Amodei, Altman and Musk all endorsed pacing the frontier this weekend, with no binding mechanism and no changed ship dates. Median time between frontier releases is now 11 days.
OpenAI's Sam Altman and xAI's Elon Musk publicly backed Amodei's call for slower frontier development on September 13, a rare moment of alignment among three rival labs. The pledge carries an offer of third-party evaluator access but no binding enforcement.
Artificial Analysis puts the median gap between frontier-model releases at 11 days in 2026, down from 37.5 days in 2023. The number gives critics their sharpest counterpoint to a voluntary pause framework.
A planned Iran and Gulf powers meeting on a proposal for the strait was postponed while Saudi Arabia kept its east-west pipeline shut through the weekend. Hot inflation plus the oil shock has moved Fed expectations from cuts toward a possible hike.
A weekly operator deep dive, the full Dojo Pro prompt packs, and the complete prompt database. Founding members lock the launch price forever.
Join the founding list →Anthropic CEO Dario Amodei published an essay on Saturday, September 12, arguing that recent capability jumps have convinced him AI companies must deliberately "pace the frontier" so safety research can catch up with increasingly capable models. The proposal includes a pause or slowdown framework and an offer to...
Continue readingDiplomacy over the Strait of Hormuz stalled as a planned Iran and Gulf powers meeting was postponed, Saudi Arabia kept its east-west pipeline closed after Friday's drone strikes, and an Iranian cargo ship was hit near Qeshm Island, killing one.
Wall Street rose Monday and oil jumped at the open even as hot inflation readings pushed traders to price a Federal Reserve rate hike at this week's meeting, a reversal from the cuts expected only weeks ago.
The same week three lab chiefs discovered the virtue of restraint, the Fed outlook flipped toward tightening; if capital gets dearer for capital-hungry labs shipping a frontier model every 11 days, a voluntary slowdown may turn out to be the cheapest position available.
Talks on a proposal for the strait were postponed while Saudi Arabia held its east-west pipeline shut through the weekend, leaving both diplomatic and physical outlets narrowed. The strike on an Iranian commercial vessel near Qeshm Island, alongside reports of Houthi drones and missiles, adds direct shipping incidents to what was already an escalating threat environment. Oil prices responded on Monday's open.
Hot inflation readings and the Middle East oil shock have moved market expectations from rate cuts to a possible hike at this week's Federal Reserve meeting. Equities rose anyway, consistent with the Greed reading, but the debate has shifted to how far tightening goes. Capital-intensive AI buildouts face a higher cost of money just as Russia's drone strike at the Dorohusk-Yahodyn crossing and a Vilnius airport shutdown add European security risk to the backdrop.
Frontier Models Now Ship Every 11 Days
Read full analysisByteAsk, a Y Combinator F2026 company, targets the code behind trading systems, cars, chips, and operating systems, and it refuses to stop at a diff that merely looks right: it runs the real toolchain, compiler, debugger, and tests on every change. It works from the terminal or inside VS Code, Neovim, and Emacs, and ships fully on-prem, which matters for firmware and finance teams that cannot send source to a cloud API. Start with a bug your test suite already catches and watch whether the agent's loop actually closes it, then widen scope.
Try itGoogle's new canvas feature in Sheets builds a visual, interactive app on top of data that already lives in a sheet, so the internal tracker your team keeps in a tab can become a dashboard or intake tool without a developer. Pick one spreadsheet that people already open daily and prompt for the single view they keep asking for, such as a filterable status board. Keep the sheet as the source of truth and treat the canvas as a front end; if the prompt output drifts from the data, the fix is in the columns, not the prompt.
Try itGemini Spark is Google's agentic layer connected to Chrome and Google Photos: it navigates pages, fills forms, completes browser errands, edits photos, and assembles albums from a natural-language request, working in the background while you do something else. Give it a bounded, reversible first job, for example building an album from last month's trip or filling a form you can review before submission. Hold off on anything involving payments or account changes until you have watched it complete three tasks and read how it reports what it did.
Try itWonder is an AI workspace aimed at freelancers and small teams that responds to a request by spinning up specialised agents and executing the work through to a finished output, rather than returning a chat answer. The failure mode of this category is not bad output but vague input, so define what done looks like in two lines, including format, sources allowed, and what must not be touched, before you hand it anything. Run a research-and-summary job first, then compare the deliverable against your written test, not against your impression of it.
Try itBlackmagic Design's DaVinci Resolve 21.1 adds native integration with Claude, Claude Code, and ChatGPT Codex alongside more than 25 Krokodove tools and workflow improvements, which puts a coding agent next to the timeline for the first time in a mainstream NLE. The practical use is scripting the repetitive parts of post: batch renames, conform checks, marker-driven exports, and Fusion node setups you rebuild every project. Start by asking the assistant to automate one chore you did by hand on your last delivery, and keep the media untouched until the script has run cleanly on a duplicate project.
Try itHand ByteAsk your oldest C or C++ bug. It compiles, debugs and tests in a loop until the change actually holds, so point it at the flaky module you have been avoiding and review the diff rather than writing it.
Turn one spreadsheet into a mini-app. Use Google Sheets canvas to generate an interactive view from a single prompt, then replace one internal dashboard you currently maintain by hand.
Delegate a browser errand to Gemini Spark. Give it one repeatable Chrome task this week, like pulling a vendor comparison or curating an image set, and log where it failed so you know the real boundary of agentic reliability.
Three rival CEOs agreeing to pace the frontier is genuinely unusual, and it still changed nothing anyone can verify today. The release interval fell from 37.5 days to 11 days while these same executives raced each other, and the only concrete concession on the table is third-party evaluator access that each lab grants itself. For builders, plan as if the cadence holds at 11 days, because nothing in this weekend's announcements slows it. Meanwhile the oil shock is repricing the Fed toward a hike, so cheap capital is not the assumption to build on either.
Method names the computation path behind the row, from a closed vocabulary: as published, yfinance close, 03B search corroboration, 04R dossier scoring, not recorded.
A figure is money, a percentage, a spelled magnitude, a multiple, a rate carrying its unit, a score over its denominator, a thousands separated number or an index close. Bare integers and years are not counted, so a phrase like "over 100 integrations" carries no mark. A chart's own scale line is the axis of the level above it and is counted once, with that level.
171 editions so far, one a day. Unsubscribe anytime.