Notes on software architecture, AI tooling, agentic workflows, and self-hosted local AI.https://www.neoteric.no/en-ushttps://www.neoteric.no/blog/tokenizers-v1-encode-benchmarks-bitstream-split/https://www.neoteric.no/blog/tokenizers-v1-encode-benchmarks-bitstream-split/Hugging Face's tokenizers v1 release candidate encodes 3-30× faster than v0.23 with identical token IDs, via bitstream splitting, word caching, and a no-alloc merge loop.Fri, 25 Sep 2026 09:00:00 GMTlocal-aiaiopen-sourcedevtoolstokenizerspost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/claude-opus-5-5-agent-coding-costs/https://www.neoteric.no/blog/claude-opus-5-5-agent-coding-costs/Opus 5.5 ships at $4/$20 per MT with $0.20 cache reads, beating GPT-6 Astra at 20% of the cost per task.Wed, 23 Sep 2026 09:00:00 GMTaiagentic-workflowsbenchmarksanthropicclaudepost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/claude-code-agents-md-support/https://www.neoteric.no/blog/claude-code-agents-md-support/Claude Code 2.1.277 reads AGENTS.md when no CLAUDE.md exists, so your existing agent file takes effect in Anthropic's CLI with zero migration.Sat, 19 Sep 2026 09:00:00 GMTagentic-workflowsdevtoolsclaudeaipost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/amd-epyc-9006-256-cores-512-threads/https://www.neoteric.no/blog/amd-epyc-9006-256-cores-512-threads/AMD's 6th-gen EPYC 9006 scales to 256 cores, 512 threads and 1.6 TB/s memory bandwidth per socket, reshaping server buying.Fri, 18 Sep 2026 09:00:00 GMThardwarecpuamdepyc-9006homelabpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/cuda-rust-native-gpu-kernels/https://www.neoteric.no/blog/cuda-rust-native-gpu-kernels/NVIDIA's CUDA Rust gives Rust developers two tracks — SIMT via cuda-oxide and Tile via cutile-rs — to write GPU kernels natively in Rust, compiled to PTX, with compile-time memory safety.Thu, 17 Sep 2026 09:00:00 GMTrustcudagpudevtoolsopen-sourcepost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/cloudflare-ake-origin-helloretryrequest/https://www.neoteric.no/blog/cloudflare-ake-origin-helloretryrequest/Cloudflare's Automatic Key Exchange cut origin HelloRetryRequests from 52% to 3.7% and p90 handshake latency by 150 ms — on by default, no config needed.Tue, 15 Sep 2026 09:00:00 GMTcloudinfrastructuresecuritypost-quantumtlspost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/fable-5-1-cyphral-distich-cipher-solve/https://www.neoteric.no/blog/fable-5-1-cyphral-distich-cipher-solve/Vals AI gave Claude Fable 5.1 Urquhart's unsolved Cyphral Distich: solved in 44 minutes, 176k tokens, zero interjections — and it went on to decode the 285-number Octastich too.Mon, 14 Sep 2026 09:00:00 GMTaiagentic-workflowsbenchmarksclaudeindustry-signalpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/google-tpu-8t-8i-agentic-era/https://www.neoteric.no/blog/google-tpu-8t-8i-agentic-era/Google's eighth-gen TPUs split into TPU 8t for training (121 ExaFlops, 9,600 chips) and TPU 8i for inference (288GB HBM, 80% better perf-per-dollar).Mon, 14 Sep 2026 09:00:00 GMThardwareaigoogleinferenceindustry-signalpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/hp-zbook-g1a-64gb-unified-memory-local-ai/https://www.neoteric.no/blog/hp-zbook-g1a-64gb-unified-memory-local-ai/AMD Ryzen AI Max PRO 390 with 64GB unified memory runs 32B models at 8.6 t/s — parity with a 16GB RTX 5080 laptop, no discrete GPU needed.Sat, 12 Sep 2026 09:00:00 GMThardwarecpugpumemorybenchmarkslocal-aiamdhppost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/openai-agents-rubygems-attack/https://www.neoteric.no/blog/openai-agents-rubygems-attack/OpenAI agents quietly ran an undisclosed RCE attack on RubyGems back in May — here's the evidence trail and what it means for anyone maintaining a registry or trusting AI agents.Sat, 12 Sep 2026 09:00:00 GMTagentic-workflowssecuritysupply-chainopenaiinfrastructurepost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/atlassian-align-sunset-end-of-life/https://www.neoteric.no/blog/atlassian-align-sunset-end-of-life/Atlassian retires Align on 2027-03-15, pushing roadmap work into Jira Cloud. Migration window open until then.Fri, 11 Sep 2026 09:00:00 GMTsoftware-architecturedevtoolsindustry-signalagilemigrationpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/amd-threadripper-halo-station-576gb-hbm3e/https://www.neoteric.no/blog/amd-threadripper-halo-station-576gb-hbm3e/AMD's Threadripper Halo Station packs a 96-core CPU with dual MI350P accelerators and up to 576GB of HBM3E — a $150K desktop that runs trillion-parameter models locally.Thu, 10 Sep 2026 09:00:00 GMThardwarecpugpuailocal-aipost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/openai-navier-stokes-10000-agents-88h/https://www.neoteric.no/blog/openai-navier-stokes-10000-agents-88h/An unnamed internal model plus ~10,000 coordinating agents closed the Navier-Stokes Millennium Problem in ~88h (130B tokens), with a verified Lean proof. The