Model Vendors Encroach On Agent Platforms
Tasklet
The real threat is that model companies can turn agent features into a default capability, which compresses standalone agent platforms toward the messy integration work that horizontals usually avoid at first. Anthropic, OpenAI, and Google already ship tool use or function calling at the model layer, so the remaining hard part is not reasoning, it is connecting thousands of apps, handling auth, and making actions run reliably inside real business systems.
-
Zapier describes its durable edge as long tail interoperability, not raw model access. It supports 8,000 apps, lets admins control which tools and endpoints agents can touch, and inserts deterministic workflow steps around LLMs so automations fail less often in production.
-
Browser automation shows why the connection layer matters. When a target system has no API, an agent has to sign in, click buttons, fill forms, and recover from popups like a human would. That is slow and expensive, but often it is the only way to complete real work.
-
Tasklet is still early enough that this matters disproportionately. Revenue reached about $5M by March 31, 2026, from roughly $385K at the end of 2025, which means execution on integrations and reliability can still create real near term separation before model vendors fully move down stack.
The next phase is a race between distribution and infrastructure depth. Model providers will keep pushing from chat into action, but the winners in workflow automation will be the ones that make agents dependable across thousands of systems, including old portals and edge case enterprise software that still break naive model driven automation.