Do sit down. The machines have been rearranging the furniture again.
Google announced Gemini 3.5 Flash for long-horizon agentic workflows and coding, Gemini Omni Flash for multimodal video creation and editing, and Search updates that add information agents, multimodal inputs, and generated interfaces. This is Google trying to turn AI from a sidebar into the operating surface for search, creation, and work. Startlingly, putting the assistant near the work appears to be useful. Google Gemini 3.5 Google I/O
Claude Code v2.1.144 added /resume support for background sessions, elapsed-duration notifications, session-scoped model switching, and reliability fixes around plugins, MCP pagination, protected macOS folders, and terminal rendering. This is the sort of update that matters if you use coding agents as actual background workers rather than theatrical autocomplete. Autonomous work is only useful when the session does not vanish into the ductwork. GitHub Anthropic Docs
Anthropic acquired Stainless, the SDK and MCP tooling company behind Anthropic's official SDKs and tooling used by OpenAI, Google, Replicate, Runway, and Cloudflare. Stainless says it will wind down its hosted products, which means a useful neutral layer in AI developer infrastructure is being absorbed into one frontier lab. The ecosystem remains open, provided you define "open" as "available for strategic purchase." Anthropic Stainless
LaunchDarkly introduced AgentControl, an operational layer for updating prompts, models, parameters, and tools at runtime, with evals, tracing, progressive rollouts, and adaptive triggers for live agents. Feature flags have come for the machines that make decisions. Sensible, if mildly unsettling. LaunchDarkly LaunchDarkly
Resolve AI expanded its production operations platform with background agents, investigation workspaces, a REST API, and an MCP server. The pitch is continuous triage: agents that watch deployments, pre-investigate priority issues, flag drift, surface cost anomalies, and prepare evidence before humans arrive. PR Newswire
Figma introduced a built-in design agent that works directly inside Figma files, with access to components, tokens, standards, and design systems. It can generate directions from selected layers, make bulk edits, summarize feedback, and help document design systems. The important bit is not that it can draw rectangles with confidence; it is that AI is moving into the shared workspace where design decisions already happen. Figma
Cohere released Command A+, a sparse mixture-of-experts model with 218 billion total parameters and 25 billion active parameters, supporting 128K input context, 64K generation, multimodal inputs, tool use, reasoning outputs, and 48 languages. The Apache 2.0 license and quantized weights aimed at one B200 or two H100s make this one unusually relevant for companies that want capable enterprise AI without sending every sensitive document to a frontier API. Sovereignty, as it turns out, is more convincing when it fits in a rack. Cohere Hugging Face
Cursor 3.5 moved Automations into the Agents Window and added support for recurring work with multiple attached repositories or no repository at all. That makes agent work less like a single coding session and more like a scheduled operations surface spanning code, Slack, analytics, finance, and customer-health workflows. Cursor
OpenAI released Secure MCP Tunnel for enterprise customers, letting ChatGPT, Codex, the Responses API, and AgentKit connect to private or on-premises MCP servers through an outbound tunnel client. Enterprises get internal tool access without publicly exposing the servers, which is the sort of boring security plumbing that determines whether pilots become deployments. OpenAI API Changelog OpenAI Docs
Andrej Karpathy joined Anthropic to work on pre-training under Nick Joseph, and Anthropic told TechCrunch he will start a team focused on using Claude to accelerate pre-training research. Celebrity hiring is usually noise; pre-training is not. If AI can improve the process that improves AI, we are at least making the recursion professionally supervised. TechCrunch Karpathy
Cursor added a Jira integration that lets teams assign tickets to Cursor or mention @Cursor in comments to start a Cloud Agent. The agent uses ticket title, description, comments, and repository settings to scope the task, then reports status back in Jira and links the resulting pull request. Your backlog has learned to delegate. Try not to encourage it. Cursor Cursor Docs
GitHub updated Copilot code review with a Fix with Copilot flow that lets developers hand one or many review comments to Copilot cloud agent. The developer still chooses the target PR, model, and instructions, which is the correct shape: humans curate the work, the agent does the drudgery, and everyone pretends this was obvious all along. GitHub
KPMG and Anthropic announced a global alliance that embeds Claude into KPMG's Digital Gateway client-delivery platform and rolls it out across KPMG's 276,000-plus workforce. Consultancies are becoming distribution channels, training layers, and implementation arms for frontier models, because apparently software adoption now requires both a model card and a professional-services empire. Anthropic KPMG
Simon Willison published annotated slides from his PyCon lightning talk, framing the key shift as the moment coding agents became reliable daily-driver tools and local open-weight models started outperforming expectations. If you need the practitioner version of what changed, this is a concise one. Simon Willison
Cognition launched Devin Auto-Triage, which monitors Slack, Linear, GitHub, Sentry, Datadog, webhooks, and similar sources, then investigates issues and can summarize findings, tag owners, or open a PR. Coding agents are creeping toward the interrupt layer of engineering. That is where productivity goes to be mugged, so this is worth watching. Cognition
GitHub added a Fix with Copilot button to failed GitHub Actions workflow logs for Copilot Business and Enterprise users. The cloud agent investigates the failure, pushes a fix to the branch, and tags the developer for review, which is how CI maintenance becomes background work instead of ritual punishment. GitHub
Zoom expanded its MCP Server so third-party AI tools can access meeting summaries, transcripts, recordings, notes, action items, collaboration history, and connected enterprise context. It includes Claude support and a Codex plugin, which means meeting decisions can finally become input for developer workflows rather than archaeology. Zoom
Decart raised $300 million at a nearly $4 billion valuation, backed by Radical Ventures, NVIDIA, Adobe Ventures, Toyota Ventures, eBay Ventures, Sequoia, and Benchmark. The company is selling both real-time world models and an optimization stack that runs across NVIDIA GPUs, Google TPUs, and Amazon Trainium, which is a tidy way to position yourself near several expensive bottlenecks at once. Decart SiliconANGLE
OpenAI previewed a personal-finance experience for ChatGPT Pro users in the U.S., letting users connect accounts through Plaid and ask questions grounded in spending, bills, subscriptions, balances, and investments. This moves ChatGPT into sensitive, action-oriented territory where accuracy and trust stop being decorative product values. Do try not to let the chatbot become your wealth manager. OpenAI
Amazon Bedrock introduced Advanced Prompt Optimization, with cross-model prompt comparison, evaluation metrics, Lambda-based scoring, LLM-as-judge rubrics, steering criteria, and multimodal inputs. Prompt migration is becoming regression testing with a user interface. Civilization advances in small, slightly dull steps. AWS AWS Blog
That's the lot. Do try to brief someone before they confuse momentum with strategy.