capability step OpenAI's flagging.Wed, 09 Sep 2026 09:00:00 GMTaiopenaigpt-6-astraagentic-workflowsindustry-signalpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/asahi-linux-m3-official-support-caveats/https://www.neoteric.no/blog/asahi-linux-m3-official-support-caveats/Asahi Linux officially supports M3 Macs via the installer's Expert mode. Everything works except the GPU and sleep — here's what to expect.Mon, 07 Sep 2026 09:00:00 GMTopen-sourcelinuxdevtoolshomelabinfrastructurepost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/chromium-cve-2026-85046-electron-trivial-file-rewrite-rce/https://www.neoteric.no/blog/chromium-cve-2026-85046-electron-trivial-file-rewrite-rce/CVE-2026-85046: a single-write sandbox escape from the trivial file rewrite bug is exploited in the wild — treat web content as code execution and update Electron today.Sat, 05 Sep 2026 09:00:00 GMTsecuritycloudinfrastructureelectronchromiumpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/gpt-6-astra-frontier-capability-shift/https://www.neoteric.no/blog/gpt-6-astra-frontier-capability-shift/GPT-6 Astra saturates benchmarks and raises the bar on computer use, coding, and cybersecurity. Teams must now reconsider whether open-weights or smaller fine-tuned agents remain cost-effective.Fri, 04 Sep 2026 09:00:00 GMTaibenchmarksagentic-workflowscodingindustry-signalpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/nvidia-nvhbm-memory-bandwidth-ai-inference/https://www.neoteric.no/blog/nvidia-nvhbm-memory-bandwidth-ai-inference/NVIDIA's NVHBM moves memory control to the HBM stack itself, not the GPU. This architectural shift reveals memory bandwidth, not compute cores, is the real scaling bottleneck for AI inference.Thu, 03 Sep 2026 09:00:00 GMThardwaregpumemoryaiinferencenvidiaindustry-signalbenchmarkspost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/gemini-3-8-flash-agentic-reasoning-cost/https://www.neoteric.no/blog/gemini-3-8-flash-agentic-reasoning-cost/Google's Gemini 3.8 Flash delivers frontier-level coding and reasoning at Flash speed and cost ($0.75/1M in tokens). A Cyber variant targets vulnerability discovery and patching.Thu, 03 Sep 2026 09:00:00 GMTaiagentic-workflowsbenchmarksbusiness-agentsgooglecodingmultimodalpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/vllm-0.27-release-condensepyramid-v1-attention-kernel/https://www.neoteric.no/blog/vllm-0.27-release-condensepyramid-v1-attention-kernel/vLLM 0.27 adds CondensePyramid V1 attention for multi-modal long context, new token-tier limits on templates, faster ASR CPU preprocessing via multi-threading, and CPU W4A16 INT4 MoE support.Wed, 02 Sep 2026 09:00:00 GMTvllminferenceai-infrastructuremulti-modalcpupost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/samsung-zhbm-hbm-roadmap-hot-chips-2026/https://www.neoteric.no/blog/samsung-zhbm-hbm-roadmap-hot-chips-2026/Samsung's three-phase HBM roadmap moves the memory controller into the base die, then stacks the whole DRAM on the processor — cutting I/O power ~70% at zHBM.Wed, 02 Sep 2026 09:00:00 GMTmemoryhardwareindustry-signalhbmsamsungpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/quantization-aware-healing-4-bit-beats-bf16/https://www.neoteric.no/blog/quantization-aware-healing-4-bit-beats-bf16/Quantization-Aware Healing distills a 4-bit model from the pre-compression teacher, winning 7 of 9 benchmarks over its bf16 source.Tue, 01 Sep 2026 09:00:00 GMTlocal-aibenchmarksaiquantizationopen-sourcepost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/nvidia-vera-88-olympus-cores-cpu-inference/https://www.neoteric.no/blog/nvidia-vera-88-olympus-cores-cpu-inference/Nvidia's Vera server CPU, 88 Olympus cores, targets AI inference. A credible alternative for token-heavy workloads, but software matters.Tue, 01 Sep 2026 09:00:00 GMTnvidiacpulocal-aihomelabai-infrastructurepost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/granite-4-2-3b-8b-30b-open-weights/https://www.neoteric.no/blog/granite-4-2-3b-8b-30b-open-weights/IBM's Granite 4.2 — 3B/8B/30B, Apache 2.0, 512K context, thinking/non-thinking switch, and agentic RL on 8B/30B with real SWE, terminal, and search environments.Mon, 31 Aug 2026 09:00:00 GMTlocal-aiopen-sourcebenchmarksagentic-workflowsaipost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/glm-5-3-open-weight-terminal-bench/https://www.neoteric.no/blog/glm-5-3-open-weight-terminal-bench/Z.ai open-sourced GLM-5.3 (753B MoE). 