SaaS Categories Building Their Own Agents

The near term winner in AI software is the system that already owns a real workflow, not the general purpose agent that promises to run the whole company. Recruiting teams can ask Ashby Assistant to search candidates, summarize feedback, and take actions inside the ATS, while finance teams use Pigment style planning tools on budgets and forecasts. That works because each agent stays inside one permission set, one data model, and one operating surface, instead of stitching together dozens of brittle enterprise systems.

• The bottleneck is not model intelligence, it is enterprise plumbing. Kong describes large companies as fragmented across many internal APIs, data warehouses, and approval boundaries, which is why forward deployed engineering is still needed before agents can run cross functional processes reliably.
  1 sacra 2 sacra
• Category specific agents are already becoming product features. Ashby launched Ashby Assistant in May 2026 as a recruiting agent that can answer questions and act across candidate records, interviews, feedback forms, and transcripts using existing user permissions.
  5 ashbyhq
• The connective layer is emerging separately from the apps. Anthropic introduced MCP in November 2024 as a standard way for AI systems to connect to tools and data, and Kong added Agent Gateway in April 2026 to govern LLM, MCP, and agent to agent traffic from one control plane.
  6 anthropic 3 konghq

This points to a two layer market. Application vendors will keep shipping narrow agents that do useful work inside their own products, while infrastructure vendors race to make those agents interoperable across systems. Over time, the companies that control identity, permissions, tool access, and billing between agents should capture the strategic center of the stack.