Samsung posted the best semiconductor quarter in 40 years and the market sold it hard enough to trip KOSPI's circuit breaker. After Iran struck commercial ships in the Strait of Hormuz, the US retaliated overnight with strikes across Iran and Iran hit back at US bases in the Gulf, shattering June's truce and pushing oil sharply higher. NATO opened in Ankara with Russia's 68-missile barrage on Kyiv as the backdrop, Trump calling the alliance "disloyal," and China conducting its first acknowledged submarine ballistic missile test into the Pacific without notifying Washington.
The US-Iran conflict reignited overnight. Iran struck commercial shipping in the Strait of Hormuz, the US launched retaliatory strikes across Iran, and Iran then hit US installations in two Gulf states. June's truce that had reopened the strait is broken, and it is the day's dominant risk driver as NATO's leaders gather in Ankara. The full picture is in Geopolitics below, the energy and rates consequences are in Markets and Macro, and the levels are in the Dashboard.
Crypto data provided by CoinGecko
Samsung's best semiconductor quarter in 40 years produced its worst single-day stock reaction in months: shares fell roughly 10%, KOSPI tripped its circuit breaker for the sixth time in 2026, and the SOX index dropped more than 6%. Two competing reads, and they are not equal. One: the earnings were fully priced in and this is positioning mechanics. Two: the market is repricing the whole semiconductor supply-chain thesis, because DeepSeek, the lab that matched frontier performance at a fraction of the cost six months ago, said the same day it is designing its own inference chip. Read two is more reliable, and the tell is breadth: a pure sell-the-news move stays in Samsung and the Korean memory names; this bled across the entire SOX. When a record print sells off market-wide, not name-specific, the market is not doubting Samsung's quarter. It is doubting who captures the profits from the next one. The framework for that repricing is in today's Take.
The 10-year yield climbed to 4.50%, its highest in two weeks, on the same day equities sold off, which is the market's way of telling you this is an inflation scare, not a growth scare. When growth genuinely wobbles, money runs from equities into Treasuries and yields fall. When inflation is the fear, both assets reprice lower simultaneously: equities on margin compression and bonds on real-rate expectations. The catalyst was the Strait of Hormuz escalation, where Iran struck commercial shipping and the US retaliated overnight, pushing oil past $72 a barrel and reintroducing shipping-lane risk to the energy complex. Weight the yields-up-equities-down combination over any individual data print: the bond market is pricing sustained energy-supply risk, not a one-day headline, and the rates traders' inflation read has been more accurate than the equity market's growth-optimism read at every inflection since 2022.
NATO's move toward 5% of GDP on defense, unveiled at the Ankara summit's opening session, is the largest peacetime fiscal commitment in European postwar history. The bill lands on the sovereign-bond market, not the income statement. European allies spend roughly 2.5 to 4% of GDP on defense today; scaling to 5% by the early 2030s adds hundreds of billions in annual spending that has to come from higher taxes, deeper deficits, or program cuts, and procurement cycles of 5 to 15 years lock it in regardless of who governs next. The part the market has not priced is the collision: this permanent upward step in European government issuance arrives exactly as the ECB unwinds its own balance sheet and stops being the buyer of last resort. More bonds, fewer official buyers, for a decade. The 2% target took ten years to approach; 5% reprices the risk-free rate that sits underneath every European asset.
Coinbase secured UK Financial Conduct Authority authorization to offer traditional investments alongside crypto, making it the first major crypto-native exchange to cross the regulatory bridge in both directions. Instead of remaining crypto-only, Coinbase can now offer stocks, bonds, and ETFs to UK customers on the same platform that handles their Bitcoin and Ethereum. The play is the same one Robinhood ran in the other direction, starting with stocks and adding crypto: own the full investment stack, not just one asset class. The structural significance is that Coinbase obtained this authorization on the same day the FCA finalized its comprehensive crypto framework, meaning it moved first in a market where competitors will need months to catch up on the compliance infrastructure. First-mover advantage in regulated markets is not about speed. It is about the accumulation of compliance precedent that makes the next approval faster and every competitor's application slower.
