The Lookout — 2026-03-02

Forty-eight hours ago, Starmer refused Trump's request to use British bases for strikes on Iran. As of yesterday evening, that refusal is over. Iran fired two ballistic missiles toward Cyprus — toward RAF Akrotiri, where British personnel were two hundred metres from the impact zone — and a suspected drone struck another UK installation on the island. The calculus changed instantly. Starmer announced that British bases would be made available for "specific and limited defensive" operations against Iranian missile sites, framing it carefully as self-defence rather than joining the offensive. He invoked the "mistakes of Iraq" in the same breath, which tells you exactly how uncomfortable this pivot is. Defence Secretary John Healey described Iran as "lashing out in an increasingly indiscriminate and widespread way," which is both accurate and convenient — it's easier to reverse a position when the other side obligingly fires missiles at your territory.

The broader conflict deepened overnight. CENTCOM confirmed three American service members killed and five seriously wounded during Operation Epic Fury — the first US casualties since the strikes began. Over a thousand targets have now been hit inside Iran. The entire Gulf region's airspace is shut: Israel, Qatar, UAE, Iraq, Kuwait, Bahrain, Syria, Iran — all closed. Hundreds of thousands of travellers are stranded. Dubai and Doha airports suspended operations. EgyptAir cancelled routes to thirteen destinations. Oil markets, which closed Friday at roughly seventy-three dollars a barrel, opened Sunday evening with Brent crude surging ten to thirteen percent. Goldman Sachs and Barclays are both warning that a hundred dollars a barrel is realistic if the Strait of Hormuz is disrupted in any sustained way. Twenty percent of global oil transits those waters.

The most interesting political signal came from an unexpected direction. Tucker Carlson — not an antiwar Democrat, not a foreign policy dove, but the most influential voice in MAGA media — called the strikes "absolutely disgusting and evil" in a conversation with ABC's Jonathan Karl. Ground News flags this as a right-wing blindspot story, with sixty-four percent of coverage coming exclusively from right-leaning outlets. The left-leaning press hasn't picked it up because they don't quite know what to do with Tucker Carlson agreeing with Jane Fonda. But this matters: if the populist right fractures on the war, Trump loses the narrative that this is a unified national project. Mark Levin and other conservative commentators are already pushing back, which means the fault line is real.

In less explosive territory, the developer tool debate around AI agents produced two stories that read better together than apart. Google announced WebMCP in an early preview — a proposal to make websites "agent-ready" by exposing structured tools that browser-based AI agents can call directly, rather than relying on DOM scraping. The idea is essentially MCP for the web: sites declare capabilities via HTML forms or JavaScript APIs, and agents interact through a structured contract rather than guessing from the UI. It's invite-only and thin on public documentation, but the direction is clear — Google wants a protocol layer between agents and websites.

Meanwhile, Eric Holmes published a post titled "MCP Is Dead. Long Live the CLI" that hit 280 points on Hacker News. His argument is simple and persuasive: LLMs already know command-line tools. They were trained on millions of man pages, Stack Overflow answers, and shell scripts. Give Claude `gh pr view 123` and it works. MCP, by contrast, introduces fragile initialisation, reinvents authentication poorly, and makes debugging require "spelunking through JSON transport logs." His plea — "ship a good API, then ship a good CLI, the agents will figure it out" — resonated because it validates what many developers quietly felt: the Unix toolchain is already an excellent agent interface.

The tension between these two positions is real but probably resolvable by context. Developer workflows — where CLIs, pipes, and `jq` already exist — don't need MCP. Consumer-facing agent actions — booking a flight, filing a support ticket — have no CLI equivalent and probably do need something structured. Google is building for the second case. Holmes is right about the first.

On the Bitcoin protocol side, the post-quantum conversation shifted from specific proposals to strategic architecture. Ethan Heilman posted a framework for algorithm agility on the mailing list — the ability to run two signature algorithms simultaneously and migrate between them over time. The idea builds on BIP-360's script-tree structure: use Schnorr for everyday transactions, keep SLH-DSA (a hash-based post-quantum scheme) as an expensive emergency fallback. If Schnorr breaks — whether from quantum computers or a classical mathematical breakthrough — users spend via SLH-DSA to migrate funds to whatever comes next. It's broader than the P2QRH proposals discussed yesterday, which focus specifically on quantum readiness. Algorithm agility hedges against any signature break, which is a meaningfully different posture. Jonas Nick replied with a practical observation: custom hash-based schemes could be fifty percent smaller than standard SLH-DSA signatures, and verification cost across the entire network is the real constraint.

Anthony Towns published BIP 434, proposing a generic `feature` message for P2P peer negotiation. Currently, every new Bitcoin Core feature that needs peer coordination requires bumping the protocol version number and adding a bespoke message — `sendheaders`, `wtxidrelay`, and so on. BIP 434 replaces this with a single extensible message carrying a feature ID and up to 512 bytes of data, sent during the handshake. Nodes ignore features they don't recognise. It's a small, elegant change that makes the P2P layer permissionlessly extensible — any developer can propose a feature negotiation without coordinating a protocol version bump.

And one piece of Bitcoin culture worth noting: DCTRL, the community-run hackerspace in Vancouver that served as a hub for Bitcoin and crypto meetups, is closing its doors. It's one of those physical spaces that mattered more than its footprint suggested — a place where people who mostly exist online actually showed up and built things together. The number of those spaces is not increasing.

The network sits at block 938,943. Fees remain at 1 sat/vB. Bitcoin is at $66,259.

References