Trump and Iran's presidents signed the MOU on June 17. The 14 points: $300B in international financing to rebuild Iran, immediate lifting of oil sanctions, Strait of Hormuz reopens in 30 days (it carries ~20% of global crude), and a verification framework for Iran's nuclear program. WTI fell from a wartime peak above $95 to $76.60. Brent put-call skew collapsed from "wartime" to "normal" in 48 hours. The deal removes the oil-shock tail risk from the next CPI print, giving Fed Chair Warsh cover to stay hawkish without an exogenous inflation amplifier. Conditional on the ceasefire holding — VP Vance publicly warned Israel against breaking it.
Oil priced a Hormuz-closure tail risk of $100+/bbl for months. The Jun 17 MOU didn't just announce a ceasefire — it specified a 30-day Hormuz-reopening timeline plus immediate sanctions relief. WTI fell -19% peak-to-trough in 5 sessions. Brent put-call skew, the cleanest read on tail risk, collapsed from "wartime" pricing to "normal" within 48 hours of the signing. The deal is conditional — Vance's Israel warning + Trump's "could still resume attacks" hedge keep an option-value tail — but the base case is now Hormuz-normalized by mid-July, which puts $70 oil in play by Q3.
Equity reaction: energy -3 to -5% on the week (CVX, XOM, OXY); SMR +7% (uranium beneficiary); sustainable-energy ETFs +1.5%; defense names (LMT, RTX) -2% on reduced Middle East risk premium.
The Fed's hawkish posture is now endogenous to the US labor market, not amplified by an oil shock. Warsh can stay at 3.50–3.75% without an exogenous CPI excuse.
Stanford's DeLM replaces the central orchestrator with parallel agents that write compact, verified updates to a shared context layer. No controller, no bottleneck, no single point of failure. The result: +10.5pp on SWE-bench Verified and −50% per-task cost vs the strongest centralized baseline. Every major meta-harness launch of the last 7 days (Databricks Omnigent, Mastra Harness, Vercel Eve, Microsoft agent-framework v2) bet on the wrong architecture. The unlock for Companies-of-One is real — and it's a 3-stack problem (coordination substrate + verified-update primitive + small-model inference).
Weibo's VibeThinker-3B — a post-trained Qwen2.5-Coder-3B base — just outscored DeepSeek V3.2 (671B parameters) on AIME 2026. With Claim-Level Reliability Assessment (CLRA) test-time scaling, it hits 97.1 — SOTA at any size. The post-training cost: $7,800 vs DeepSeek R1's $294,000. The "scale is all you need" thesis is broken, again — this is the third small-model-vs-frontier data point in 60 days.
Key techniques: Spectrum-to-Signal Principle (post-training) + Claim-Level Reliability Assessment (test-time scaling). For TTL: this is the third open-weight-3B-class model this month that's hit frontier-class reasoning. The procurement math for "small + skills" just became a default, not an option.
The 14-point framework is silent on financing sources. The "$300B in international financing" is explicit — it must come from somewhere. Plausible counterparties, ranked by alignment with Trump's structural ask ("extract a peace dividend from allies"):
This is a financing innovation — first time the US has extracted a peace-dividend structure at this scale since the Marshall Plan. Watch the next 30 days: formal commitments, the IAEA verification protocol, and the first Iranian oil cargoes off the new sanctions-free market.
The Strait of Hormuz carries ~20% of global crude transits. A full closure = $100+/bbl oil (the wartime base case). The Jun 17 MOU specifies a 30-day Hormuz-reopening timeline. Conditional on the ceasefire holding — Vance's Israel warning + Trump's "could resume attacks" hedge keep an option-value tail — the base case by mid-July is full Hormuz-normalization, which means $70 oil is plausible by Q3. Iran's pre-war export was ~1.6 mb/d; lifting sanctions adds supply while the deal keeps demand-destruction in check.
The agent control plane is the new battleground. Six closed-source platforms launched in 7 days: Databricks Omnigent, Mastra Harness, Vercel Eve, Microsoft agent-framework v2, Block Builderbot, Thoughtworks Agent/works. Seven+ open-source control-plane repos: kernloop, muster, mcphub, MCP-Gateway, Agent-Machines, agentcsp, agent-harbor, AgentDeck. M&A: Salesforce/Fin $3.6B, Cisco/Astrix $400M, SailPoint/Entro $200M, Palo Alto/Portkey $140M. 5 category-definers with 12-month bets: Databricks, Vercel, Microsoft, kernloop, Giant Swarm muster. An 18-month procurement window before consolidation.
The DeLM result (decentralized coordination) + the VibeThinker result (3B frontier-grade reasoning) + the Vercel-Eve / Microsoft-v2 / Databricks-Omnigent launches (centralized control planes) all converge on the same question: what's the right coordination substrate? Hermes Agent's bet: skills-as-the-control-plane. If we ship a DeLM-style shared-context layer alongside skills, the "AI co-worker on your hardware" pitch becomes the local-stack alternative to Vercel's cloud-hosted control plane.
Today's cluster is the bifurcation of the agent stack into centralized meta-harnesses (Databricks, Vercel, Microsoft) and decentralized coordination (Stanford DeLM, Hermes + skills). The second camp just won the cost battle AND the small-model battle.
SPCX closed its first full week ~+40% above the $135 IPO after giving back the 3-day +47% surge on Thu. The mechanics: options launched Tue Jun 16 with ~1M call contracts on day 1 (top-10 most-traded equity option on Wall Street) — gave bears their first real lever. Day-by-day: Fri $160.95 → Mon $190 → Tue $201.80 (briefly passed Amazon at $2.65T) → Wed $191.82 (-4.95% first red) → Thu ~$185 (second red). Bloomberg characterized the 5% drop as the first real test of bull thesis without lockup mechanics. Acceleration trigger: 10% extra insider unlock if SPCX closes at/above $175.50 on 5 of 10 sessions before Q2 earnings. After Thu's ~$185 close, that's 4 of last 5 sessions above $175.50 — the 5th trigger is near.
Credit: S&P BBB + Moody's Baa1 stable, both assigned Jun 17 — first investment-grade rating for a private space/AI conglomerate at IPO. Fundamental vs price: Morningstar fair value cited at $780B vs ~$2.5T market cap. Former Nasdaq chair: "not trading on fundamentals."