The same FCA framework that authorized Coinbase sets a hard clock for everyone else: a mandatory authorization gateway opens September 2026, and firms without approval cannot serve UK customers after October 2027. It forces crypto firms to meet the same capital-adequacy, conduct, and consumer-protection standards as traditional financial institutions, the price of admission to the most structured post-MiCA market in the world. The contrast with Washington is the story: the US has stablecoin-specific KYC rules under the GENIUS Act but still no comprehensive market-structure framework, which means the UK has, for the first time in this digital-asset cycle, overtaken the US on regulatory clarity. Capital, exchanges, and compliance teams migrate toward wherever the rules are already written. The rules just got written in London.
Solana's tokenized real-world asset ecosystem hit $3.41 billion, an all-time high driven by on-chain treasury tokens and structured credit products. Real-world asset tokenization, putting treasuries, corporate credit, and structured products on a blockchain ledger, has been promised for years; the milestone is that a single chain now holds enough to matter for institutional portfolio allocation. The growth is concentrated in tokenized treasuries (Ondo, Maple) and structured credit, which suggests the use case is not retail speculation but institutional cash management: firms parking working capital on-chain for 24/7 settlement and yield. If RWA volume on Solana doubles again by year-end, the chain stops being a trading venue and starts being a settlement layer, and the economic model shifts from transaction fees to custody and clearing.
Security researchers at Sysdig documented JADEPUFFER, the first observed autonomous AI ransomware operation, in which an AI agent independently discovered a vulnerability, wrote and deployed the exploit, and managed the ransom negotiation without human intervention. The distinction from prior AI-assisted attacks is full autonomy: no human operator wrote the payload, selected the target, or managed the extortion. The AI did it end to end. The significance is not the sophistication of the individual attack but the economics it changes: human ransomware operators are expensive, slow, and leave behavioral signatures that defenders can fingerprint. An autonomous agent can run thousands of simultaneous campaigns at near-zero marginal cost with no sleep schedule and no operational pattern to detect. The cybersecurity industry's entire detection model is built around human attacker behavior: keystroke cadence, working hours, lateral-movement patterns. JADEPUFFER is the first documented case where none of those signals exist, which means the detection paradigm itself, not just the defenses within it, needs to be rebuilt.
Illinois Governor J.B. Pritzker signed the AI Safety Measures Act, making Illinois the third major state to pass AI-specific regulation after California and New York, and collectively these three states cover roughly 40% of the US AI market by revenue. The act requires algorithmic impact assessments for "high-risk" AI systems in hiring, lending, insurance, and healthcare, and creates a state AI Safety Board with enforcement authority. The signal is not the specific provisions, which are comparable to the EU AI Act's risk-tiering approach, but the regulatory fragmentation: with no federal framework, AI companies now face three different state-level compliance regimes covering their largest markets. The compliance cost of three divergent state regulations exceeds the cost of one federal law, which makes the Illinois signing the strongest catalyst yet for a preemptive federal bill. Watch whether Congress accelerates its timeline in response.
Reuters reported that DeepSeek is developing its own custom AI inference chip, targeting the fastest-growing segment of AI computing and aiming to reduce its dependence on Nvidia and Huawei hardware. The chip is designed for inference, not training, which means DeepSeek is going after the segment where volume is exploding as AI deployments scale beyond research labs into production. DeepSeek's cost-engineering track record (it matched frontier training performance at a fraction of competitors' spend) makes this more credible than a typical "we are building a chip" announcement from a software company. If DeepSeek secures manufacturing capacity, likely through SMIC's most advanced processes, the combination of a lab that already demonstrated extreme cost efficiency with custom silicon purpose-built for inference creates a structural challenge to the assumption that all AI chip demand flows to incumbents indefinitely. The Take explores the framework underneath.
Iran's attacks on three commercial vessels in the Strait of Hormuz, including the Qatari LNG carrier Al Rekayyat, drew overnight US retaliation: CENTCOM says American forces struck more than 80 Iranian targets, and Washington reimposed oil sanctions on Tehran. Iran counter-struck US installations in Bahrain and Kuwait, where missile-alert sirens sounded, shattering June's truce that had reopened the strait. Hitting a Qatari ship was escalation by target selection: Qatar hosts Al Udeid, the largest US air base in the Middle East, and is the world's largest LNG exporter, so the fighting now sits on top of the Gulf basing architecture Washington depends on. The last time shipping was struck in Hormuz at this rate, the 2019 tanker attacks, Brent spiked $10 within 48 hours; this time the shooting is two-way. The question is how wide the target set gets, because Iran hitting US bases in two Gulf states puts every basing agreement in the region in play at once, not as a bargaining chip but as a target.
