Onchain everything, Aave's risk purge, and the self-custody bill: July 26-August 2 (2026)

The week in one sentence

Crypto's biggest claims got more operational this week. Brian Armstrong put Coinbase in the path of every onchain asset and future agent-to-agent payment, while Stani Kulechov cut low-adoption Aave markets before pushing further into securities finance. Hayden Adams showed Uniswap moving toward yield and fee capture. Changpeng Zhao made the bill for self-custody explicit: users gain control, but they also inherit the patching, recovery, and incident-response problem. 1 2 3 4

Leader	Public statement in the July 26-August 2 window	What to check next
Brian Armstrong, Coinbase co-founder and CEO	Every asset will move onchain; he also asked when agent-to-agent payment volume will exceed human-to-human volume. 1 5	Actual agent payment volume, authorization failures, and who controls the rails.
Stani Kulechov, Aave founder and CEO	Aave is removing low-adoption reserves and winding down six deployments, affecting his reported $98.1M of supply and $15.6M of debt. 2	Whether risk-adjusted growth improves after the reduction, and whether securities finance produces real demand.
Hayden Adams, Uniswap founder	Uniswap Earn now accepts USDC, USDT, and ETH deposits in the app and wallet; Adams also reported a 106,000 UNI burn day. 6 3	Product usage, the source of yield, and whether fee capture persists beyond launch attention.
Changpeng Zhao, Binance founder	He backed self-custody while warning that a patched bug may leave old wallets exposed, and that the burden remains with the user. 4	Recovery paths, support responsibility, and the real tradeoff between venue risk and user risk.
Vitalik Buterin, Ethereum co-founder	His Diamond iO article describes obfuscation that is less theoretically remote than older constructions, but still infeasible today. 7	Whether security analysis and optimization turn the construction into a usable primitive.
Arthur Hayes, BitMEX co-founder and Maelstrom CIO	Hayes read the Fed as permission for buying distressed DeFi tokens with product-market fit. 8	Which tokens, what monetary mechanism, and whether the claim survives without a rising market.

Armstrong: Coinbase wants the rails, not just the venue

Brian Armstrong's July 30 earnings-day post was a product thesis disguised as a market thesis. He wrote that "every asset on earth" - stocks, bonds, commodities, and real estate - will move onchain, and that a small number of companies will own the rails. He positioned Coinbase as one of those companies. In the same post, he reported that more than 90% of agentic stablecoin transaction volume was on Base, that Coinbase's crypto trading-volume market share reached 10.3%, and that prediction-markets revenue rose 106% quarter over quarter. These are Armstrong's company-reported figures and forecasts, not independent measurements. 1

The follow-up was more revealing than the slogan. On July 31, Armstrong said AI was helping Coinbase update financial infrastructure while Coinbase built infrastructure for AI agents. On August 2, he defined the "Agentic Finance flippening" as the point when agent-to-agent payment volume exceeds human-to-human payment volume, then asked readers to predict when it would happen. He has supplied a testable definition, but not a date or a volume forecast. 9 5

The investor question is not whether every asset will move onchain. Armstrong's post assumes that outcome. The useful questions are narrower:

Does agent payment activity become a measurable revenue stream, or stay a wallet-and-infrastructure narrative?
Can Base's reported concentration in agentic stablecoin volume coexist with credible interoperability and recovery paths?
If a few firms own the rails, do they earn from custody, settlement, distribution, trading, or all four?

The August 2 post had 4,047 likes and 454,816 views in the retrieved snapshot. That is attention, not adoption, but the definition gives future editions a cleaner metric to track. 5

Loading content card…

Stani: shrink the risk surface before expanding the addressable market

Stani Kulechov, Aave's founder and CEO, announced a deliberate contraction on July 30. Aave planned to deprecate 50 low-adoption asset reserves, wind down deployments on Sonic, Scroll, zkSync, Metis, Soneium, and Aptos, and retire 21 matured Pendle PTs in favor of newer maturities. Kulechov said the changes affected $98.1 million in supply and $15.6 million in debt. The figures are his snapshot of the affected markets. 2

He then closed off the easiest interpretation. The wind-downs, he said, should not be read as a judgment on any L1 or L2. The stated goal was to reduce Aave's operational, technical, and economic risk surface so the protocol could grow high-value markets and expand into securities finance. Kulechov still described L2s as important to Ethereum's user experience, and pointed to Stable Vaults using an L2 as their accounting layer. 10

That is a more disciplined growth pitch than the usual chain-count metric. Aave is choosing fewer markets, fewer long-tail assets, and more attention on collateral and borrowers it considers worth the operational burden. The tradeoff is visible: less surface area for incidental liquidity, more concentration in the markets that remain.

For builders, three checks matter:

Risk accounting: Do the removed reserves lower oracle, liquidity, and maintenance risk by more than the lost supply and debt reduce revenue?
Securities finance: What collateral, legal wrapper, price feed, and liquidation process would make tokenized securities lendable?
Distribution: Can Stable Vaults and institutional partnerships bring users without recreating the long-tail complexity Aave is now removing?

The reserve post carried 1,513 likes and 286,143 views in the retrieved snapshot. That does not prove the policy is correct. It does show that Aave's public story has moved from "deploy everywhere" toward an explicit risk budget. 2

Loading content card…

Hayden: Uniswap adds a product layer to the trading venue

Hayden Adams's posts this week connected two changes that are easy to read separately. On July 31, he said users could deposit USDC, USDT, or ETH to earn yield directly in the Uniswap web app and wallet. The post did not specify the yield source, rate, or risk controls, so those details should not be inferred from the launch announcement. 6

On July 30, Adams said a 106,000 UNI burn was the third-highest burn day ever and estimated that amount at $170 million of annual UNI burn at the current price. He described it as the highest burn on a "normal" day, while noting that the two larger days came during major market moves. The annual figure is a run-rate calculation based on the day's burn and current price, not a guaranteed future cash flow. 3

Together, the posts show Uniswap trying to make the app useful between trades and make protocol activity legible as token value capture. That is a stronger commercial posture than a DEX that only waits for users to swap. It also creates new tests: yield products need transparent sources and loss boundaries, while burns need durable fee-generating activity rather than a single burst of volume.

