Summary
Today's ten signals cluster around three fronts of the enterprise agent buildout. The data and productivity platforms are racing to become the governance chokepoint for AI agents: Snowflake made Cortex Agents and MCP servers generally available as Native Apps with a new Cortex AI Gateway, Databricks made its Unity AI Gateway generally available as a runtime governance layer, and Microsoft brought multi-tenant agent management into the Microsoft 365 admin center in public preview. In parallel, the coding-agent and model-cost race accelerated: xAI shipped Grok Build 1.0 as a Claude Code-compatible CLI, OpenAI brought Codex to Linux and its GPT-5.6 family into AWS's Kiro (cutting completed-task cost ~82% in joint tests), Google launched the low-cost Gemini 3.7 Flash, and Vercel's AI SDK added an ACP harness to make agent runtimes interchangeable. Underneath both, agent infrastructure and the agent economy advanced with AWS Bedrock AgentCore Runtime instances for persistent production agents and Cloudflare Wallets letting agents hold stablecoins and pay for what they use.
Key themes
- The data and productivity platform becomes the agent-governance chokepoint: Snowflake's Cortex AI Gateway and Native App agents, Databricks' Unity AI Gateway GA, and Microsoft's multi-tenant agent management in the M365 admin center all push control, spend limits, and permissions to where the data and tenants already live.
- The coding-agent field is consolidating around shared interfaces: xAI's Grok Build ships drop-in compatible with Claude Code, Vercel's AI SDK adds an Agent Client Protocol harness to swap runtimes without code changes, and OpenAI extends Codex to Linux — the CLI/harness layer is trending toward a de facto standard.
- Model cost is the competitive axis: Google's Gemini 3.7 Flash debuts at aggressive introductory pricing, and OpenAI's GPT-5.6 in AWS Kiro reports a roughly 82% cut in completed-task cost — price-per-capability for agent and coding workloads is the pressure point.
- Agent infrastructure and the agent economy mature: AWS Bedrock AgentCore Runtime instances give production agents persistent compute, GPU support, and 14-day sessions, while Cloudflare Wallets and cloudflare.pay let agents hold stablecoins and pay for resources they consume.
Notable items
- Cloudflare launches Wallets and cloudflare.pay (Agents Week, Aug 4), letting AI agents hold stablecoins and pay for what they use with agent-initiated flows like x402 — flagged high impact for the emerging agent payment economy.
- Microsoft brings multi-tenant AI agent management to the Microsoft 365 admin center in public preview (Aug 18): a consolidated cross-tenant agent inventory with install, permission review, and block controls (risk/activity views require a Microsoft Agent 365 license) — high impact.
- xAI ships Grok Build 1.0 (Aug 7), an agentic CLI built on Grok 4.6 that is fully compatible with Claude Code with zero configuration — signaling the CLI agent interface is becoming a portable standard.
- Google launches Gemini 3.7 Flash (Aug 13) at $0.75/M input and $3.75/M output introductory pricing (rising to $1.50/$7.50 after Dec 31, 2026), a direct lever in the coding/agent model price war.
- OpenAI brings the GPT-5.6 family (Sol, Terra, Luna) to AWS's Kiro (Aug 24), with joint Terminal-Bench 2.1 testing showing roughly 82% lower cost per completed task — a model vendor shipping onto a rival cloud's coding surface.
- AWS adds Runtime instances to Bedrock AgentCore (Aug 6): persistent managed compute for production agents with multi-agent collaboration, GPU support, and sessions up to 14 days.
Source coverage
Source rows used: 10