ChainFeeds on Aug. 14 released a new edition of its daily research roundup, collecting five Web3 features from its Aug. 13 briefing. The selection covers Bitcoin market structure, the shape of the next crypto cycle, wallet permissions for AI agents, Ethereum’s post-quantum transition, and startup lessons from Y Combinator CEO Garry Tan.
Glassnode says Bitcoin is still compressed, with demand lagging bullish positioning
In the Bitcoin section, ChainFeeds cites Glassnode as saying that a clearer improvement in market conditions would require BTC to reclaim the Short-Term Holder Cost Basis around $68,700 on stronger volume, together with a renewed return of ETF inflows.
The report says July core inflation fell to 2.5%, headline inflation was broadly unchanged, and Federal Reserve policy did not shift. The effective federal funds rate has stayed unchanged since last December and remains more than one percentage point above core inflation, leaving real rates in restrictive territory. What drew attention, according to the piece, was the market response: Bitcoin showed little meaningful rebound after the data release, while U.S. equities weakened. In what it describes as a relatively mild macro backdrop, BTC still failed to draw support. If price cannot rise on that setup over the next several trading sessions, the report says it may point to weak demand.
It also notes that U.S. consumer confidence remains at the lower end of the past decade, while U.S. stocks hit a record high on Aug. 7 and are still hovering near those levels. Investors are rotating from cash into assets, with AI-linked equities taking most of the flow, but Bitcoin has not shared much in that move. Since the summer, BTC has underperformed U.S. stocks and spot activity has weakened sharply. ETF flows would be one of the first indicators to watch if capital returns, but the piece says no clear comeback in demand is visible yet.
ChainFeeds says Bitcoin is currently pinned between two important on-chain cost levels: the Median Realized Price around $63,000 on the downside and the Short-Term Holder Cost Basis around $68,700 on the upside. Short-term holders are, on aggregate, sitting on unrealized losses. Historically, that cohort has often sold when price rebounds back toward its cost basis, making $68,700 a notable overhead resistance zone. For nearly three months, BTC has traded inside this narrow band, with falling volatility bringing the two cost lines closer together.
Spot market activity has dropped to what the report calls an extreme level. Measured in BTC, exchange spot volume has fallen to the lowest reading since the dataset began in 2019. Even excluding Binance, volume is close to the 2023 bear-market low. In other words, turnover has fallen to one of the lowest levels in seven years. Such low participation usually does not last for long, the piece says, and can magnify price swings: a small amount of fresh demand could move the market sharply higher, while limited selling pressure could also break support quickly. The contracting range and depressed spot volume together create the setup for an expansion in volatility.
Seller pressure is easing, but actual buyer demand has not returned. Only about half of circulating BTC supply is now in unrealized profit, a level the article says is close to zones seen near prior bear-market bottoms. Seller Exhaustion Constant has also dropped to a cycle low and sits at a relatively low level by standards going back to 2013, signaling that sellers are tiring. Even so, the report says the market has not seen the kind of final panic flush that often marked earlier bear cycles.
On ETFs, net flows turned positive at the end of July for the first time in months, but the scale was only a small fraction of earlier institutional accumulation phases. The institutional buying that helped drive the market in 2024 and 2025 has not truly returned, according to the piece. At the same time, BTC continues to flow onto exchanges. The pace is down from the peak seen in early June, but the direction still points to additional potential sell supply.
The derivatives side is another source of concern. ChainFeeds says Hyperliquid whales have maintained net long exposure since mid-March. Futures open interest is now larger than a full day of futures trading volume and is close to the record set last September. With spot turnover very low, ETF buying still weak, and order-book bid depth thinner, leveraged longs have already positioned for recovery. If the range breaks lower, the June low near $58,500 becomes the main level to watch, and long liquidations could amplify the move.
A debate over the next crypto cycle: VC, perpetuals, and whether prediction markets are overrated
The second feature is a long-form post by Lao Bai titled, in ChainFeeds’ summary, a panoramic breakdown of the next crypto market. The framing is blunt: from early-stage VC to cross-asset traders, what remains after putting down blockchain as the default hammer.