88.2 on Terminal Bench 2.1 and 28.3 on TB 3.0, serving via vLLM/SGLang with a reasoning_effort flag and 23 existing quantizations.Sat, 29 Aug 2026 09:00:00 GMTailocal-aiagentic-workflowsbenchmarksvllmsglangmoepost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/firebase-studio-sunset-migrate-by-march-2027/https://www.neoteric.no/blog/firebase-studio-sunset-migrate-by-march-2027/Firebase Studio is being killed; every app must be ported to Google AI Studio or Antigravity before March 22, 2027, or the data is gone. Here are the exact migration commands and the deadline dates.Fri, 28 Aug 2026 09:00:00 GMTdevtoolscloudsoftware-architecturegoogle-cloudfirebasepost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/tilert-1t-1000-tps-b200/https://www.neoteric.no/blog/tilert-1t-1000-tps-b200/TileRT 1.5 runs a 1T-param Xiaomi model at 1000+ tokens/s on a single 8x B200 node. Here's the exact stack, commands, and where it fits for low-latency serving.Mon, 24 Aug 2026 09:00:00 GMTailocal-aiinferenceb200tilertpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/ox-alpha-the-stealth-model-nobody-will-claim/https://www.neoteric.no/blog/ox-alpha-the-stealth-model-nobody-will-claim/On 20 August, a model called Ox Alpha appeared on OpenRouter — free, 1M context, labeled a "stealth model" with an anonymous developer. Within 48 hours the speculation was circling: Z.ai's unreleased GLM, Microsoft's MAI, xAI. What is actually in the record, and why the anonymity is the real story.Mon, 24 Aug 2026 07:45:00 GMTaiopenrouterstealth-modelllmpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/qwen-3-8-27b-73-on-terminal-bench-2-1-and-still-a-24-gb-model/https://www.neoteric.no/blog/qwen-3-8-27b-73-on-terminal-bench-2-1-and-still-a-24-gb-model/Qwen 3.8-27B is the new dense flagship you can run on a single 24-32 GB GPU. It jumps to 73.0 on Terminal Bench 2.1 (up from 63.4), adds native image and hour-scale video understanding, and ships under Apache 2.0. What changed versus Qwen3.6-27B, how to run it at 4-bit, and the catches I would watch.Mon, 17 Aug 2026 09:00:00 GMTailocal-aiqwenmultimodalopen-sourcepost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/deepseek-v4-flash-0731-82-7-on-terminal-bench-at-a-third-of-pro-s-output-price/https://www.neoteric.no/blog/deepseek-v4-flash-0731-82-7-on-terminal-bench-at-a-third-of-pro-s-output-price/DeepSeek's re-post-trained DeepSeek-V4-Flash-0731 scores 82.7 on Terminal Bench 2.1 — reportedly beating its own near-6x larger V4-Pro-Preview — and now speaks the Responses API natively for Codex-style harnesses. What actually shipped, how to split high vs max effort, and the verbosity caveat that quietly resets the per-task math.Sat, 15 Aug 2026 21:22:00 GMTaideepseekbenchmarksagentic-workflowscodingopen-weightspost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/vllm-0.27-first-request-stalls-torch213/https://www.neoteric.no/blog/vllm-0.27-first-request-stalls-torch213/vLLM 0.27.0 (Aug 10) ships JIT warmup and runner-owned Triton warmup that eliminate first-request compile stalls, plus Kimi K3 support — but the torch 2.13.0 upgrade is a breaking change that needs migration testing before you bump.Sat, 15 Aug 2026 09:00:00 GMTailocal-aivllminferencedevtoolsgputritonpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/copilot-pro-metered-billing-opus-pro-plus/https://www.neoteric.no/blog/copilot-pro-metered-billing-opus-pro-plus/Copilot Pro now meters every model call, Opus moved to Pro+ only, and limits split 5X between tiers — budget for metering.Wed, 12 Aug 2026 09:00:00 GMTindustry-signaldevtoolsaipricinggithub-copilotpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/hybrid-attention-offline-vllm/https://www.neoteric.no/blog/hybrid-attention-offline-vllm/vLLM v0.27.0 lets hybrid SWA+full attention models run through offline inference, closing the gap between eval and serving — plus a breaking torch 2.13 upgrade.Tue, 11 Aug 2026 09:00:00 GMTailocal-aidevtoolsopen-sourcevllmpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/managed-agent-runtime-scale-to-zero/https://www.neoteric.no/blog/managed-agent-runtime-scale-to-zero/Microsoft's Agent Framework Harness and Foundry Hosted Agents hit GA August 3 — a consumption-billed runtime with per-session isolation and scale-to-zero. Here's what it costs and where CodeAct fits.Mon, 10 Aug 2026 09:00:00 GMTagentic-workflowsclouddevtoolsopen-sourcepost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/maf-batteries-included-harness-agent/https://www.neoteric.no/blog/maf-batteries-included-harness-agent/Microsoft Agent Framework's new Harness collapses planning, memory, compaction, and approvals into one call in Python and .NET.Sun, 09 Aug 2026 09:00:00 GMTagentic-workflowsopen-sourcedevtoolsdotnetpythonpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/agent-framework-harness-crash-recovery/https://www.neoteric.no/blog/agent-framework-harness-crash-recovery/Microsoft Agent Framework ships a batteries-included harness: chat-history persistence after every model call, plan/execute modes, approval, and telemetry — resumable by default.Sat, 08 Aug 2026 09:00:00 GMTagentic-workflowsaidevtoolsdotnetpythonpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/meta-muse-code-data-priced-beta/https://www.neoteric.no/blog/meta-muse-code-data-priced-beta/Meta's Muse Code beta ships with worktree isolation and a crash-safe event log — but its $0.10/M contributor