Russia launched 68 missiles and 351 drones at Kyiv overnight, killing at least 19 people, in the largest combined aerial attack in weeks, timed to the opening day of NATO's Ankara summit. Zero ballistic missiles were intercepted, which means either Ukraine's Patriot batteries were depleted, repositioned, or overwhelmed by the volume. The timing is not coincidence but communication: Russia struck on the eve of the summit where allies were set to pledge $140 billion in additional support, sending the message that pledges do not equal protection. Zelensky invoked the strike to push for additional interceptor-missile deliveries, which changes the summit conversation from aid volume to aid composition, specifically the gap between money committed and air-defense hardware available to absorb it.
Trump arrived at NATO's Ankara summit calling allied nations "disloyal" and pushing for sharply higher defense-spending targets, while the communique labeled Russia the "most significant and direct threat" to the alliance for the first time in those terms. The contradiction at the summit's core is that the alliance's most powerful member is simultaneously demanding its allies spend more on collective defense and questioning whether the collective commitment is worth honoring. The 5% target faces the same constraint the 2% target did: production capacity. European defense manufacturers cannot absorb the spending even if budgets authorize it, because factory lead times for precision munitions, air-defense systems, and armored vehicles run 24 to 36 months and the plants to fill orders at 5%-of-GDP scale do not exist yet. The gap between spending commitments and deliverable capability is the window an adversary times its moves into.
China conducted what New Zealand, Australia, and Japan described as its first publicly acknowledged submarine-launched ballistic missile test into the Pacific Ocean, and the United States was not notified in advance. The test changes the deterrence geometry in the Indo-Pacific: a submarine-launched ICBM can be fired from under the ocean, making it nearly impossible to preemptively strike, which means China's nuclear second-strike capability just became credible in a way it was not when its arsenal was entirely land-based. The absence of US notification is the signal within the signal: during the Cold War, the US and USSR maintained hotlines specifically to prevent accidental escalation from unannounced tests. China choosing not to notify suggests it views the Pacific test as a message, not a routine exercise, and the intended recipients are the AUKUS partners whose submarine cooperation agreement is the declared reason for the test's location.
A 37-story Manhattan high-rise, the former Pfizer global headquarters being converted into luxury apartments, began buckling on July 7 when two structural columns on the 21st floor failed during construction work. Nine surrounding buildings were evacuated, including a school running a summer camp for 400 children. Fire officials said the steel-framed building, constructed in the 1960s, has continued shifting since the initial failure and remains unstable. The event is a physical reminder that adaptive reuse of midcentury commercial buildings, the dominant play for converting emptying office towers into residential space, carries structural risk that the financial models for these conversions rarely price. The question is not whether this building falls. It is whether the next decade of office-to-residential conversions has underpriced the engineering complexity of repurposing structures designed for uniform commercial loads to handle the uneven loads of residential layouts, plumbing stacks, and HVAC systems they were never built for.
Researchers reported that ancient bees carved nest chambers into the tooth sockets of mammal bones roughly 20,000 years ago, the first evidence that a living species used the structural cavities of dead vertebrates as reproductive architecture. The finding reframes how insects colonize landscapes after megafauna extinctions: the bones of the animals that vanish become the infrastructure for the species that follow, a literal case of one extinction's debris becoming another lineage's foundation.
Physicists demonstrated an ultrafast quantum control technique that makes a quantum system appear to run backward in time, reversing the evolution of quantum states with enough precision to be useful for error correction. The technique is not time travel; it is the ability to undo quantum decoherence, the process by which quantum information leaks into the environment and becomes classical noise. If the method scales, it attacks quantum computing's central engineering problem from a direction no one was working: instead of preventing errors, you reverse them after they occur.
Cuba's national power grid collapsed for the third time in 2026, plunging roughly 10 million people into darkness and forcing hospitals to cancel surgeries. The first collapse was January, the second was March, and each recovery has been shorter and more fragile than the last. The pattern is not random failure but structural decay: Cuba's generation fleet averages over 40 years old, runs on heavy fuel oil it can barely afford, and has no capital pipeline for replacement. A grid that collapses three times in six months is not experiencing outages. It is demonstrating that the system has passed the point where maintenance can restore stability, a phase transition from "unreliable" to "non-functional" that infrastructure engineers recognize as irreversible without full replacement.
The legal machinery that rations water for the American Southwest expires this December with no replacement agreed, and the market still prices Western water as free and infinite.
