructure. That sentence will tell you more about the next five years than any revenue figure. In the silence of the chain, we hear the future — and right now, Adobe's chain is silent about its own meter. Art is the glitch that proves we are human; the credits are the meter that pretends we are not.