On September 10, Dee Goens took the CEO seat at Zora, replacing co-founder Jacob Horne, who had held the position for more than six years. The announcement went up on a Wednesday. Horne is heading into a new chapter and, per Goens, will not stray too far from crypto โ the sentence founders write when they are leaving the building but keeping the keys.
The succession is the headline. The number underneath it is the story.
Zora has run layoffs this year. Headcount is now under ten. The operating model is being rebuilt around AI. Ten people, one live Layer 2, one token, an NFT mint rail, a creator economy, a social feed, and a sequencer that has to produce blocks whether or not anyone is awake to watch it.
I have audited small teams before. The Mumbai sprint in 2017 was a team of six sitting on an integer overflow in their liquidity pool logic. I sent the pull request with the proof attached, they merged it two days before mainnet, and roughly two million dollars of early deposits did not evaporate. Six people was fine there, because their contract surface was one file. Headcount is not a proxy for capability. It is an exact proxy for the number of things you can do at the same time.
To read this correctly, separate the protocol from the company that ships it.
Zora launched in 2020 as an open NFT protocol built around the ask โ a minting primitive that let a creator price work on a curve instead of an auction, letting the market discover value continuously rather than in a single cliff. The contracts were public from day one. The front end was the business.
Then came 2023 and Zora Network, an OP Stack rollup with its own sequencer and its own fee market. Then came the migration of attention. Zora stopped being a marketplace and became a place. Posts became mintable. Mintable posts became tradeable. Tradeable posts became a social graph with a price printed on every edge. The token arrived in 2025 and gave that graph a unit of account.
Every one of those steps added surface area. Every one of those steps also added a dependency on a team. Contracts do not resign. Contracts also do not write their own upgrade proposals, watch their own dashboards, or answer their own incident pages.
The token is the other half of the arithmetic. A treasury funded in a bull market pays for a bull-market headcount, and when the market turns, the same treasury has to service the same obligations at a fraction of the value. Cutting to nine is not a philosophy. It is division: nine people on a two-year runway beats forty people on a five-month one, and the layoffs are the sound of someone doing that sum correctly and early rather than late and loudly.
So the question is not whether Zora is dying. It is narrower and more answerable. Which functions were being performed by people, which of them now have nobody, and which have been handed to a model.
Start with the part nobody prices.
Protocols at Zora's scale run upgrade authority through a multisig โ a Safe, 3-of-5 or 4-of-7, some threshold in that band. The signers are employees. When you go from a few dozen people to nine, you either rotate the signer set or you accept that the same three humans handling on-call are the same three humans who can move the contract. The threshold is a protocol parameter. The signer set is a personnel fact. When personnel changes, a protocol parameter changes without a governance vote.
That is not a Zora-specific criticism. It is true of nearly every protocol with a team-held admin key, and it is the single most under-checked item in the asset class. The signer set is public if you know where to look โ read the Safe, trace the owners, check the timelock delay. Almost nobody does. I would want that checked this week, before the next upgrade proposal lands, because a proposal authored and approved by three people who also wrote the code is not a governance process. It is a commit with extra steps.
The token holders have one lever here, and it is thin but not zero: a governance forum, a snapshot, and the reputational cost of ignoring both. That is worth less than a timelock and considerably more than nothing. It is the only instrument the community holds if the signer set drifts while everyone is looking at the price chart.
Then there is incident response. A sub-ten team cannot run a 24/7 rotation. They run best-effort. In a bear market the attacker pool narrows but the unemployed searcher pool widens, and best-effort against a widening pool means a four-hour detection window on a drain. Speed is a feature, not a bug, until it breaks. Zora's speed is its sequencer cadence. Its break condition is a human who is not awake.
Now the AI operating model, which deserves to be taken seriously rather than reflexively mocked.
Mechanically, AI-driven operations at a protocol of this shape means a handful of concrete things: automated moderation of an onchain feed, spam and duplicate detection on a minting rail, metadata generation for collections that arrive with no description, support triage, and routing of creator payouts. Some of that is genuinely better done by a model than by a person. Filtering is the clearest case. The kill rate on a public mint rail is brutal โ wash mints, copy-paste collections, refund exploits, gas-griefing spam. A deterministic ruleset fails on all four because the adversaries read the ruleset. A model that tracks the current spam pattern is a real upgrade. That is not hype. That is maintenance.
Curation is where maintenance becomes governance.