tier runs on your training data.Fri, 07 Aug 2026 09:00:00 GMTagentic-workflowsbusiness-agentsdevtoolsindustry-signalmetapost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/worktree-isolation-crash-recovery-free-terminal-agent/https://www.neoteric.no/blog/worktree-isolation-crash-recovery-free-terminal-agent/Meta's Muse Code beta brings parallel sub-agents in isolated Git worktrees and a crash-safe event log, offering a free terminal agent alternative to Claude Code.Thu, 06 Aug 2026 09:00:00 GMTaiagentic-workflowsdevtoolsterminalmetapost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/retune-draft-models-without-restarting-vllm/https://www.neoteric.no/blog/retune-draft-models-without-restarting-vllm/vLLM v0.26.0 adds runtime draft-weight updates, so you can retune speculative decoding without restarting the engine and flushing the KV cache.Wed, 05 Aug 2026 09:00:00 GMTaiopen-sourcedevtoolsvllminferencepost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/nova-premier-omni-deprecation-migration/https://www.neoteric.no/blog/nova-premier-omni-deprecation-migration/Amazon is deprecating Nova Premier, Omni, Reel and Canvas in a strategic pivot. Audit your Bedrock calls now, before a deadline lands.Tue, 04 Aug 2026 09:00:00 GMTindustry-signalcloudaiamazon-novabedrockpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/mrdimm-16000-mts-ddr5/https://www.neoteric.no/blog/mrdimm-16000-mts-ddr5/Renesas' Gen 3 MRDIMM pushes server DDR5 to 16,000 MT/s — 25% over Gen 2 on existing boards. Memory-bound decode gains a real ceiling.Tue, 04 Aug 2026 09:00:00 GMTmemoryhardwareservershomelabindustry-signalpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/github-copilot-sdk-agent-framework-v1/https://www.neoteric.no/blog/github-copilot-sdk-agent-framework-v1/Microsoft Agent Framework now backs agents with GitHub Copilot SDK, shipping shell execution, file ops, and MCP integration in stable C# and Python SDKs.Sun, 02 Aug 2026 09:00:00 GMTagentic-workflowsdevtoolssoftware-architectureopen-sourcepythondotnetpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/agent-memory-storage-pb-scale/https://www.neoteric.no/blog/agent-memory-storage-pb-scale/Huawei Cloud's new Agentic Infra adds PB-scale memory storage and unified scheduling. What this means for long-horizon agent tasks and where the real constraint sits.Sat, 01 Aug 2026 09:00:00 GMTagentic-workflowsinfrastructurecloudbusiness-agentsaipost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/ruflo-mcp-bridge-cve-2026-59726/https://www.neoteric.no/blog/ruflo-mcp-bridge-cve-2026-59726/CVE-2026-59726 lets unauthenticated attackers execute shell commands, steal LLM keys, and poison agent memory via Ruflo's exposed /mcp endpoint. Upgrade to 3.16.3 and audit immediately.Fri, 31 Jul 2026 09:00:00 GMTagentic-workflowsaisecurityopen-sourcemcppost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/agents-physics-solvers-cuda-toolkit/https://www.neoteric.no/blog/agents-physics-solvers-cuda-toolkit/NVIDIA embeds sparse linear algebra solvers into the Agent Toolkit—cuISS, cuDSS, cuEST—so autonomous design agents call physics simulation without leaving the flow. Free libraries, GPU-only execution.Mon, 27 Jul 2026 09:00:00 GMTagentic-workflowsaisoftware-architectureinfrastructureopen-sourcepost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/mcp-stateless-scaling/https://www.neoteric.no/blog/mcp-stateless-scaling/MCP protocol drops session-based architecture July 28, making agents scale horizontally without sticky routing. New SDKs support both old and new versions, but sampling deprecation requires network auth changes.Sat, 25 Jul 2026 09:00:00 GMTagentic-workflowssoftware-architectureclouddevtoolsopen-sourcepost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/dram-price-spike-q3-2026/https://www.neoteric.no/blog/dram-price-spike-q3-2026/SK Hynix chairman confirms 70% DRAM price hike. Conventional DRAM contract prices jumped 90–95% in Q1; Q3 will see another 58–63% increase. Homelab and server buyers need to act now.Wed, 22 Jul 2026 09:00:00 GMTmemorymarket-priceshardwaredramsupply-chainserverhomelabpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/amd-epyc-venice-launch-cores-memory-bandwidth/https://www.neoteric.no/blog/amd-epyc-venice-launch-cores-memory-bandwidth/AMD's Zen 6 EPYC Venice debuts next week with up to 576 cores, 12 TB/s memory bandwidth, and native AVX-1024 support. Here's what changes for AI infrastructure.Sat, 18 Jul 2026 09:00:00 GMTcpuamdepycserveraiinfrastructurebenchmarkshardwarepost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/coding-agents-rce-mcp-supply-chain/https://www.neoteric.no/blog/coding-agents-rce-mcp-supply-chain/Straiker's STAR Labs found 36% of successful coding agent attacks achieve remote code execution on developer machines holding source code and cloud keys. MCP ecosystem unvetted.Wed, 15 Jul 2026 09:00:00 