Almost nobody outside the basin has noticed that a hard deadline is bearing down on the water supply for forty million people. The 2007 Interim Guidelines that govern how Lake Powell and Lake Mead are operated, the 2019 Drought Contingency Plans layered on top, and the companion water-sharing minutes with Mexico all expire on December 31, 2026, and there is no successor deal. This is not a distant risk; the machinery to replace it has already stalled in the open. The Bureau of Reclamation released its Post-2026 Draft Environmental Impact Statement on January 9, 2026 and could not even name a Preferred Alternative, because the seven basin states are deadlocked: the Upper Basin and Lower Basin have filed competing operating plans, Arizona, California, and Nevada submitted a separate Lower-Basin-only proposal on May 1, and lawyers are now openly discussing a Supreme Court fight over the century-old Colorado River Compact itself. The physical backdrop is a river that has been in structural drought for two decades and reservoirs that keep testing shortage-trigger elevations. Here is the structural tell most investors miss: the river irrigates on the order of five million acres and agriculture uses roughly seventy to eighty percent of the water, but much of that farm water sits on senior priority rights, so when cuts come the pain lands first on cities, industry, and the data centers now siting into Arizona and Nevada on the assumption that Western water and power are cheap and boundless. The market has filed the whole thing under "environmental story, not a trade." If the basin states miss the end-2026 deadline and the allocation slides into litigation or federally imposed cuts, expect a durable re-rating toward the companies that sell the way out of scarcity: water-treatment and water-recycling names like Xylem (XYL), Veralto (VLTO), and Ecolab (ECL), and regulated water utilities like American Water (AWK) and Essential Utilities (WTRG), which gain years of demand as reuse and desalination stop being optional. The losers are the water-intensive users who penciled in cheap water: Southwest agriculture (farmland owners such as Farmland Partners (FPI) and Gladstone Land (LAND), growers such as Limoneira (LMNR)) and the AI data-center buildout that assumed the desert would always have water to cool it. Watch: the Bureau of Reclamation's Final EIS and Record of Decision on post-2026 operations, expected during 2026, and Lake Mead's projected elevation heading into the 2027 water year. A deadlocked negotiation combined with a falling reservoir elevation is the trigger; the moment the Bureau imposes its own terms, the scarcity trade stops being theoretical.
Context signal: connecting an ordinary phone straight to a satellite is about to stop being a novelty and become a standard feature, and the first companies it breaks are the specialized satellite-comms firms whose whole business was being the only option off the grid.
The comfortable read is that satellite-to-phone service is a neat wilderness gimmick. The structure says it is a substitution event. Direct-to-device, an ordinary, unmodified smartphone connecting straight to a satellite with no special hardware, went from experiment to shipping product in barely a year. T-Mobile and Starlink's "T-Satellite" launched commercially in mid-2025, expanded from texts to app data (maps, weather, WhatsApp) by late 2025, and Starlink had more than 650 direct-to-cell satellites in orbit by early 2026; AST SpaceMobile, partnered with AT&T and Verizon, plans to launch 45 to 60 next-generation satellites by the end of 2026, enough, it says, to offer 5G data across the United States. The reason this is structural, not cyclical, is what it does to pricing power: once "no dead zones" is a cheap add-on bundled into a normal phone plan, the premium that specialized satellite-communications incumbents charged for being the only way to connect off-grid collapses. You cannot sell scarcity that a mass-market carrier is giving away. The consequence: if AST hits its end-2026 launch cadence while Starlink keeps widening direct-to-device data coverage, the commodity mass-market service eats the high-priced niche. Iridium (IRDM), whose satellite-phone and satellite-IoT business is built on being the specialist, faces its premium competed away, while the firms that own spectrum or exclusive carrier deals, Globalstar (GSAT, Apple's satellite partner) and AST SpaceMobile (ASTS), plus the carriers that resell direct-to-device as a paid upgrade, capture the value; the cell-tower landlords, American Tower (AMT), Crown Castle (CCI), and SBA Communications (SBAC), keep the urban-capacity business but lose the "coverage everywhere" moat that justified the economics of remote towers. Watch: AST SpaceMobile's actual satellite launch count and first commercial 5G-data direct-to-device service through the end of 2026, and Iridium's subscriber and average-revenue-per-user trend in its personal-communications and IoT segment. If AST's constellation reaches continuous US coverage while Iridium's niche revenue per user rolls over, mass-market direct-to-device has commoditized the premium and the substitution is underway, not theoretical.
