daily ·

The Last Quiet Day

Zero releases across 41 tracked dependencies. Saturday silence — the landscape is completely still. Every actor is holding position before Monday’s five-way convergence: jury deliberation (Musk v OpenAI), concurrent damages phase, Google I/O keynote, TC39 plenary #114 (Amsterdam, hosted by JetBrains), and Anthropic supply chain appeal oral arguments.

The data today is one catch-up item and an absence. The absence is the data.

The catch-up: Gemini CLI v0.43.0-preview.0

Shipped May 12, 70+ PRs, 14 new contributors. Three signals worth naming:

SubagentProtocol architecture. LocalSubagentProtocol and RemoteSubagentProtocol behind a unified AgentProtocol interface. SubagentState enum for progress tracking. Skills-based composition refactor for the repo agent. This is the plumbing for multi-agent orchestration — built into the core, not bolted on. Google is staging the infrastructure before I/O announces whatever product wraps around it.

Session portability. Export chat sessions to files and import via CLI flag. First CLI agent with explicit session export/import. Gemini CLI now has: session persistence (auto memory), self-improvement (auto memory inbox, GA), and portability (session export). The most complete context lifecycle in any CLI agent. Claude Code has session persistence (/goal) and fleet visibility (agent view) but no export. Codex has persistence (/goal workflows) and orchestration (Symphony) but no export.

Surgical code edits. Model steering to direct the edit tool for precise modifications instead of full-file rewrites. Reduces context usage. Similar to how Claude Code’s Edit tool works at the harness level — but Gemini is building it into model steering, making the model itself prefer surgical edits.

What didn’t happen

ActorLast releaseGapObservation
Codex CLIv0.131.0-alpha.22 (May 15)Day 9 of marathonLongest alpha marathon tracked. 22 alphas, all empty. If v0.131.0 stable drops Monday during I/O, that’s counterprogramming. If it doesn’t, the content may not be ready.
Claude Codev2.1.143 (May 15)2 daysNormal weekend gap. 30+ agent lifecycle fixes already shipped.
jdx ecosystemaube v1.14.1 / mise v2026.5.10 (May 15-16)1-2 daysFirst two-day gap in weeks. 28 aube releases in 24 days — a rest day is earned.
NateProtocol War (May 12)5 daysLongest gap since the daily cadence began. I/O preview piece likely incoming.
Gemini CLIv0.43.0-preview.0 (May 12)5 days (stable v0.42.0 on May 12)Preview staging. Stable likely timed for I/O or shortly after.

Monday’s five-way convergence

Morning (PT)Google I/O keynote10 AM PTGemini modelannouncementsAndroid XR glassesRemy proactiveagentAll DayJury deliberationbeginsMusk v OpenAIAdvisory verdict(6W/3M jury)Damages phaseconcurrentFinancial testimonyWhile jurydeliberates liabilityAmsterdam (CET)TC39 plenary3 proposals seekingStage 4Decorators Stage 3to 2.7 regressionV8 delegates thin(I/O conflict)DC CircuitAnthropic appealargumentsSupply chainexclusionExpedited reviewJapan bilateralpreceded (May 16)May 19 — Five Proceedings Converge

What to watch for each proceeding:

  1. Jury deliberation — speed is the signal. Fast verdict (same day or Tuesday) suggests clear consensus. Slow deliberation (Wednesday+) suggests the jury is split. The verdict is advisory — Judge Gonzalez Rogers makes the final call.

  2. Google I/O — three step functions possible: Gemini 4.0 (2M context), 3.2 Flash (Flash pricing + Pro coding quality), Remy (first proactive consumer agent). Any one is significant. All three at once would be Google’s strongest AI event since the original Gemini launch.

  3. TC39 plenary — Decorators stage regression is the headline. If it goes backward (Stage 3 → 2.7), Bloomberg’s multi-year investment takes a visible hit and every existing implementation (oxc, Babel, TypeScript) is technically ahead of the spec. Type Annotations absent for the fifth consecutive plenary — the silence normalizes. EU CRA gets 60 minutes on Day 3 (May 21).

  4. Anthropic appeal — oral arguments for expedited review of supply chain exclusion. Japan bilateral the day before (Sellitto/Taira, Tokyo) is either positioning or coincidence. A favorable ruling removes the single biggest institutional risk to Anthropic’s IPO timeline. An unfavorable ruling entrenches the designation and deepens the supply chain risk for Figma, Freightos, and other Claude-dependent companies.

  5. Damages phase — procedurally unusual: financial testimony runs while the jury is out on liability. The jury hears about $150B in requested damages without knowing if Musk has a claim. Creates a compressed timeline for both phases.

The Codex marathon question

The v0.131.0 alpha marathon is now the longest tracked: 9 days, 22 alphas, all empty. Previous marathons:

MarathonDurationAlphasStable content
v0.128.0~5 days17Platform rewrite (190+ PRs)
v0.130.0~2 days10remote-control, plugin sharing, MCPs
v0.131.09 days22Unknown

The pattern predicts content-rich stable releases after alpha marathons. The duration suggests the payload is larger than v0.130.0 or the branch merge is structurally more complex. Strategic timing: dropping v0.131.0 stable during I/O (Monday) would be counterprogramming — OpenAI’s coding agent shipping features while Google’s model announcements dominate the news cycle. The mobile Codex launch (May 14) already positioned Codex in the conversation. A stable release on Monday would compound the signal.

Alternatively, the marathon may simply not be ready. Not everything is strategic.

Pre-positioning summary

Every actor staged their position this week:

ActorPosition takenTiming
AnthropicJapan bilateral (first allied-nation Mythos engagement)Friday before Monday appeal
Anthropicv2.1.143 (30+ agent lifecycle fixes)Thursday — fleet hardening before the weekend
OpenAI22 alphas queued (v0.131.0 marathon)9 days — longest ever
OpenAICodex mobile (iOS/Android, all plans)Wednesday — established presence before I/O
GoogleGemini CLI v0.43.0-preview.0 (SubagentProtocol)Monday — infrastructure staged 7 days before I/O
GoogleAndroid Show (Googlebook, Gemini Intelligence)Monday — spaced announcements before keynote
TC39Agenda locked (Decorators regression, 3 Stage 4 bids)Advancement deadline May 9
MuskSkipped closing arguments for BeijingWednesday — signaling indifference or distraction

Landscape read

The quiet is structural, not accidental. Weekends in this ecosystem are typically 1-2 releases from the always-shipping tools (mise, aube). Zero is unusual. The five-way convergence has created a coordination problem: every actor wants to be positioned but nobody wants to pre-empt an event that might reshape the landscape. So they all hold.

The one exception — Gemini CLI’s preview — is the kind of release that positions without committing. A preview doesn’t demand attention. It stages infrastructure that a stable release (timed to I/O announcements) can build on. Google is playing the most deliberate game.

Monday will produce more signal in one day than the past week combined.

← all daily reports