Blog
Research findings, compliance guidance, and engineering insights.
A local agent in front of a cloud agent: gpt-oss-120b on llama.cpp
A self-hosted gpt-oss-120b handles two classification jobs on an always-on ops host and decides whether the cloud agent runs at all. The llama.cpp systemd unit, the three-gate ladder, per-job effort settings, and where the line between local and cloud sits.
We Shipped A2A 1.0 Support Twice. The First Time Was Wrong.
Adopt, revert, negotiate: we implemented the A2A 1.0 wire format from the spec, tore it out six weeks later because no deployed client could talk to it, then reintroduced it as a per-request negotiated dialect. What that arc taught us about specs versus ecosystems.
Lessons from an In-Place PostgreSQL 17 to 18 Upgrade (the uuidv7 Default That Did Not Switch Over)
An in-place Aurora 17.9 to 18.3 upgrade kept inserting with the uuidv7 polyfill because column defaults are OID-pinned. The trap, the re-point-then-drop fix, the benchmarks (native is 2.6x faster and packs a 33%-smaller index), and a ~10-minute write-outage that self-healed.
Your people are already using AI. Make being compliant the easy path.
Shadow AI is a path problem, not a discipline problem: 90% of executives feel in control while 52% of workers use unsanctioned AI. Pave the compliant path so agents take it on their own, and the audit-ready record accrues as a byproduct.
Settlement Evidence for Agentic Commerce: What AP2, ACP, and x402 Leave Out of Scope
The 2026 agentic-payment rails sign the authorization moment, then declare dispute evidence and settlement out of scope - in their own spec language. AP2, Verifiable Intent, Stripe SPTs, and x402 mapped against the settlement-evidence layer none of them carries.
Can an AI Agent Be Trusted to Write Its Own Audit Log? We Measured It.
Four production-tier models processed payment batches with forced write failures, then wrote their own audit reports. Three of four claimed success for writes that never happened - up to 47%. The independent signed chain caught every phantom record.
Durable Intent, Measured: We Wiped Four Agents and Asked Them to Finish Their Own Work
A cold-recovery experiment: four agents restarted with zero memory had to discover and finish their own work from the signed ledger alone. All four stalled behind a judgment gate; a deterministic gate flipped the weakest model from 0/3 to 3/3.
The 88,000-Token Crash: Making gpt-oss-120b Survive Its Full 128K Context on Strix Halo
Past ~80k tokens of prefill, llama.cpp on Strix Halo trips the amdgpu GPU watchdog: ring reset, vk::DeviceLostError, core dump. Flash attention does not fix it; -ub 512 does. The crash matrix, the throughput bill, and 3/3 needle-in-a-haystack at 110k tokens.
SCITT as the Article 12 Implementation Pattern for Autonomous AI Agents
EU AI Act Article 12 requires logs conformant to recognised standards. SCITT is the IETF-track standard designed for this evidence shape. The article-by-article mapping, the gaps SCITT explicitly leaves open, and a concrete implementation pattern.
The advisory-lock self-deadlock Postgres can't see
PostgreSQL's deadlock detector walks the row-lock wait graph. Advisory locks held across `await` boundaries can form cycles through application code that the detector cannot see. Reproduction, real diagnostic output, and the structural fix.
Near Frontier-Quality LLM, No Cloud, No Subscription, Unlimited Tokens: gpt-oss-120b on Strix Halo + Ubuntu 26.04
A $2,300 96 GB Strix Halo box runs gpt-oss-120b at ~48 tok/s with a stable full 128K context on llama.cpp + Vulkan. Rewritten after a full rebuild - including what the May version of the post got wrong.
pg-boss in production: footguns we hit and how to avoid them
Four operational footguns we hit running pg-boss at AGLedger for about a year. One is fixed upstream; the other three still bite. Self-contained reproductions, citations, and the patterns we settled on.
Cutting PostgreSQL Audit-Report Query Time 44% with GROUPING SETS and Materialized CTEs
Six aggregations per request became two. Total DB time dropped 44% under sustained load. The anti-pattern, the rewrite, and the honest caveat about why wall-clock p50 did not move.
AGLedger Performance at Scale: A Simulated Quarter, Measured and Verified Offline
A quarter of claims operation on the published 1.2.0 image: 353,861 API calls in 86 minutes, zero failures, then 265k records verified offline in under a minute on a desktop CPU.