When Your Best Customer Starts Making Its Own. When a customer's spending on a supplier crosses a threshold, the customer starts designing its own version. This is not diversification. It is the demand itself creating the substitution, and the pattern has run through every major technology platform transition of the last two decades.
Apple silently designed M-series chips for years while being Intel's most prestigious laptop customer. In 2020, Apple switched. Intel's PC division has not recovered. Amazon designed Graviton server processors while being one of Intel's largest cloud customers. By 2025, Graviton powered a substantial fraction of AWS compute workloads at lower cost and higher performance. Google designed TPUs while buying thousands of Nvidia GPUs. TPU now handles the majority of Google's internal AI inference. In each case, the customer did not leave because the supplier failed. It left because the supplier succeeded: margins got high enough and the product became important enough that internalizing the supply chain penciled out.
DeepSeek is now running the same play against the broader semiconductor supply chain. Reuters reported July 7 that the Chinese AI lab is developing its own custom inference chip, targeting the fastest-growing segment of AI computing. DeepSeek is not doing this because it is unhappy with its current chips. It is doing this because it demonstrated, six months ago, that it could match frontier AI performance at a fraction of the training cost, and the logical next step for a lab that already optimizes ruthlessly on compute cost is to own the silicon. The chip targets inference, not training, which means DeepSeek is going after the segment where volume is exploding as AI moves from research labs into production deployments.
The structural implication for the semiconductor bull case: the consensus thesis holds that AI demand is insatiable and all roads lead to incumbent chipmakers. DeepSeek's move puts a timer on that assumption. Not a kill switch. A timer. The demand does not disappear. It internalizes. And the companies that lose are not the ones that fail to innovate but the ones whose best customers innovate past them.
The counter-case is that chip design is extraordinarily hard and manufacturing is harder, and nearly every company that has tried custom silicon has been humbled by the gap between announcing a chip and shipping it at scale. DeepSeek faces US export controls that block access to TSMC's leading-edge foundries (3nm and 5nm) and high-bandwidth memory, forcing it onto SMIC's less advanced processes or sanctioned workarounds that may not deliver competitive performance. Custom chips have a poor hit rate historically: Google's TPU is the exception, and even TPU has not displaced Nvidia outside Google's own infrastructure. Nvidia's CUDA ecosystem creates massive switching costs that no custom inference chip has replicated, because developers have built years of tooling and optimization on CUDA and will not migrate to a new stack without overwhelming economic incentive. Meanwhile, training, where the largest GPU spend lives, remains untouched by DeepSeek's inference-only chip. And inference itself is commoditizing on existing architectures: Nvidia's inference pricing has fallen significantly over the last two years as competition from AMD, Intel, and cloud-custom silicon (AWS Inferentia, Google TPU) intensifies, which shrinks the margin available for a new entrant to capture. The falsification test is concrete: if DeepSeek secures a manufacturing partner and reaches tape-out on a competitive process node by end-2027, the substitution is real and the semiconductor moat has a structural leak. If the chip stalls in design, gets blocked by export restrictions, or reaches production but underperforms Nvidia on a total-cost-of-ownership basis, the demand-creates-substitution thesis fails for the same reason most custom-chip projects fail: the gap between a lab's ambition and a foundry's capability is where billions go to disappear.
"The most common outcome after potentially traumatic events is not lasting disorder. It is a stable trajectory of healthy functioning."
— George Bonanno, The End of Trauma (2021)
You assume that when something difficult happens, you need to process it. Sit with the feelings. Work through the layers. The culture reinforces this at every turn, and the assumption runs so deep it feels biological: pain must be metabolized or it stays toxic.
Bonanno spent two decades at Columbia following thousands of people through bereavement, terrorist attacks, serious injuries, and job loss. The most common trajectory was not prolonged struggle followed by gradual recovery through processing. It was stability. The majority returned to healthy functioning relatively quickly, without therapy, without structured emotional processing. They were not in denial. They were recovering, which turns out to be the default human response, not the exceptional one.
The more uncomfortable finding was that forced processing could make things worse. People pulled into mandatory debriefing after traumatic events sometimes showed worse outcomes than those left alone. The intervention assumed pathology where none existed. The inversion is not that emotions do not matter. It is that the instinct to immediately analyze and narrate every difficult experience is a learned behavior, not a biological necessity, and it can interfere with the natural recovery that would have happened on its own. The people who bounce back fastest are often the ones who do the least deliberate emotional work in the first day or two, not because they are avoiding their feelings but because they are letting their nervous system do what it evolved to do.
