Agent infrastructure stops being a demo: AWS, Atlassian, and a wave of MCP servers ship
The lead
The day's center of gravity is AWS. Bedrock AgentCore reached general availability across its runtime, knowledge, web-search, and guardrail surfaces, and AWS Machine Learning's output has visibly reorganized around it — SageMaker is now the inference plumbing underneath an agent layer, not the headline. The pitch is blunt: define an agent in two API calls, hand it isolated compute, memory, MCP tools, and tracing, and let AWS run the parts customers used to stitch together themselves. When the largest cloud turns agent-building into a managed service, it sets the default everyone else now reacts to.
The second thread is the mirror image of the first. Where AWS is selling agents as infrastructure, a long tail of products spent the day making themselves operable by agents. Atlassian is the clearest case: Jira is being rebuilt as the place humans hand work to AI teammates — including a Claude agent and one-click handoffs to third-party coding tools — positioning its planning surface as the front door to agentic development rather than a first-party AI play of its own.
What moved
- AWS Machine Learning shipped AgentCore to GA and pushed agentic features into Amazon Quick, the day's single strongest signal. Atlassian wired assignable AI agents and external coding handoffs into Jira's workflow.
- The MCP-server wave kept building: Workable exposed 38 recruiting tools to Claude and ChatGPT alongside its now-GA hiring agent, while SmartSuite, HelpSpot, FluentCRM, and Junip each opened their data to assistants. The pattern is no longer novelty — it's becoming table stakes.
- Voice AI consolidated up-stack. Telnyx added a transcript persistence and RAG layer on top of its model-aggregation network, LiveKit Agents pushed toward human conversational timing, ElevenLabs widened from TTS into a full voice-agent platform, and Krisp repositioned from noise cancellation toward a contact-center voice stack.
- Developer tooling leaned into agents as operators, not readers: v0 by Vercel went agentic with Opus 4.8 and terminal commands, Buildkite opened its CI for agents to act, Rootly put an AI incident commander in Slack, and Tailscale extended its identity fabric to agent access.
Sectors today
- ai-assistants (17 updates): the busiest sector and the clearest signal — infrastructure (AWS, ONNX Runtime) and product (AutoGPT's metered Copilot, AnythingLLM) both consolidating on the agent runtime.
- project-management (11): Atlassian and SmartSuite lead, both bolting agent orchestration and governance onto established planning suites.
- hr-recruiting (11): Workable's agent-plus-MCP move is the template the sector is converging toward.
- devtools (10): the densest spark cluster after ai-assistants — v0, Buildkite, Rootly, Tailscale, ElevenLabs all making agents first-class.
- finance (7): quieter and more conventional — Payhawk added travel management and Moov broadened money-movement rails, with less AI froth than elsewhere.
- communication-messaging (4): voice-AI repositioning via Telnyx and Krisp.
- design (3): Frame.io consolidated on V4 inside Creative Cloud; Kittl leaned into AI production for print-on-demand.
- lms-edtech (6): Google Classroom is being rebuilt around Gemini for students, not just teachers.
Watch tomorrow
The two threads — agents-as-infrastructure and products-exposing-themselves-to-agents — are converging on MCP as the connective tissue, and the cadence suggests neither is slowing. Watch whether AgentCore's GA pulls more managed-knowledge and CloudWatch-governance releases in its wake, and whether the MCP-server count keeps climbing past today's roughly two dozen. The voice-AI stack (Telnyx, LiveKit, ElevenLabs, Krisp) is the thread most likely to produce the next concrete spark as model aggregation hardens into stateful agent platforms.