The article argues that two things are true at once. Crypto has matured, and its original ambition to remake the world has narrowed. In that view, the industry moved from trying to build a native crypto world to reshaping, and in part serving, the existing financial system. The piece says that if someone in 2018 or 2019 had been told that stablecoins would become a major part of global dollar infrastructure at a scale of several hundred billion dollars, that Bitcoin would get ETFs and attract pension and institutional money, that crypto companies could pursue IPOs, that real-world assets would begin entering traditional finance, and that firms such as Stripe, Visa, and BlackRock would be building crypto infrastructure, public chains, and PayFi products, many would have concluded that the industry had won.
Yet the piece says that, from an asset-performance perspective, crypto did not win in the same way. It says the market no longer offers another DeFi Summer or a setup where altcoins rise 50x or 100x in a year. Over the past decade, from NFTs to inscriptions to various meme assets, native on-chain asset creation did not deliver the broader victory many expected. What crypto ultimately built, the author argues, is a financial rail on-chain, and the next stage is to bring more real-world native assets onto that rail. Two products are singled out as major crypto-native product-market fits: stablecoins and perpetual futures.
On specific platforms, the author says Variational stands out. Outside Hyperliquid, the favored second tier in the piece includes Aster, Lighter, and Variational. Variational’s RFQ model combined with over-the-counter hedging is described as logically sound, and the team, founder, capital backing, and institutional resources are all cited as strengths.
The article goes on to argue that the market only needs four or five major perpetual venues, much as centralized exchange activity ultimately concentrates around a handful of names. It points to Binance and OKX, alongside user-favored platforms such as Bitget, Bybit, MEXC, and Gate, as the sort of structure already seen in exchanges. Perpetual venues, in the author’s view, may settle into a similar pattern, with Hyperliquid in a T0 position and Lighter, edgeX, and Variational in the next band. The piece adds that as perpetuals and centralized exchanges in crypto start to look more like leveraged brokerage products in traditional finance, many less liquid mid- and small-cap stocks could eventually adopt a playbook similar to Binance Alpha.
When AI agents get wallets, the control question moves to the center
The third feature comes from imToken Labs and asks what happens when AI agents begin to hold wallets. ChainFeeds summarizes the shift this way: the question is no longer only how to let an agent act, but how to let people delegate authority without losing control.
The article says that although agent and agent-cluster capabilities have improved sharply this year, they still get stuck in traditional internet workflows: finding a service, opening the website, registering an account, adding a credit card, buying a plan, and obtaining an API key before any actual call can be made. For a human, that process is inconvenient. For software expected to complete a task on its own, any step involving login, registration, payment, or identity checks can force a handoff back to a human operator. The broader point is that the “brain” of the agent has advanced much faster than the internet’s payment infrastructure, which remains built around human users.
x402 is presented as an attempt to change that layer. The piece says it revives the long-existing but rarely used HTTP status code “402 Payment Required” and embeds payment requests directly into the basic request-response loop. Under Coinbase’s design for x402, when an agent requests a paid API, the server can tell it how much to pay, which assets are accepted, and where the payment should go. After the agent pays and resubmits the request with proof of payment, the server verifies the transaction and returns the resource.
The article then contrasts that with MetaMask Agent Wallet. If Cloudflare mainly addresses how an agent buys something, MetaMask Agent Wallet moves one step further and looks at how an agent can use assets directly. In this setup, the shift is no longer just from asking AI whether ETH looks attractive, but toward allowing an agent to execute on-chain actions inside pre-set permission boundaries. The article gives an example instruction: buy 0.2 ETH if ETH falls toward $3,000 and gas sits below the 24-hour average. In that model, the user defines the goal, conditions, and permissions, while ongoing monitoring, condition checks, trade preparation, and even final execution can be delegated in part to the agent.