The next evidence is operational:

repeat deposits and withdrawals in Earn, including which assets users keep there;
the net fees and risks behind the yield product;
whether UNI burns remain high when token launches and market volatility cool down.

Loading content card…

CZ: self-custody removes one dependency by adding another

Changpeng Zhao used a Trust Wallet incident to describe the part of self-custody that product pages tend to hide. On August 1, he said Trust Wallet had faced a similar pseudo-random-number-generator bug years earlier, causing $12 million in losses that the company covered for every user. His larger point was that software will always have bugs; the meaningful variable is who stands behind the product. This is CZ's account of the earlier incident, not an independent reconstruction of its losses. 11

Twelve minutes later, he made the user-side consequence explicit: developers patching a bug cannot repair wallets that were generated before the patch, and they may have no way to reach users on air-gapped devices. "I'm a believer in self custody, but it puts the burden on you," he wrote. That is not an argument for abandoning self-custody. It is an argument for treating recovery, notifications, and incident response as part of wallet design. 4

The two posts also sharpen the comparison between centralized and self-custodied products. A venue can sometimes patch, contact, reimburse, or absorb losses. A self-custodied wallet can reduce venue exposure, but it cannot make an old key safe after the fact. The reader's real choice is therefore not "custody or no custody"; it is which party can detect, communicate, and pay for failure.

CZ's self-custody post had 3,715 likes and 742,951 views in the retrieved snapshot. His separate August 1 post said the market might be bearish while money was still looking for investments, a useful reminder that capital availability and product quality are different claims. 4 12

Loading content card…

Vitalik: Diamond iO is closer to useful, not useful yet

Vitalik Buterin published "Obfuscation (Part II): Diamond iO" on July 28. The article explains cryptographic obfuscation as a way to let someone run an encrypted program on plaintext inputs without seeing the program's internal logic. That matters when a program contains a secret key and must expose a narrow capability without exposing the key or the rest of the logic. 7

The important qualifier is efficiency. Vitalik describes older obfuscation protocols as having a "galactic" runtime, making them theoretical curiosities despite advances in feasibility. Diamond iO relies on more aggressive and less tested cryptographic assumptions, but moves the runtime closer to "planetary": still infeasible today, yet potentially within reach of a few practical use cases after further optimization. 7

This is not an Ethereum release announcement. It is a long-horizon infrastructure signal. If the construction survives security analysis and its efficiency improves, obfuscation could let developers expose narrowly controlled computation without handing over the full program logic. The immediate follow-up is research, not a token launch: assumptions, proof quality, runtime, and concrete use cases must all improve before builders treat it as production infrastructure. 13

The announcement post had 973 likes and 276,633 views in the retrieved snapshot. The audience response is easy to measure; the technical distance to deployment is not. 13

Loading content card…

Hayes: a one-line liquidity thesis

Arthur Hayes posted on July 30 that "The Fed green lit buying of bombed out DeFi shitcoins with PMF." PMF means product-market fit. This is Hayes's interpretation of the monetary backdrop, not a policy announcement and not a list of assets he endorsed. The post had little supporting detail, so its value is as a directional bet rather than a complete investment argument. 8

The useful question is what would falsify it. A real version of the thesis needs a defined monetary channel, a way to identify product-market fit, and evidence that the projects can retain users when liquidity is less forgiving. Without those, "PMF" is just a label attached to a distressed-token trade.

What changed this week

The strongest statements were not pointing at one new sector. They were assigning ownership of the hard parts:

Armstrong assigned Coinbase the job of operating rails for tokenized assets and agent payments.
Stani assigned Aave a smaller, more deliberate risk surface before securities finance expands.
Adams added yield and token value capture to Uniswap's trading surface.
CZ assigned recovery and incident response back to the individual self-custodian.
Vitalik's Diamond iO work showed how far a cryptographic primitive can move toward feasibility while still being unusable today.

That leaves a practical reading for investors and builders. Follow the claims that can be checked in product behavior: agent payment volume, stablecoin concentration, Earn deposits and yield sources, Aave reserve performance after pruning, UNI burns outside launch spikes, and wallet response during real incidents. The posts are not proof that any thesis will win. They are better treated as public test plans from the people building the systems.

No qualifying in-window podcast release or conference keynote from the tracked leaders was verified. Andre Cronje's returned account timeline contained no current-window first-party post, so he is not represented by a secondhand claim here. This issue is therefore concentrated in verified X posts and Vitalik's July 28 official article. Engagement counts and protocol figures are time-stamped snapshots from the retrieved source records, and figures such as UNI's annualized burn and Aave's affected supply remain speaker-attributed unless independently corroborated.

Onchain everything, Aave's risk purge, and the self-custody bill: July 26-August 2

The week in one sentence

Armstrong: Coinbase wants the rails, not just the venue

Stani: shrink the risk surface before expanding the addressable market

Hayden: Uniswap adds a product layer to the trading venue

CZ: self-custody removes one dependency by adding another

Vitalik: Diamond iO is closer to useful, not useful yet

Hayes: a one-line liquidity thesis

What changed this week

References

Related content