GitHub Copilot app sessions can now be opened directly in VS Code, carrying over the conversation, code changes, and worktree so developers can review diffs, edit by hand, and test locally without restarting. The tutorial walks through opening a session, reviewing the agent's changes line by line, continuing the same conversation in the Copilot sidebar, and pushing updates to the session's pull request.
Google released a TPU fork of Meta's TorchTitan that runs PyTorch training on Cloud TPUs through TorchTPU's native torch.device("tpu") backend, with TPU-native kernels and a step-by-step optimization playbook. On TPU v7x (Ironwood), Llama 3 405B holds about 57% MFU from 128 to 1,024 chips, Qwen3 235B reaches 32.42% MFU on one 1,024-chip slice, and FLUX.1 reaches 19.4% MFU on 1,024 chips at 93% weak-scaling efficiency. The fork is at https://github.com/google-pytorch/torchtitan/, and a Transformers modeling backend plus an agent onboarding skill extend the same trainer to other checkpoints.
Why it matters: Google 与 Meta 合作把 TorchTitan 移植到 TPU，并给出从 FLUX.1 到 Qwen3 235B 的逐级调优路径与 MFU 数字，可作为 PyTorch 团队评估 TPU 训练可行性的参考。
Legora now runs each of its legal agent evals inside a separate Modal Sandbox, booting its full application pinned to a specific commit so every run reflects the exact product version. A typical experiment is 100 cases, 4 harness variants, and 5 trials each — 2,000 runs in parallel — with Modal Volumes snapshotting document, database, and permission state. Legora also moved from gVisor to Modal VM Sandboxes, removing its Docker networking and FUSE workarounds, and cut agent cost by 30% via smarter compaction and custom tool search.
Modal detailed how it evolved its filesystem stack to back VM Sandboxes, routing guest requests through virtio-fs and virtiofsd onto its ImageFS lazy-loading filesystem. The first version booted fast but ran slow: recursive chown went from four seconds to over a minute, and pnpm and bun installs were about three times slower. Microbenchmarks showed chown at 9.87 s over virtio-fs versus 2.31 s on gVisor and 1.58 s on ImageFS.
PJ, an engineer-turned-pest-control owner in Ocala, Florida, now consults for small businesses—roofing, pool and automotive contractors among them—setting up OpenClaw and teaching them to run it themselves. His clients want a tool they own outright rather than feeding company data into rented software, and one client, 66-year-old admin Donna, uses her OpenClaw daily.
LiveKit's guide distills lessons from solution architects and forward deployed engineers on taking AI voice agents from demo to production, organized around three questions: whether the agent is smart and natural enough, whether its performance is observable, and whether it scales reliably and cost-efficiently. It recommends measuring the first turn separately, since one company found roughly 3 out of 4 calls ended at the greeting step, and notes clients moving from GPT-4.1 to Gemma 4 31B on LiveKit Inference saved 300–500ms at p50.
The Claude Code team describes its layered code review flow, in which most reading is done by machines and human attention is reserved for risk-based approval. From June to September, merged pull requests to the Claude Code codebase more than doubled while the share that had to be reverted fell by more than 40 percent. The process stacks verification before the PR, required CI, Claude review, security review, risk-tiered human approval and a merge queue with flags and fast rollback, and the team turned past incidents into a daily eval set for the reviewer.
Why it matters: Anthropic's Claude Code team details how it moved most code reading to automated layers while merged PRs doubled and reverts fell over 40 percent, offering a transferable review model for agent-heavy teams.
Modal explains how to run LiveKit Agents with open-weights or custom models on its serverless GPUs, using BitHuman's open-weights Expression Video Avatar voiced by Qwen3-TTS to clone Charles Frye. The setup exposes the avatar container's /launch endpoint via a Modal Server and uses Sticky Sessions to keep WebSocket clients on the same container, with GPU snapshots hiding cold starts.
GitHub trained a fine-tuned ModernBERT classifier to replace its previous LLM-based pipeline for generic secret detection, improving owner-weighted alert precision by 11.5% and cutting median post-push scan execution time by 86%. The two-stage pipeline uses regex candidate generation followed by context-aware classification, which also lowered serving cost by more than 98% versus remote LLM calls. The efficiency gain made it practical to extend generic secret detection to push protection, returning results before changes are pushed.