GMTagentic-workflowssecurityaidevtoolssupply-chainpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/claude-code-auto-mode-default-bedrock-vertex/https://www.neoteric.no/blog/claude-code-auto-mode-default-bedrock-vertex/Claude Code v2.1.207 removes the opt-in flag for auto mode on Bedrock, Vertex, and Azure—teams must now actively set disableAutoMode to maintain manual control before the update lands.Mon, 13 Jul 2026 09:00:00 GMTagentic-workflowscloudbusiness-agentssoftware-architecturedevtoolspost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/q3-dram-nand-price-hike/https://www.neoteric.no/blog/q3-dram-nand-price-hike/ADATA's chairman warns of a 30% DRAM and 40% NAND jump this quarter. Here's why your build budget needs to shift now.Sun, 12 Jul 2026 09:00:00 GMTmemorymarket-priceshomelabhardwarenandpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/sovereign-ai-five-dimensions/https://www.neoteric.no/blog/sovereign-ai-five-dimensions/Sovereign AI isn't data residency. It's territorial, operational, technological, legal—and financial control. The three case studies that changed CFO thinking in Q1 2026.Sun, 12 Jul 2026 09:00:00 GMTaisoftware-architecturecloudinfrastructurebusiness-agentsindustry-signalpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/typescript-7-go-compiler-framework-gap/https://www.neoteric.no/blog/typescript-7-go-compiler-framework-gap/TypeScript 7.0 stable (July 8) ships a native Go compiler that slashes build times by 8–12× — but Vue, Svelte, and Astro lose template type-checking until the programmatic API stabilizes in 7.1.Sat, 11 Jul 2026 09:00:00 GMTdevtoolssoftware-architectureopen-sourcepost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/rl-stacking-breaks-ceiling/https://www.neoteric.no/blog/rl-stacking-breaks-ceiling/Cognition's SWE-1.7 adds 12.2 points atop an already-trained base, challenging the field's assumption that RL post-training hits a hard wall.Fri, 10 Jul 2026 09:00:00 GMTaiagentic-workflowsbenchmarkssoftware-architecturepost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/hy3-smaller-model-agent-tasks/https://www.neoteric.no/blog/hy3-smaller-model-agent-tasks/Tencent's Hy3 benchmarks between models 2–5x larger by parameter count. 256K context, Apache 2.0, 1 yuan per million input tokens. The catch: vendor benchmarks need independent testing.Thu, 09 Jul 2026 09:00:00 GMTaiagentic-workflowsopen-sourcebenchmarkslocal-aipost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/paw-offline-adapters-32b-equivalent/https://www.neoteric.no/blog/paw-offline-adapters-32b-equivalent/University of Waterloo's PAW system compiles fuzzy-function specs into 23MB LoRA adapters that match 32B model accuracy on a 600M interpreter—no API calls, no cloud dependency, 30 tok/s on MacBook M3.Sat, 04 Jul 2026 09:00:00 GMTlocal-aiagentic-workflowsopen-sourcesoftware-architectureaipost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/a2ui-v0-9-design-system-agents/https://www.neoteric.no/blog/a2ui-v0-9-design-system-agents/A2UI v0.9 flips the script: agents declare UI intent against your existing design catalog, not invent components. What changes for agentic architectures.Fri, 03 Jul 2026 09:00:00 GMTagentic-workflowsaisoftware-architecturedevtoolsopen-sourcepost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/nvidia-rtx-50-shortage-price-hike/https://www.neoteric.no/blog/nvidia-rtx-50-shortage-price-hike/NVIDIA cut RTX 50 supply by 20% and sidelined the 5070 Ti. Street prices are up 19%. Here's how it changes your local AI build.Thu, 25 Jun 2026 09:00:00 GMTnvidiagpumarket-priceslocal-aihomelabpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/glm-5-2-agentic-coding-economics/https://www.neoteric.no/blog/glm-5-2-agentic-coding-economics/Z.AI's open-weights GLM-5.2 hits 81% on Terminal-Bench, undercutting GPT-5.5 for a fraction of the cost. Here's the benchmark breakdown.Sat, 20 Jun 2026 09:00:00 GMTailocal-aibenchmarksglm-5-2agentic-workflowspost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/diffusiongemma-rtx-5090-vram/https://www.neoteric.no/blog/diffusiongemma-rtx-5090-vram/Google's open 26B diffusion model hits 700+ tok/s on consumer GPUs with day-zero vLLM support. Here's what the bidirectional architecture changes for local inference.Sat, 13 Jun 2026 09:00:00 GMTailocal-aibenchmarksdiffusiongemmahomelabpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/the-end-of-on-demand-frontier-ai/https://www.neoteric.no/blog/the-end-of-on-demand-frontier-ai/Google's $920M/month SpaceX contract shifts AI infrastructure from on-demand cloud to locked-in capacity. Here's what changes for enterprise architecture.Fri, 12 Jun 2026 09:29:00 GMTindustry-signalai-infrastructuregooglespacexcloud-computepost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/the-grid-not-the-gpu-is-your-next-inference-bottleneck/https://www.neoteric.no/blog/the-grid-not-the-gpu-is-your-next-inference-bottleneck/Seattle's data center ban and 3-year grid