Today's practice: the next time something difficult lands and you feel the pull to immediately work out what it means, stop and give the feeling 24 unnarrated hours instead. Watch how often your baseline returns on its own. That return is not you avoiding the work. It is the recovery system you already have, doing the job the culture insists it cannot.
The evolutionary biologist Richard Lewontin proposed in the 1980s what Darwin's framework left out: organisms do not merely adapt to their environments; they construct them. An earthworm does not simply tolerate soil. It creates soil. It pulls organic matter underground, aerates the earth with its tunnels, and chemically transforms the substrate through its digestive processes. The soil it creates then selects for earthworms: better drainage, better microbial life, better conditions for the species that made them. The organism changed the environment, and the changed environment selected for more of that organism. The standard evolutionary story, "adapt or die," is incomplete. The fuller version is "adapt, then reshape the landscape so it rewards your adaptations."
The concept was formalized by Odling-Smee, Laland, and Feldman in "Niche Construction: The Neglected Process in Evolution" (2003), and it runs far beyond earthworms. Beavers build dams that create wetland ecosystems benefiting dozens of species, including future generations of beavers. Coral polyps secrete calcium carbonate skeletons that build reefs, which create the sheltered, nutrient-rich environment that coral itself thrives in. Humans cook food, which pre-digests nutrients, which reduces the selective pressure on jaw and gut size, which frees caloric budget for brain development, which enables the invention of more cooking techniques. In each case, the organism is not passively selected by a fixed environment. It is actively constructing the environment that does the selecting.
The decision tool is to ask, when evaluating any competitive strategy, whether the actor is adapting to the game or rewriting the game to select for its own strengths. A company that competes on price within existing market rules is adapting. A company that sets an industry standard, builds an ecosystem around its product, or creates regulatory conditions that favor its architecture is constructing a niche. Apple did not make a better phone; it built the App Store, which created an ecosystem that selected for Apple's hardware as the default platform. Amazon did not build a better store; it built the logistics network, which changed what "competitive retail" meant and selected for Amazon's scale. The most durable advantages come not from winning the game but from changing what the game rewards, so that your existing strengths become the criteria for success rather than merely an advantage within someone else's criteria.
In 2018 three researchers, Ethan Bernstein, Jesse Shore, and David Lazer, handed groups a genuinely hard problem, the kind with one excellent solution hidden among many mediocre ones, and let them solve it under three different rules of contact. One group worked in total isolation. One saw each other's work constantly. One saw each other only intermittently, stretches of solitude broken by brief windows of sharing. The result overturned the intuition that more collaboration is better. Constant interaction raised the group's average solution but destroyed its best one: the moment everyone could see who was ahead, they copied that person, the group converged, and the diversity that might have produced something excellent evaporated before it could form. Total isolation did the reverse: it preserved the occasional brilliant answer but left the average low, because most people floundered alone with no one to learn from. Only the intermittent group got both at once: the average climbed and the best solution survived, because the breaks kept a variety of half-finished approaches alive long enough for a good one to be found, and then the brief contact let it spread. Connection, it turns out, is not a quantity you want to maximize. It has a sweet spot, and "always on" sits past it.
The uncomfortable part is that constant contact does not merely fail to help; it actively erases a group's peak capability by making everyone rush to consensus before the better ideas exist. Always-on optimizes the mediocre and kills the exceptional. That is the hidden tax on real-time everything: the shared dashboard, the open channel that never closes, the meeting that runs until agreement. The more continuously a group watches its own work, the faster it herds, and herding trades away the outlier answer that was the entire reason to assemble a group in the first place. What feels like tight coordination is often just premature convergence wearing a productivity costume.
So when your team is stuck on something hard, do not put everyone in constant contact; impose intermittency on purpose. Have people work the problem alone, come together briefly to share, then separate again before the consensus hardens. You can test the discipline inside a week: on the next real decision, kill the running group thread, collect everyone's independent take first, and only then let people see each other's work, then watch whether the best idea comes from someone who had not yet been pulled toward the group's early favorite. Rival labs that work in parallel and compare notes at intervals advance faster than those reading the same preprint the hour it drops. Product teams that prototype alone before the group review ship better work than those making every decision live in a shared channel. The lever is not more communication. It is the rhythm of connection, not the volume.