self.md radar — 2026-10-30
Jules goes read-only in November
Jules is winding down. Google has published an export procedure alongside two different deadlines.
1. Jules becomes read-only before its December shutdown
Google’s October 29 notice gives users until November 11, 2026 to finish sessions. after that, Jules becomes read-only until the service winds down on December 31.
the code, branches and pull requests already in GitHub stay there. the export is for what Jules holds separately: configuration and session history, including prompts, plans, activity logs and patches. repository memories and setup scripts are part of it too.
one exported file needs particular care: env-variables.json contains the environment-variable values. Google’s notice tells users to keep the folder private and out of repositories. an export intended to help you leave can also carry the credentials you gave the service.
the published procedure uses Jules Tools and a script that calls undocumented configuration and session endpoints. the notice says to report failed calls rather than guess at fixes. this export has not been tested here.
reading: Google’s Jules turndown notice and export procedure
2. ChatGPT’s API credits leave Codex API-key calls out
the exclusion is explicit: Codex usage through API keys is not covered. the credits also leave ChatGPT and signed-in Codex usage limits unchanged.
OpenAI’s benefit lets eligible Pro and Business owners link a subscription to an API organisation they own. the Pro tiers grant $100, $200 or $500 per month. Business grants $20 per paid Standard seat or $100 per paid Premium seat, capped at $500 per workspace per month.
the paid subscription must be at least seven days old. both accounts need the same email address, and claiming requires ownership on both sides. activate once; later eligible renewals grant credits automatically.
the first grant is prorated. unused credits expire at the billing-cycle boundary without rolling over; annually prepaid Business workspaces receive an annual grant with annual expiry. supported API usage spends free credits before purchased credits. when neither balance remains, requests fail with 429 and credit_balance_exhausted.
reading: OpenAI’s eligibility, claim flow and billing rules
3. Amazon calls AI-usage metrics flawed, then keeps them
Amazon’s engineering memo calls input and throughput metrics deeply flawed and warns against making them targets. it also says those measures correlate with useful outputs and endorses their careful use.
the memo asks managers to investigate the 10th percentile of AI adoption for missed opportunities, and the 90th percentile for whether heavy use produces matching delivery gains. outcomes remain the stated goal; usage remains a diagnostic.
there is a usable exception in the same document: engineers whose work does not benefit from current tools should document why, share the finding and bring it to a manager. leaders are expected to have used the tools on production work and be able to describe a task where they helped and one where they did not.
reading: Amazon engineering leadership’s SDE memo
4. a Tapestries canvas can leave as a zip
photos, audio and video sit on a canvas, with source objects that a reader can open. Tapestries lets authors keep that arrangement private, publish it, or allow someone else to copy and revise it.
Internet Archive’s October 29 introduction follows an October 21 launch and a limited release in 2025. Bob Stein and Dan Visel developed the interface; Ben Vershbow now directs the project.
Vershbow says a whole tapestry can be exported as a zip and could be deposited in the Internet Archive as its own cultural object. the announcement does not specify whether every external media embed survives offline.
reading: Internet Archive’s Tapestries announcement
left on the table
Vershbow proposes Tapestries assignments as an alternative to research papers vulnerable to AI shortcuts. the announcement supplies classroom accounts, not a comparative test of shortcutting or learning outcomes.