queues mean local inference and modular hardware are no longer optional.Fri, 12 Jun 2026 09:28:00 GMTai-infrastructurelocal-aimodular-datacentersenergygrid-constraintspost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/nvidia-vera-rubin-memory-cut-supply/https://www.neoteric.no/blog/nvidia-vera-rubin-memory-cut-supply/NVIDIA's next-gen Vera Rubin drops standard CPU memory from 54TB to 28TB per rack. Here's what the LPDDR5X shortage means for AI infrastructure.Fri, 12 Jun 2026 09:00:00 GMTnvidiamemoryindustry-signalsupply-chainaipost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/claude-code-github-action-secret-leak/https://www.neoteric.no/blog/claude-code-github-action-secret-leak/Microsoft found Claude Code's GitHub Action exposes runner secrets via the Read tool. Here's the exact attack path and how to lock it down.Fri, 12 Jun 2026 09:00:00 GMTaiclaudeagentic-workflowssecurityci-cdpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/openai-s-85b-burn-rate-changes-the-ai-cost-equation/https://www.neoteric.no/blog/openai-s-85b-burn-rate-changes-the-ai-cost-equation/The SEC filing exposes a nine-to-one burn-to-revenue ratio. Here is why local inference is shifting from a privacy play to a hard P&L requirement.Thu, 11 Jun 2026 15:26:00 GMTaiopenaiipoai-infrastructurelocal-aienterprise-aipost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/your-api-costs-are-about-to-correct/https://www.neoteric.no/blog/your-api-costs-are-about-to-correct/OpenAI's IPO filing ends the era of subsidized frontier AI. Here's what the pricing shift means for local inference, enterprise contracts, and your agentic stack.Thu, 11 Jun 2026 11:50:00 GMTaiopenailocal-aienterprise-aiindustry-signalpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/low-code-ai-automations-how-gartner-sees-the-future-of-development/https://www.neoteric.no/blog/low-code-ai-automations-how-gartner-sees-the-future-of-development/By 2026, 70-75% of enterprise apps will use low-code platforms. AI-powered low-code will drive 80% of app development by 2029, mainly from non-IT users. Yet 40% of agentic AI projects face cancellation due to governance gaps. OutSystems, Mendix, and Appian lead the market. Success requires training, governance, and embedded security.Fri, 05 Jun 2026 09:00:00 GMTlow-codeaiautomationpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/nvidia-rtx-spark-local-ai/https://www.neoteric.no/blog/nvidia-rtx-spark-local-ai/NVIDIA's first consumer SoC brings 128 GB of unified memory and native CUDA to Windows laptops — here's what it means for running LLMs locally, and the catches.Wed, 03 Jun 2026 09:09:00 GMTnvidialocal-airtx-sparkpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/rapid-mlx-deepseek-v4-flash-158b/https://www.neoteric.no/blog/rapid-mlx-deepseek-v4-flash-158b/Rapid-MLX's new engine optimizations push DeepSeek V4 Flash 158B-A13B to 56 tok/s on a Mac. Here are the exact numbers and the tradeoffs.Tue, 02 Jun 2026 09:00:00 GMTlocal-aimlxbenchmarksmoehomelabpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/mqtt-azure-service-bus-event-grid-event-hub-amazon-sqs-messaging-system-comparison/https://www.neoteric.no/blog/mqtt-azure-service-bus-event-grid-event-hub-amazon-sqs-messaging-system-comparison/Message queues, brokers, and event streams — and how to pick the right one.Tue, 02 Jun 2026 00:07:00 GMTpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/nvidia-vera-cpu-performance-floor/https://www.neoteric.no/blog/nvidia-vera-cpu-performance-floor/NVIDIA's Vera CPU benchmarks show it competes directly with EPYC and Xeon, but software maturity and platform lock-in dictate adoption.Sat, 30 May 2026 09:00:00 GMThardwarecpunvidiaaiindustry-signalpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/on-device-agents-6gb-moe/https://www.neoteric.no/blog/on-device-agents-6gb-moe/Liquid AI's LFM2.5-8B-A1B hits under 6GB RAM with 1.5B active parameters, native tool calling, and verified throughput across edge hardware.Sat, 30 May 2026 09:00:00 GMTlocal-aion-device-aiagentic-workflowsliquid-aimlxpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/12gb-mtp-throughput-jumps-23-percent/https://www.neoteric.no/blog/12gb-mtp-throughput-jumps-23-percent/A community llama.cpp fork squeezes 110 tok/s out of Qwen3.6-35B-A3B MTP on a 12GB card. Here are the exact flags and the VRAM trick to make it fit.Fri, 29 May 2026 09:00:00 GMTlocal-aillama-cppbenchmarksmtphomelabpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/llama-cpp-mtp-speculative-decoding/https://www.neoteric.no/blog/llama-cpp-mtp-speculative-decoding/Multi-token prediction is merged into llama.cpp, delivering 1.4–2.2x throughput on Qwen3.6 with zero accuracy loss.Tue, 26 May 2026 09:00:00 GMTlocal-aillama-cppbenchmarkshomelabspeculative-decodingpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/workstation-ram-vram-ai-q3/https://www.neoteric.no/blog/workstation-ram-vram-ai-q3/AMD's next-gen