LoginSign Up
← Back to all sparks
Daily Brief · May 10, 2026

Every category rebuilds for AI agents — Gumloop, Speakeasy, Composio claim the runtime layer.

mcp-everywhereagent-runtimevoice-aienterprise-hardeningembedded-biai-first-rebuilds
Y
By Yahya Tür
LinkedIn ↗
Generated 11d agoDrawn from 71 products

The lead

If you read SparkPulse to find out where product surface area is being rebuilt for AI agents, today is the densest day of evidence in months. Gumloop, Composio, Speakeasy, AssemblyAI, Cube, GitBook, Folk, Replicate, Superhuman, Submagic, BugHerd, Scribe, Strapi, and Raycast all shipped MCP-shaped moves — server endpoints, agent skills, tool-callable inboxes, agent runtimes — in the same window. The pattern is not that one category went agentic; it's that every category did.

The single sharpest move is Gumloop's reposition (three sparks): not "agent automation tool" but "enterprise agent platform" — owning MCP, governance, and multi-agent orchestration. Speakeasy echoes it at the infrastructure layer, reframing Gram from MCP packaging into a full enterprise agent runtime with governance baked in. The companies that previously sold "automation" or "SDK tooling" are converging on the same target description.

What moved

  • Agent-runtime claim-staking. Gumloop, Speakeasy, and Composio each made directional moves on the enterprise-grade agent execution layer; Composio paired it with Webhook Triggers V2, an auth migration, and security primitives.
  • Voice and multi-LLM gateways. AssemblyAI shipped a full voice-agent pipeline plus a multi-LLM gateway. Saleshandy added voice and infrastructure on top of cold email. OpenPhone made Sona deployable as a front-line AI agent. Respond.io built escalation paths for its Voice AI Agents.
  • AI-first product rebuilds. Air rebuilt around Canvas. UXPin is racing past per-screen prompting toward whole-flow generation in Forge. Recraft (three sparks) broadened from image generator into a multi-modal creative platform. Userflow is reshaping into an AI-first product adoption layer.
  • Enterprise hardening sprints. Devin's quarter has been one long enterprise pass, headlined by stacked review permissions and network policy; Carta rebuilt equity flows around an agentic ERP; Cohere consolidated around enterprise-specialized models (Rerank, Transcribe, Translate).
  • Embedded BI meets agents. Cube shipped Creator Mode and a Slack agent. Lightdash collapsed the SQL-vs-spreadsheet divide for analysts. Kubit pivoted from query builder to agentic analytics with Lumos. Mode is converging spreadsheets, SQL, Python, and cross-source joins into one analyst surface.

Sectors today

  • Design (24 products / 26 sparks): the most active category — Recraft did the most directional work; Air, UXPin, Kittl, Balsamiq, BugHerd, ProtoPie, and Visme all leaned harder into AI generation.
  • Devtools (22 / 21): Composio and Neon ran the boldest plays; Bunny.net, GitBook, Render, Statsig, and Replicate all pointed their roadmaps at AI-coding-tool consumption.
  • Finance (26 / 20): Shift4 absorbed SkyTab, Carta rebuilt around an agentic ERP, Rho expanded into A/R, Paystack widened across Africa, and Credit Repair Cloud went mobile-first.
  • Marketing (28 / 18): Arcade went generative video, Saleshandy rebuilt as a multi-channel sales platform, Clay matured into a real GTM platform, and Nightwatch + SEOmonitor both reframed for the LLM-search era.
  • Analytics (22 / 18): Cube, Lightdash, Kubit, and Mode pushed analytics surface area toward agents; June, Sprig, Chord, and Maze layered AI through CDP and survey workflows.
  • Development (13): Appwrite leaned into IaC and runtime breadth to challenge Firebase; Devin and Bunny.net continued enterprise hardening; Strapi rebuilt content schemas around AI.
  • Collaboration (13): Skedda extended beyond desk booking into a hybrid office OS; Tella added a Free plan and editor redesign; Document360 shipped MCP and SCIM.
  • CRM (19): Folk wrapped autonomous AI around data hygiene; Oracle NetSuite stitched AI into close-month workflows; Act! pivoted into payments.
  • Project management (11): SmartSuite leaned into ITSM/GRC with Internal Forms and AI Field Agents; Hostfully stepped into guest screening; ClickUp's post-4.0 push centers on Super Agents.
  • Communication-messaging (14): AssemblyAI, Speechmatics, SuprSend, OpenPhone, Respond.io, and Superhuman — voice and tool-callable surfaces dominated.
  • HR-recruiting (6): Spark Hire shipped AI proctoring for video interviews; Culture Amp layered AI synthesis on its survey core; Workyard became a banking layer for field crews.
  • Customer-support (6): Hatz AI and Thread both shipped multi-agent MSP workplaces; Olark rebuilt around v2.
  • Marketing-automation (5): Gumloop carried the day; Submagic stacked auto-edit, native publishing, and a Claude MCP server in the same window.
  • LMS-edtech (8): Thinkific layered AI tutoring and B2B commerce; Tutor LMS 4.0 betas continue; LifterLMS jumped to 10.0; Scribe wired SOPs into the agent era.
  • Ecommerce (7): Subbly rebuilt churn prediction in-house with a measurable lift; Starshipit moved up into warehouse management.
  • Video-conferencing (2): Tella's redesign and Nextcloud Talk 24's call-first beta — small sample, both directional.

Watch tomorrow

The agent-runtime story has hit critical mass — almost every spark today either shipped an agent surface, an MCP server, or an enterprise-hardening pass to support one. Watch the second-order moves: pricing repackages tied to agent consumption (Balsamiq and Mentimeter already started), governance shipped before the agent features it gates (Composio, Devin, Culture Amp), and the first wave of products that retreat because the integration math doesn't pencil. Gumloop, Composio, and Speakeasy are the three to watch — whichever lands an enterprise reference customer first will set the template the rest copy.