LiveKit published a troubleshooting guide for Egress, covering common errors such as the 412 "start signal not received" from templates that never call START_RECORDING, 404 participant or track not found within 30 seconds, and 400 upload failures from invalid S3/GCP/Azure credentials. It also addresses layout, webhook, and self-hosted Egress issues, and notes the guidance applies to both the current StartEgress API and the deprecated source-specific APIs.
GitHub's Copilot for Beginners series explains four ways to give Copilot project context: custom instructions in `.github/copilot-instructions.md` plus path-specific files in `.github/instructions`, agent skills in `.github/skills`, custom agents in `.github/agents`, and MCP servers that connect Copilot to external tools such as Playwright. The post walks through each mechanism and how they combine, and points readers to the community Awesome Copilot collection for ready-made examples.
Vapi 工程师 Lat Ware 复盘自建 Pedantic Review Bot 的实践，该机器人现已审查每个 PR，平均合并时间从 130 小时降至 65 小时。
Serve AI rebuilt its phone agent for a $150M pest control customer on LiveKit, collapsing a 224-node Retell flow graph into a single prompt and validating it with Agent Simulations. About 100 real production calls were converted into simulation scenarios, reaching a 98% pass rate over 10–15 iterations before cutover, with median turn latency of 1.3 seconds against a 1.5 second ceiling. The first migration took one week with no rollback, and containment rose 15%, which Serve estimates saves the customer an additional $320,000 a year.
Anthropic published a reference implementation for Claude Managed Agents (beta) that reads Slack channels and GitHub pull requests on a cron schedule, tracks changes since the last run via per-source bookmarks, and posts a brief to Slack. The walkthrough covers vault-held credentials, read-only access and a per-run spending cap, plus rules for reporting a failed source as unreadable rather than a quiet day and confirming a Slack post before updating bookmarks and the ledger. A Claude Code command with the claude-api skill configures the agent from the article's guidance.
Why it matters: Anthropic shares a reference implementation for scheduled Claude Managed Agents that read Slack and GitHub, with concrete rules for bookmarks, failed reads and post confirmation.
Addy Osmani walks through Claude Code cloud sessions, where each task runs on a fresh Anthropic-managed VM with the repository cloned onto its own branch, and cloud sessions are included in Pro, Max, Team and Enterprise plans at no extra compute charge. In a four-session test on a sample Node repository, three sessions started within 16 seconds of each other finished 87 seconds after the first began, and the flaky-test session ran the suite 40 times with zero failures. The guide covers setup scripts, network levels, GitHub connection paths and seven workflows suited to cloud sessions, while noting that local sessions remain the better fit for tasks needing local data, VPN services, GPUs or Zero Data Retention.
Why it matters: A hands-on field guide to Claude Code cloud sessions, showing how per-task VMs enable parallel work and where local sessions still fit better.
GitHub Blog outlines three skills developers should strengthen as AI changes their work: directing AI agents rather than just using them, critically reviewing AI output instead of trusting the first answer, and using time saved by AI to tackle broader problems. It cites GitHub Copilot's built-in Rubber Duck agent, which uses a second model to critique plans, code, and tests before a developer proceeds.
Claude Code mods are small JavaScript or TypeScript files that run inside a session as hooks shipped in plugins, and can observe events, rewrite behavior or draw custom UI in the terminal or desktop app. The guide builds Token Weather, an roughly 80-line mod that reads $.session.usage() after each turn and draws a weather-style context-window band above the prompt, then tours Blast Radius, which holds risky Bash commands, and Replay Theater, which steps through a turn's edits. Mods require Claude Code 2.1.287 or later and are on by default; finished code for all three is in anthropics/claude-code-playground.
Why it matters: A hands-on walkthrough of Claude Code mods that builds a working context-window meter in about 80 lines, showing how hooks, state and custom UI fit together.
Anthropic added two commands to the claude-api skill inside Claude Code: /claude-api build-eval builds an evaluation set in your codebase, and /claude-api hillclimb iterates on an application against that evaluation with a held-out set to catch overfitting.
Why it matters: Anthropic adds build-eval and hillclimb commands to its claude-api skill, showing how Claude Code can design evaluations and iterate against them while guarding against overfitting.
Vapi ran early experiments using TypeSafe AI's early-access "System One" model Jev as a guardrail that scores whether a voice assistant's response violates a developer-defined rule before it is spoken. In one early test, Jev caught at least three out of four violating responses, though Vapi says it is not a 100% compliance guarantee and only an additional check on top of prompt guardrails.