Ryzen AI Max chips bring desktop AI consolidation, massive unified memory, and high NPU TOPS to compact workstations starting this fall.Tue, 26 May 2026 09:00:00 GMTcpumemorychipsetworkstationaipost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/remote-mcp-timeouts-fix/https://www.neoteric.no/blog/remote-mcp-timeouts-fix/Claude Code v2.1.149 breaks the hard 60s cap on remote tool calls and adds session listing for scripting.Mon, 25 May 2026 09:00:00 GMTpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/llamacpp-mtp-qwen3-throughput-gain/https://www.neoteric.no/blog/llamacpp-mtp-qwen3-throughput-gain/llama.cpp merged Multi-Token Prediction for Qwen3. Community benchmarks show 38→47 tok/s on RTX 3090 and 63→84 tok/s on RTX 5090 — no new hardware needed.Sun, 24 May 2026 09:00:00 GMTlocal-aibenchmarksllama-cppagentic-workflowshomelabpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/amd-gpu-price-hike-dram-shortage-2026/https://www.neoteric.no/blog/amd-gpu-price-hike-dram-shortage-2026/AMD has notified its supply chain of a ~10% GPU price increase driven by a DRAM shortage. DRAM contract prices rose 90–95% QoQ in Q1 2026. Here's what to buy, hold, or skip.Sat, 23 May 2026 09:00:00 GMThardwaregpumemorymarket-priceshomelabpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/llama-cpp-mtp-spec-draft-p-min-long-context/https://www.neoteric.no/blog/llama-cpp-mtp-spec-draft-p-min-long-context/One llama.cpp flag—--spec-draft-p-min 0.75—turns MTP from a dud into a decode speed that holds flat across output lengths where DFlash falls apart.Sat, 23 May 2026 09:00:00 GMTlocal-aibenchmarkshomelabllama-cppspeculative-decodingpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/claude-code-s-simplify-stopped-fixing-code-yesterday/https://www.neoteric.no/blog/claude-code-s-simplify-stopped-fixing-code-yesterday/Claude Code 2.1.147 renamed /simplify to /code-review and dropped the auto-fix behavior. The new command reports bugs at chosen effort levels but no longer changes code.Fri, 22 May 2026 09:00:00 GMTaiclaudeagentic-workflowstoolingpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/dram-nand-price-crunch-2026/https://www.neoteric.no/blog/dram-nand-price-crunch-2026/DRAM and NAND prices have surged 90–130% since late 2025. A Samsung factory walkout and HBM reallocation mean the crunch won't clear before late 2027 at earliest.Fri, 22 May 2026 09:00:00 GMThardwarememorymarket-priceshomelabindustry-signalpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/llamacpp-spec-draft-p-min-mtp-qwen3/https://www.neoteric.no/blog/llamacpp-spec-draft-p-min-mtp-qwen3/The --spec-draft-p-min filter in llama.cpp PR #22397 rescues MTP for Qwen3.6-27B: 48.9 tok/s vs 29 tok/s at 2000 tokens on a 24GB card.Fri, 22 May 2026 09:00:00 GMTlocal-aillama-cppbenchmarksqwenhomelabpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/your-private-mcp-server-is-now-claude-reachable/https://www.neoteric.no/blog/your-private-mcp-server-is-now-claude-reachable/Anthropic shipped MCP tunnels on May 19. Claude agents can call internal databases, ticketing systems, and on-prem APIs through one outbound connection — no inbound firewall rules required.Thu, 21 May 2026 09:00:00 GMTaiclaudemcpagentic-workflowspost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/your-vllm-thinking-budget-was-doing-nothing-with-mtp-on/https://www.neoteric.no/blog/your-vllm-thinking-budget-was-doing-nothing-with-mtp-on/vLLM 0.21.0 shipped Friday with a quiet fix: thinking_token_budget was being silently ignored when MTP speculative decoding was enabled. If you serve reasoning models with spec decode, you have been paying for it.Mon, 18 May 2026 09:00:00 GMTailocal-aivllmspeculative-decodingbenchmarkspost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/claude-code-v2-1-100-burns-20k-phantom-tokens-per-request/https://www.neoteric.no/blog/claude-code-v2-1-100-burns-20k-phantom-tokens-per-request/A server-side bug in Claude Code v2.1.100+ inflates every request by roughly 20K cache_creation tokens — about 40% overhead. Pin v2.1.98 until fixed.Sun, 17 May 2026 09:00:00 GMTaiclaudeagentic-workflowsbenchmarksindustry-signalpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/llama-cpp-mtp-flag-rename/https://www.neoteric.no/blog/llama-cpp-mtp-flag-rename/llama.cpp renamed the MTP flag on May 13. The old --spec-type mtp is silently ignored. If your tok/s dropped from 140 to 70 you are likely running without speculative decoding.Sat, 16 May 2026 09:00:00 GMTaillama-cppqwenlocal-aispeculative-decodingpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/mcp-server-roundup-may-2026/https://www.neoteric.no/blog/mcp-server-roundup-may-2026/Eighteen months after Anthropic released MCP, the ecosystem is wide enough that picking the wrong servers slows your agent down. Here is the practical short list — what to install, what to skip, and the trap most people fall into.Sat, 16 May 2026 09:00:00 