That is what the article calls an agent’s “economic autonomy.” It does not mean the agent truly owns its own property. It means the agent can be given an account, a discretionary budget, and a set of economic permissions it can call on as conditions change. It may buy outside information or computing resources on its own, and it may deploy real assets within rules set by the user.
imToken Labs says the core wallet question over the past decade has been private-key management. Wallet interfaces have changed, but the basic relationship has not: the human initiates the action, inspects the transaction, and gives the final signature. The wallet’s key job is to protect the private key that determines control of assets and final authorization. Once an agent is inserted into that flow, another layer appears. Many actions no longer require the user to build each transaction by hand. The structure shifts from humans operating assets directly to humans expressing goals first and then delegating part of the execution path.
Under imToken’s design approach, each agent that receives execution rights gets a separate agent account. Its session key is generated and isolated inside a trusted execution environment, or TEE, and the key does not leave that secure environment. The account also has to be bound to a clear policy covering allowed protocol whitelists, per-transaction limits, daily limits, operating frequency, and validity periods. The result is not an open-ended wallet but an execution account boxed in by policy guardrails. The user still retains the higher level of control and can change the policy, pause or resume the agent, revoke authority, and pull funds back. AI may help parse intent, plan routes, estimate costs, and flag risks, but actions outside the approved boundary still have to return to the user for confirmation.
Recursive STARKs are framed as a key lever for Ethereum’s next decade
The fourth piece, from DengChain Community, argues that recursive STARK proofs sit near the center of Ethereum’s long-term transition. ChainFeeds says the article looks at post-quantum security risks and alternatives across consensus, accounts, rollups, and data availability, and also discusses four dimensions of privacy and how recursive STARK aggregation could underpin base-layer scaling.
The article says Ethereum is approaching an inflection point. Its first decade validated the idea that a public blockchain can support meaningful applications at scale. Smart contracts, DeFi, stablecoins, appchains, and tokenization turned Ethereum from an experimental decentralized computing platform into a settlement layer for a growing digital economy. The second decade presents a different challenge: Ethereum must support economic activity that society actually depends on. Stablecoins are becoming payment rails, tokenized assets are moving beyond pilot programs, and institutions are building directly on public rails.
As Ethereum becomes critical infrastructure, the piece argues, it has to be hardened for the years ahead so that it remains robust, secure, efficient, and able to operate at decentralized scale. It lays out three concrete requirements: withstand threats that do not yet fully exist, especially cryptographically relevant quantum computers; provide enough privacy for real economic activity that requires confidentiality; and scale without giving up neutrality, openness, or censorship resistance. The article puts that in a simpler line: Ethereum must, above all, remain censorship-resistant, open source, private, and secure. Those properties do not become less important as adoption grows. They become the basis for adoption itself.
A post-quantum transition, the article says, must cover every part of the current protocol that is vulnerable to quantum attacks. That includes ECDSA signatures, BLS consensus proofs, KZG commitments used for blobs, and pairing-based SNARKs used by rollups when settling on L1. Their common dependency is the hardness of the elliptic-curve discrete logarithm problem, which Shor’s algorithm could solve in polynomial time on cryptographically relevant quantum computers.
The article reviews several candidate replacements. Lattice-based signature schemes such as ML-DSA and Falcon can keep signatures relatively small, but on-chain verification costs are high, above 1 million gas, and aggregation theory remains immature. Hash-based signatures rely only on the security of the underlying hash function, but they pay for that with size: signatures can run to several KB, and verification today may still cost hundreds of thousands of gas. STARK proof systems are built on hash functions in the random oracle model. They are transparent, natively post-quantum, and do not require a trusted setup. Their cost comes in proof size and verification cost. A STARK carries thousands of Merkle-path hashes, making each proof tens to hundreds of KB, and on-chain verification with current provers can cost millions of gas, roughly an order of magnitude above pairing-based SNARKs.
The piece says post-quantum signatures and privacy systems share the same hard feature: they are cryptographically expensive. PQ signatures are larger and costlier than the elliptic-curve schemes they would replace. Privacy protocols depend on heavy proving end to end, raising complexity and cost for users. If deployed at protocol scale, both would increase validation costs for every node in the network. The same answer appears over and over: fold post-quantum signatures into recursive STARKs, wrap quantum-safe data-availability encoding into a STARK, batch privacy-transfer proofs into another STARK, and note that rollups are already settling in this general way.
