You’re reading an issue of “The AI Economy,” my newsletter exploring the forces shaping the AI era—tracking how AI is rewriting business, work, technology, and culture. Subscribe to get expert insights and curated updates delivered straight to your inbox.
As the AI industry builds agents with greater capabilities and freedom to do what they want, Twilio is choosing to build the layer that decides whether they should. On Thursday, the communications infrastructure platform unveiled Ola, a tool that lets people approve or block actions taken by their AI agents on their behalf. Created by the company’s incubation lab, Twilio Forward, Ola is powered by the Agent-to-Human (A2H) protocol, an open framework defining how agents securely communicate with and request authorization from the humans they work for.
Twilio argues that existing communication channels weren’t built for the agentic era. Services like messaging and email were designed for human-to-human interaction, forcing developers into a trade-off: rigid compliance requirements on one end and inadequate spam protection on the other—neither is suited to autonomous agents operating at scale.
The idea behind Ola is that it’s a single point of contact between humans and the agents they work with. As Twilio’s Vice President of Product and Engineering, Rikki Singh, Senior Engineering Manager for Applied Research, Ryan Ferguson, and Product and Partnership Manager, Ryan Skinner, explained in a blog post, when an agent needs to take action on behalf of their human, a structured request will be sent through Ola describing what it wants to do.
“Ola evaluates the action against your permission preferences and either auto-approves, routes it for your authorization, or blocks it,” the trio said. “Every approval is cryptographically signed, creating a verifiable record of exactly what was authorized and when.”
Twilio claims Ola provides everything needed to oversee AI agents, from communication and permissioning to approvals and activity auditing. Developers get a single channel to interact with all their agents—prompting new tasks, receiving status updates, and reviewing authorization requests—while also ensuring each agent carries a verified identity, scoped permissions, and approvals matched to the level of risk involved. Every intent is logged, so there are no surprises. And if an agent goes rogue, Twilio has built in a killswitch that lets developers shut it down with a single action.
Underpinning all this is A2H, Twilio’s open-source protocol specification introduced in February. The AI industry has moved quickly to standardize how agents interact with tools (Model Context Protocol), communicate with each other (Agent-to-Agent), and handle commerce and payments, but Twilio saw a gap no one had closed: a dedicated standard for agents communicating with the humans overseeing them. A2H is built to fill that space, providing what Twilio describes as “a single, channel-agnostic, auditable surface for agents to communicate with their human principals.”
For Ola, A2H makes the oversight layer trustworthy rather than just functional. Because every agent request flows through a standardized, auditable protocol, developers don’t have to just take Ola’s word that an action was authorized—they now have it cryptographically verified, tied to a defined standard, and not lost in a proprietary black box.
Twilio is launching Ola today as a web app for non-commercial use in the United States. Targeting developers first is a page from Twilio’s own playbook—builders rarely control enterprise budgets, but they shape what companies eventually buy. Early adoption also seeds the A2H protocol itself, giving it a foothold before rivals can establish competing standards. The company said that connecting an agent “takes seconds,” and it uses the same MCP tool-calling patterns AI agents already use.
Of course, Twilio isn’t building Ola in a vacuum. Salesforce, Okta, and ServiceNow are among the companies staking claims in agent oversight, though each approaches the problem from a different angle. For example, Salesforce introduced observability tools as part of Agentforce 360 in November, but it’s designed primarily to manage agents built and deployed within the Salesforce ecosystem. Okta is focused squarely on identity, answering questions like where agents are, what they can access, and what they’re permitted to do. ServiceNow, meanwhile, has been expanding its AI Control Tower into a broader governance layer that spans agent discovery, risk scoring, and now enforcement across AWS, Google Cloud, Microsoft Azure, and enterprise applications.
What appears to separate Ola from the rest is its focus on the human side of the equation. Twilio is building a communication channel between agents and the individual humans they work for, not the control plane for IT and security teams. It’s a layer that didn’t previously exist.