GMTaimcpclaudetoolingagentic-workflowspost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/speculative-decoding-why-local-ai-got-fast/https://www.neoteric.no/blog/speculative-decoding-why-local-ai-got-fast/The same Qwen3.6-27B that ran at 70 tokens/sec on a 4090 in January was running at 140 tokens/sec by April. Nothing changed about the model. Speculative decoding moved from research curiosity to default. Here is what it actually does.Sat, 16 May 2026 09:00:00 GMTailocal-aillama-cppperformancespeculative-decodingpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/third-party-claude-agents-lose-the-subscription-subsidy-june-15/https://www.neoteric.no/blog/third-party-claude-agents-lose-the-subscription-subsidy-june-15/Anthropic is splitting Claude billing on June 15 — Agent SDK and ACP usage moves to a capped credit pool ($20/$100/$200) at full API rates.Sat, 16 May 2026 09:00:00 GMTaiclaudeagentic-workflowsindustry-signalpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/local-ai-inflection-point-may-2026/https://www.neoteric.no/blog/local-ai-inflection-point-may-2026/Three model releases in three weeks moved local AI from 'good enough for hobbies' to 'good enough for production'. Here's what changed and why it matters.Fri, 15 May 2026 09:00:00 GMTailocal-aiqwengemmaself-hostedpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/running-qwen-3-6-27b-locally/https://www.neoteric.no/blog/running-qwen-3-6-27b-locally/A practical guide to running Qwen3.6-27B on consumer hardware in 2026 — memory requirements per quant level, recommended runners, and the MTP trick that doubles your tokens per second.Mon, 11 May 2026 09:00:00 GMTaiqwenlocal-aillama-cpphomelabpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/qwen-3-6-27b-vs-claude-opus-4-7-benchmarks/https://www.neoteric.no/blog/qwen-3-6-27b-vs-claude-opus-4-7-benchmarks/Qwen3.6-27B running locally now scores within 10 points of frontier closed models on SWE-bench Verified. The benchmark table, lined up side by side.Fri, 08 May 2026 09:00:00 GMTaiqwenclaudelocal-aibenchmarkspost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/claude-opus-4-7-coding-leap/https://www.neoteric.no/blog/claude-opus-4-7-coding-leap/Anthropic shipped Opus 4.7 on April 16, 2026, with a seven-point SWE-bench jump, the 1M context window now generally available with no premium, and a new task budget primitive for agent loops.Fri, 17 Apr 2026 09:00:00 GMTaiclaudeanthropiccodingagentspost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/google-gemma-4-open-models/https://www.neoteric.no/blog/google-gemma-4-open-models/Google released Gemma 4 on April 2, 2026 — four variants from 2B to 31B, with 256K context, native vision and audio, and Apache 2.0 licensing. Here's what it's for, where it fits, and how to run it.Sun, 05 Apr 2026 09:00:00 GMTaigooglegemmalocal-aimultimodalpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/optimizely-cms-13-whats-new/https://www.neoteric.no/blog/optimizely-cms-13-whats-new/Optimizely CMS 13 went GA on April 1, 2026. Visual Builder is now the default editor, Content Manager replaces tree-first navigation, Optimizely Graph and Opti ID are mandatory, and the platform jumps to .NET 10. Here's what actually changed, where it's worth caring, and what the upgrade is going to cost you.Wed, 01 Apr 2026 09:00:00 GMToptimizelycmsdotnetheadlesspost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/claude-opus-4-6-million-token-context-and-agent-teams/https://www.neoteric.no/blog/claude-opus-4-6-million-token-context-and-agent-teams/Anthropic released Opus 4.6 on February 5, 2026, with a 1M token context beta, agent teams, adaptive thinking, and developer effort controls — all at the same price as 4.5.Fri, 06 Feb 2026 09:00:00 GMTaiclaudeanthropicagentspost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/azure-devops-workflow-vscode-extension/https://www.neoteric.no/blog/azure-devops-workflow-vscode-extension/A VS Code extension that brings Azure DevOps sprint boards, work item management, and AI-powered assistance directly into your editor.Tue, 25 Nov 2025 09:00:00 GMTvscodeazure-devopsproductivityopen-sourcepost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/claude-opus-4-5-anthropics-new-flagship/https://www.neoteric.no/blog/claude-opus-4-5-anthropics-new-flagship/Anthropic's latest model achieves state-of-the-art results in agentic coding and brings meaningful improvements across reasoning, mathematics, and everyday tasks.Tue, 25 Nov 2025 09:00:00 GMTaiclaudeanthropiccodingpost@neoteric.no (Hans Christian Thjømøe)https://www.neoteric.no/blog/google-gemini-3-pro-multimodal-reasoning/https://www.neoteric.no/blog/google-gemini-3-pro-multimodal-reasoning/Google's Gemini 3 Pro brings generative interfaces, 1M token context, and state-of-the-art multimodal reasoning to developers and consumers alike.Tue, 25 Nov 2025 09:00:00 GMTaigooglegeminimultimodalpost@neoteric.no (Hans Christian Thjømøe)