The repeated question, in the article’s telling, is how to afford stronger cryptography. Each time, it narrows to a more specific one: how cheaply and how efficiently can STARKs be aggregated and verified? Ethereum’s rollup-centric roadmap has already moved execution away from L1 and left the base layer focused on consensus, settlement, and data availability. But scaling does not stop at Layer 2. Every rollup inherits what the base layer provides, including consensus, finality, verification, and data availability. Improving the base layer improves everything built on top of it. Recursive proving and aggregation can compress many proofs, or many expensive cryptographic checks, into one succinct proof. Since one STARK can verify another, those aggregations can be composed without a fixed limit. In the end, each node may only need to verify a single proof rather than thousands.
a16z interview with YC’s Garry Tan focuses on hot trends, AI coding, and managerial bandwidth
The fifth feature, from TechFlow, summarizes an a16z interview with Y Combinator CEO Garry Tan. The headline message is that chasing what is hot can become an expensive habit for founders.
According to the article, Garry Tan says the biggest mistakes of his career were not isolated product calls but repeatedly going after whatever looked fashionable instead of sticking with fields he truly understood and cared about. He recalls graduating from Stanford in 2003 with solid experience in web programming, then leaving that area because Web 1.0 had just crashed, the Nasdaq had fallen sharply, and the prevailing mood said the web was dead. He turned to Windows Mobile instead. As he tells it, Mark Zuckerberg was likely working on Facemash around that time and Facebook did not yet exist, while he had stepped away from web programming at exactly that moment.
He also points to a second missed opportunity that he says was even more expensive. When Palantir was being founded, Peter Thiel flew to Seattle to meet him and offered him a $70,000 check representing his salary if he joined. He declined. In the interview, Garry Tan estimates that the decision now looks like a mistake worth roughly $2 billion to $4 billion. He says the root cause behind both errors was the same: he was looking at the map rather than the terrain. He cared too much about what looked cool and what would win approval from respected investors, instead of paying attention to what the smartest people around him were actually building.
On the present moment, the article highlights Garry Tan’s view of vibe coding and agentic coding. A founder today, he says, can be equivalent to 400 versions of themselves from two years ago. The claim is meant literally, not as hype. A single person can now build things that previously required a full engineering team. He also says the golden age of pure SaaS is nearing its end. For pure per-seat SaaS products with no data moat and no network effects, the model may stop working within five to ten years. The reason is that AI is pushing the cost of making software close to zero, so software itself is no longer the moat. Data, user relationships, and network effects are.
His advice, as relayed by the article, is that if someone is still building a pure SaaS business in 2026, it should serve as a bridge toward a deeper moat rather than the final destination. The piece also stresses one line from the interview: code is no longer precious. It interprets that as a shift in what is costly. When writing code was expensive, teams used project approvals, reviews, specs, and QA to reduce the cost of being wrong. If code is now close to free, that operating assumption changes. What becomes more expensive is taste and agency: knowing what to build and what deserves sustained effort.
The article ends with a management example involving Brex co-founder Pedro. Garry Tan says Pedro uses an AI agent to read the meeting notes of all direct reports and see information two levels down. That means he does not need to sit through a given compliance team meeting for the agent to summarize what was discussed over the prior three weeks, where disagreements surfaced, and which people were in conflict. He can then enter a meeting with full context, make a decision, and leave. In Garry Tan’s account, that gets at a deeper management problem AI may help solve: once a company grows beyond the limits of one person’s cognition, many issues disappear into the middle layers because nobody has enough bandwidth to know what is happening everywhere. The article cites the psychology shorthand of “7 plus or minus 2” for human working memory and contrasts it with a person plus a configured agent being able to hold “three Harry Potter books’ worth of context” in mind.
At the top of the newsletter, ChainFeeds also listed several headline items for Aug. 14, including Gemini’s second-quarter earnings, a further delay to the U.S. SEC’s tokenization innovation exemption, the Ethereum Foundation’s move away from the Poseidon hash algorithm toward SHA2 or BLAKE2, Pump.fun’s launch of a referral-based reward program, and ether.fi’s launch of tokenized asset trading, portfolio-backed loans, and fiat account functions.
