Agent infrastructure ships in waves as GitHub, Rivet, and Weaviate converge on production agents
The lead
Sixty products updated in a single day, but the signal is directional: the scaffolding for production AI agents is shipping in coordinated batches across the developer tool stack. GitHub added sandboxed agent execution alongside new frontier models; Rivet landed OTel tracing, BYOC deployment, and MCP integration simultaneously; Buildkite shipped an MCP server designed to let AI agents operate — not just observe — CI pipelines. The story today isn't which models are available. It's whether agents can run reliably in production, and the answer is becoming yes.
The second thread sits in the memory layer. Weaviate shipped Engram — a purpose-built agent memory system — alongside 4-bit quantization and HFresh disk indexing. Together these moves push Weaviate well past fast vector database into infrastructure for long-running agent systems.
What moved
- GitHub added Claude Opus 5.5 and GPT-6 to Copilot while shipping sandboxed agent execution with OTel observability and shared org-level skills. The model additions expand demand; the sandboxing is the structural change.
- Rivet shipped four sparks this cycle: OTel tracing for cross-span observability, BYOC deployment for enterprise compliance, and MCP integration for AI-native developer experience. It is now production-grade AI agent infrastructure.
- Weaviate released Engram agent memory, HFresh disk-based indexing, and 4-bit quantization in quick succession. The vector database is becoming the memory backend for agent systems.
- Buildkite's MCP server now supports dynamic pipeline inspection and automated failure summarization — primitives that let AI agents debug a broken build without a human in the loop.
- Zluri launched a SoD compliance module and a visual canvas workflow builder, entering IGA territory and competing with Saviynt and SailPoint at the mid-market.
- Ashby shipped Scheduled Agents — recurring autonomous jobs that run without manual triggers and push results to Slack or email. Recruiting automation is now asynchronous.
- Recraft released a new Studio UI, an agentic Chat mode, and 1.3-second image generation in a single push, alongside a Design Agent in open testing that generates brand identity kits from a single prompt.
Sectors today
Devtools (14 products): The densest sector. Buildkite's MCP server and GitHub Copilot's three-vendor model expansion (Claude Opus 5.5, two GPT-6 variants) define the week. Chromatic moving beyond Storybook to Vitest Browser Mode and React Native visual testing is the quieter structural shift.
Development (7 products): Rivet and Weaviate are the sparks. Sanity's MCP server is shipping near-daily toward agents as first-class CMS operators. n8n's AI Assistant can now execute workflow nodes as tools — Agents on by default in 2.41 — collapsing the distance between visual automation builder and agent runtime.
HR & recruiting (6 products): Ashby's Scheduled Agents lead the sector. Tanda's skills tracking now feeds its Roster Agent for AI-driven scheduling decisions — HR structured data becoming a live input to automated shift coverage.
Customer support (6 products): ChurnZero is building a named-agent catalog where every CS workflow has an AI author. DoneDone shipped custom fields and workflow automations, methodically closing the gap with mid-market ticketing platforms.
Analytics (4 products): dbt Core 2.0 is the structural headline — the proprietary/OSS fork formalizes, with agent skills (SKILL.md packages) landing as a plugin surface. Lightdash shipped AI-described custom chart types and a content governance overhaul, continuing its move toward dbt-optional enterprise BI.
CRM (5 products): Primarily infrastructure work. Twenty shipped a native Desktop Recorder and opened messaging channels to third-party apps, moving toward a programmable CRM platform.
Watch tomorrow
Rivet, Buildkite, and Weaviate each shipped dense infrastructure cycles this month — all three are building toward an agent execution + observability + memory stack that doesn't yet exist as a unified product. Whether integrations across these layers emerge is the thread to watch. Separately, Plausible tracking ChatGPT and Claude as dedicated referral channels is an early signal of how quickly AI-sourced traffic is becoming a first-class analytics category for product teams.