Modal explains how it optimized Kimi K2.6 inference for coding agents, reaching 2.8x per-user and 5.6x cross-user performance per replica and running services at hundreds of billions of tokens per day. The work combines custom DFlash speculative decoding fine-tuned on coding traces, FP8 KV cache and NVFP4 shared-expert quantization, HiCache CPU-tier caching, and KV cache-aware load-based routing. Modal upstreamed its SGLang patches and released the configuration behind Modal Auto Endpoints.
This guide compares LiveKit Agents and Pipecat, the two open source frameworks it says matter most for building production voice and video AI agents, across architecture, adoption, developer experience and production readiness. LiveKit Agents runs on LiveKit's open source WebRTC and SIP infrastructure and wires the pipeline for you, while Pipecat assumes no transport and has developers assemble the pipeline stage by stage. It also covers where TEN Framework, Dograh, Vision Agents and Vocode fit in.
LiveKit explains that the "inference is slower than realtime" warning in its voice agents refers to the local VAD model, not the STT, TTS, or LLM models used through LiveKit Inference. Each 32ms audio window must be processed in under 32ms, and the reported delay shows how far the VAD has fallen behind live audio. Blocked event loops and CPU saturation are the main causes, and the fix is to rule out synchronous code rather than add threads.
Voicemail detection, also called answering machine detection, decides within roughly three to five seconds whether an outbound call reached a live person or a recording, relying on timing and behavior signals rather than voice audio. A 2026 arXiv preprint found timing features alone can match beep- and keyword-based approaches while avoiding their added latency. Vapi lets users choose a detection provider, tune timing, and switch back to conversation when a human answers.
Botika runs its full-stack generative AI operation on Modal, covering a 100-terabyte image data pipeline, in-house foundation models with tens of billions of parameters, and roughly 15 models serving production inference. The pipeline processes that dataset with under a 1% error rate across thousands of concurrent containers, and a single researcher can now run 50 experiments a day instead of one or two.
LiveKit's blog breaks down why voice agents often go silent after short replies like "yes" or "ok," tracing it to turn detection failures. The most common cause is STT providers missing short speech and never emitting a final_transcript, which blocks end-of-utterance detection entirely. The post also covers overly aggressive or permissive endpointing delays and misconfigured noise cancellation that drops short audio as noise.
Vapi published a seven-criteria rubric for evaluating AI voice agents in customer experience, covering resolution, real-time voice quality, action-taking, escalation, model flexibility, lifecycle trust, and platform control. The guide argues containment and deflection overcount success, since contained callers may simply hang up, and cites ContactBabel data showing phone preference for complex interactions rose from 28% in 2018 to 47% in 2024. Vapi positions itself as a model-agnostic platform layer with configurable endpointing, interruption handling, and background denoising.
LiveKit is extending its SFU architecture beyond single peer-to-peer links so robot deployments can run multi-participant sessions, where a Room lets each camera, model, operator, or service join as its own participant. LiveKit describes three architectures: human-in-the-loop data recording with robot, policy, operator, and recorder in one session; VLM-guided skill orchestration where a vision-language model selects skills at roughly 1 Hz while separate policy participants control the robot at tens of hertz; and policy escalation, where a reward model at 30 Hz or higher flags stuck or unsafe states and the robot revokes authority from the specialist policy.
A vendor-neutral guide lays out five criteria for picking a conversational AI platform for enterprise voice agents: developer control, model choice and fallbacks, testing and observability, the full agent lifecycle, and compliance at scale. It cites a Zapier survey of 542 US C-level executives in which 81% worried about depending on a specific AI vendor and only 6% could drop their primary vendor without disruption. The guide argues latency and orchestration matter more than feature lists, noting AssemblyAI's finding that delays past 500–700 ms feel unnatural.
Vapi argues that voice agents are judged on five real-time dimensions: latency, turn-taking, interruption handling, prosody, and a resilient pipeline. It cites research putting human turn gaps near 200ms and notes delays beyond roughly 500–700ms feel unnatural, while naive sequential pipelines can cost about 1.9 seconds versus 600–900ms for optimized ones. Vapi positions its model-agnostic pipeline as tunable and testable via simulated calls.