In 2021 I curated an exhibition in Mumbai's creative district โ fifty works from decentralized artists, royalty splits negotiated contract by contract, ten percent of secondary sales retained by the people who made the work. I went in thinking curation was a ranking problem. It is not. It is a values problem wearing a ranking interface. Curation is the new consensus mechanism, and the moment you delegate it to a model, you have delegated the question of what counts as legible culture. Art is the metadata of human emotion โ and a model trained on engagement can only ever recover the metadata of attention. Those are different distributions. They are not close.
There is a version of this that is fine. A model that filters spam out of a feed is a janitor. A model that decides which feed a post lands in is an editor. Zora's public materials will not tell you which one was built. The observable will: watch whether the top-of-feed distribution concentrates or disperses over the next two quarters. Concentration means an editor. Dispersion means a janitor. That test is public, and it costs nothing to run.
The third thing nine people have to keep alive is the chain.
An OP Stack rollup is not a set-and-forget artifact. Someone owns the sequencer. Someone watches the proposer. Someone notices when the batcher stops posting. Someone pays the blob bill, and someone decides what happens when that bill prints red for a month straight in a market where the token is down seventy percent from listing. That last one is where I have watched more chains die than anywhere else โ not in an exploit, in a spreadsheet.
Here is the argument I have been making since the 2022 audit, the one where I pulled over a hundred thousand transactions across Optimism and Arbitrum and wrote up the state root calculation inefficiencies I found. Data availability is the most oversold line item in rollup economics. The overwhelming majority of rollups do not generate enough data to justify a dedicated DA layer. They generate enough to justify blobs. A chain with Zora's throughput profile does not need a DA committee, a DA token, or a DA narrative. It needs to post blobs to Ethereum and get on with it โ or, better, to stop being a chain at all and settle back onto the L2 it forked from.
Running your own sequencer buys fee revenue and control. It costs a permanent operational liability that scales with uptime expectations, not with headcount. Nine people can absolutely run an OP Stack chain. What they cannot do is run it and run a social product and run a token and run creator support and rotate their own multisig. Something on that list is currently under-served. You do not have to guess which. Watch the sequencer uptime page, the upgrade cadence, and the support latency. The under-served item will show up as the flat line.
The last structural note is about liquidity, because Zora's product model is the purest example of a narrative I have pushed back on for years.
Every mintable post is its own market. Every market is thin. The reflexive read is that Zora suffers from liquidity fragmentation and needs routing, aggregation, shared pools, some new primitive โ and conveniently, every such primitive arrives with a token attached. The fragmentation is not the disease. It is the shape of the asset. Micro-markets over cultural objects are supposed to be illiquid, because the objects are unique and the demand is bursty. Aggregating them does not deepen them. It relocates the thinness to a layer where it is harder to see. I do not predict trends; I ride the volatility. But I do not mistake a volatile asset for a broken market, and I do not buy a product because someone rebranded thinness as fragmentation.
Now the part where I disagree with the consensus read on this whole announcement.
The reflexive interpretation of layoffs, a sub-ten team, and an AI pivot is decline โ a protocol winding down under a caretaker CEO. That read is lazy, and it is wrong in an interesting way. A lean protocol team is structurally less able to capture its own protocol. Nine people cannot credibly run a values-extracting operation across a network they also police. Not because they are virtuous. Because they do not have the headcount. The protocol is neutral; the user is the variable. The smaller the operator, the less the operator can bend the thing it operates. There is a real chance the sub-ten Zora is more credibly neutral than the forty-person Zora ever was. And Goens is a co-founder stepping into the seat rather than an outside operator โ continuity of intent, not a hired hand managing a wind-down.
The blind spot in that optimistic read is upgrading. A small team does not stop upgrading. It upgrades faster, because consensus inside nine people is one thread. The risk profile inverts: not stagnation, but velocity without review. The 2017 Mumbai lesson is that review is the only thing standing between velocity and a two-million-dollar hole, and review requires people who are not the author. Nine people is not enough to guarantee that separation. That is the exposure nobody has priced, and it only becomes visible in hindsight.
Yields are transient; infrastructure is permanent โ and this market is currently working out which parts of the last cycle were infrastructure and which were payroll.
The question I would put to any Zora holder this week is not whether Goens is the right leader. It is simpler. If the last two engineers leave in March, what still works? If you can answer that from public artifacts โ the Safe, the timelock, the sequencer, the blob postings, the commit log โ you own infrastructure. If your answer needs an org chart, you own a company, and you should price it like one.