Wolven Tech: Rust consulting. AllSource Prime: durable AI memory → https://nitter.gq/t.co/CfdDxoKhs5 Longhand: AI workflows & training → https://nitter.gq/t.co/Rbml5TOtFF
London, United Kingdom
Joined January 2014
- Tweets4.6K
- Following2.8K
- Followers1.1K
- Likes9.3K
Pinned Tweet
1,000 followers. Thank you.
Marking it by opening something up: I am taking on fractional full stack AI developer work through Wolven Tech Ltd.
Five spots. One is already taken.
How it works below.
The agent changes its answer. Your conversation test carries on with a reply written for the old one. Check the next user turn before trusting its score. 1/5
🤖 Made with AI
ALT Football tactical pieces redirect a pass towards a moved receiver, leaving a faint unused route. Caption: Respond to the reply.
Version the simulator, retain complete conversations and review failures with domain expertise. Separate agent mistakes from simulator mistakes before changing a release decision. An implausible follow-up can be an evaluation defect. 4/5
Which conversation in your test suite becomes incoherent when the agent changes its first answer? Inspect that branch.
langchain.com/blog/how-numan…
5/5
Systemd accepted Rust for new leaf programs, keeping shared core code in C. Support merged October 24; the November 3 explanation sets out the limits.
Rust & AI Weekly #20 covers that decision and three tool evaluations. A thread. 1/6
ALT A silver mechanical wolf inspects a copper connector between a small replaceable module and a large graphite machine.
Sysinfo 0.40.1: Trial for telemetry. Review removed APIs, OsString interface names and Windows load-average handling alongside GPU collection.
Use the refresh workload your collector needs. No author benchmark was independently reproduced here. 5/6
Issue #20 has version dates, licence checks, separate maintenance/adoption signals and proposed acceptance checks. Radar: 98 tools.
The systemd merge and earlier crate releases are dated explicitly.
decebaldobrica.com/blog/2026…
6/6
The profiler shows a blank device lane. Before changing capacity, prove the capture contains the device’s work. Start with a known operation. 1/5
🤖 Made with AI
ALT A wolf silhouette watches an open observation frame as a final event reaches a trace tray. Caption: Complete the capture.
Use a known operation to verify that device events appear. Retain the capture boundary, synchronisation point and warm-up exclusions. Then measure the intended workload with its normal execution behaviour. 4/5
Bring the complete diagnostic trace and representative benchmark to the next performance review. These are proposed checks. No local profiling experiment was run.
huggingface.co/blog/amazon/p…
5/5
A successful Claude request can carry a profile ID the provider never looked up. Trace one customer through a retry and a batch before adding a provider route. 1/5
🤖 Made with AI
ALT A sculpted wolf carries a red identity ribbon along a branching path towards a silver checkpoint. Caption: Keep tenant attribution.
Trace a tenant through retries, batches and supported fallback routes. Compare intended identity with the outgoing profile and the platform’s checks. Attribution loss is a proposed application failure test, not a reported provider defect. 4/5
Which path would you inspect first for lost attribution: retry, batch or fallback? Reuse provider enforcement where it supplies the guarantee you need.
platform.claude.com/docs/en/…
5/5