Vapi published answers to questions from its webinar with Anthropic on voice AI, detailing the live ride-booking demo agent built on Vapi with Anthropic's Claude Haiku and the Layla v2 voice. The agent used five tools — estimate, book, check status, cancel and end call — plus smart endpointing and a conservative interruption threshold to reach sub-second turns. Agents running Claude on Vapi handled nearly 5M calls in the past 30 days across 50+ languages.
AI voice agents cut customer wait times in two ways: removing work from the queue through self-service deflection, intelligent routing, agent assist, and demand prediction, and answering instantly with no hold. SQM Group found customers abandon calls at 2 minutes 36 seconds on average, while ContactBabel put the UK mean abandonment rate at 8.4% in 2024. Vapi says a first voice agent can be stood up from its dashboard in minutes, then measured on ASA, abandonment, and first-call resolution.
Modal published an interactive tool for estimating serverless versus reserved GPU costs, defaulting to current rates for a B200 GPU serverlessly on Modal and via three-year hyperscaler reservation. The model shows that when an application's peak-to-average ratio exceeds the reservation discount rate, serverless GPUs cost less overall, and Modal cites observed discounts of 2-5x against peak-to-average ratios of 5-10x.
Vapi published answers to the seven most-asked questions from its webinar with Spring Venture Group, where Homer Kay described how SVG uses Vapi voice agents to qualify Medicare leads before transferring callers to licensed agents. The answers cover outbound calling compliance, API-first integrations, swapping STT, LLM and TTS components independently, human handoff, and HIPAA-eligible configurations with Zero Data Retention.
Voice AI success starts with use-case selection: tasks ending in a link, code, or number belong in chat, while urgent, hard-to-type, or emotionally loaded problems suit voice. Chat scripts must be rewritten for the ear—split long passages into single-idea turns, restate collected details like dates aloud, and limit spoken options to two or three. One staffing company cut downstream placement errors by 30% by confirming each slot as it was collected, and another team trimmed 40 seconds off average call duration by turning open questions into choices.
Vapi built the 15 voice agents on its homepage — 3 use cases across 5 languages — by generating them from a single ROUGH_DRAFT.md file instead of hand-editing each in the dashboard. The four-phase framework (plan, draft, bootstrap, harden) scaffolds a TypeScript project with OpenAI gpt-4.1 at temperature 0.5, ElevenLabs and Deepgram/Soniox voice stacks, and an idempotent upsert loop that stores assistant IDs in .env.local. The repo is public: clone it, add VAPI_PRIVATE_KEY, and run bun run bootstrap.
Modal contributed optimizations to the FlashAttention-4 kernel to speed up decode-heavy LLM inference workloads, where memory bandwidth is the bottleneck. The changes include porting the "split KV" parallelism technique to FA4 and replacing TMA cp.async.bulk loads with cp.async for irregular memory accesses. An FP8 input PR (merged April 17, 2026) delivers up to 1.16x throughput over the bf16 baseline and enables smaller KV caches.
Modal says the bottleneck in RL post-training of LLMs is now infrastructure, not algorithms, and has released an open-source library drawn from running RL at scale. It details three pain points — glue code, cluster queueing and GPU under-utilization — and reports RDMA weight sync cutting Qwen3 8B update time from 2.62s to 41ms.
Modal published an engineering account of how it cut AI inference replica spin-up from roughly 2,000 seconds to about 50 seconds using four components: a buffer of idle healthy GPUs, a lazy content-addressed filesystem called ImageFS, host-side checkpoint/restore via gVisor's runsc, and CUDA checkpoint/restore built on Nvidia driver support. Across over ten thousand cold starts, mean vLLM boot latency fell from 95,679 ms to 13,797 ms and SGLang from 83,713 ms to 17,486 ms for a ~1 GiB Qwen 3 0.6B model in bf16. The post also notes limits, including single-GPU-only snapshots and host-environment sensitivity that requires multiple snapshots per deployment.
Vapi's team reviewed thousands of voice deployments and identified five characteristics that predict whether a voice agent use case will succeed: high call volume, repetitive and predictable conversations, clear success criteria, and two more. Missing one trait can be compensated for, but missing three means exploring alternative use cases.
OpenAI published a realtime prompting guide for voice agents, covering Realtime 2 reasoning, preambles, tools, and exact entity capture. The guide explains how to structure prompts so realtime voice agents handle reasoning and tool use while capturing entities precisely.