Most AI tools in IoT so far have focused on the data: asking questions of sensor readings, spotting anomalies, producing dashboards. Soracom is pushing further. Its new Soracom Agent is designed to help with the whole IoT project, from working out what a customer needs to building and running the system.
The IoT connectivity company introduced Soracom Agent globally in July 2026, after launching it in Japan as a technology preview earlier that month. The agent can work with Soracom's platform through its APIs, command-line tools and a Model Context Protocol (MCP) server, which means it can take actions, not just offer advice.
Soracom says each customer's agent runs in an isolated, secure container and keeps a project memory that builds up as work progresses. Customers retain ownership of their data and of what the agent learns about their project. Pricing has not been announced in the English-language release.
From Answering Questions To Doing The Work
Soracom Agent builds on two earlier AI services. Soracom Query lets users ask questions about their IoT data in plain language instead of writing SQL. Soracom Flux is a low-code tool for building automated workflows that react to device data. The agent sits above both, able to use them as part of larger tasks.
"Soracom began by making connectivity accessible to IoT builders," said Kenta Yasukawa, Soracom's co-founder and chief technology officer, describing a path from connectivity through Query and Flux to the new agent.
What It Looks Like In Practice
Japanese coverage of the launch gives a clearer picture of the intended use. According to gihyo.jp's report on the launch, a user could instruct the agent in natural language, for example asking it to understand sensor alerts when anomalies appear and suggest responses. The agent then assembles the workflow itself, combining devices such as Soracom's cameras with data collection, image analysis and notifications.
The same report says users can give instructions through a console or by voice, and that the agent can operate devices, APIs, command-line tools and MCP servers to complete tasks. Soracom has also linked it with Wisora, its AI chatbot service, for tasks such as checking stock levels from office camera images. Future plans include notifications through Slack, email and voice, support for third-party APIs and compatibility with more devices.
Why IoT Projects Need This Kind Of Help
IoT projects are notoriously hard to get right. They cut across hardware, connectivity, cloud services, data handling and business processes, and many organisations lack people with expertise in all of them. Japanese media have described IoT development as a kind of mixed martial art for that reason.
That complexity is a big reason pilot projects stall before reaching production. Soracom's bet is that an agent with deep knowledge of its platform can close some of that gap, helping non-specialists design and deploy systems that would otherwise need a team of engineers. We looked at how industrial connectivity pieces fit together in our explainer on industrial IoT connectivity.
A Pattern Across The Industry
Soracom is not alone in adding agents that can act on IoT platforms. Fleet and operations platform Samsara has taken a similar step by exposing its data and tools through MCP, as we examined in our look at what Samsara is doing with MCP. The shift from AI that describes the physical world to AI that acts on it is one we explored in our piece on how physical AI turns sensing into action.
The Questions Agents Raise
An AI agent that can change device settings, build workflows and call APIs brings clear benefits, but also new risks. If it misunderstands an instruction, it could misconfigure devices, trigger false alerts or change settings across a fleet. And any agent that reads data from devices or documents can potentially be misled by manipulated inputs.
Soracom's design choices address part of this. Isolated containers per customer limit the chance of data leaking between accounts, and keeping project memory under customer control addresses concerns about intellectual property. What the public materials do not yet detail is how permissions and approvals work: which actions the agent can take on its own, and which require a person to confirm.
The security community has started to catalogue these risks. The OWASP GenAI Security Project's Top 10 for Agentic Applications lists problems such as goal hijacking, tool misuse and privilege abuse, all relevant to agents that can act on connected devices.
What Customers Should Ask
Organisations considering Soracom Agent, or similar tools from other vendors, will want answers to several questions:
· Which actions the agent can take automatically, and which need human approval.
· How its access is scoped, for example by project, device group or environment.
· Whether its actions are logged in a way that supports auditing.
· How project memory is stored, and whether it can be exported or deleted.
· How pricing works as usage grows.
Part Of A Broader Soracom Push
The agent arrives as Soracom expands its platform in other directions too. Earlier in 2026, the company opened pre-orders for SGP.32 eSIM orchestration, supporting the new GSMA standard that lets companies switch operator profiles remotely. We covered that standard's wider impact in our piece on the fight over the IoT SIM.
Taken together, the moves show Soracom trying to become more than a connectivity provider: a platform that handles connectivity, data, automation and now project delivery.
The Bottom Line
Soracom Agent extends AI in IoT beyond analytics into designing, building and operating projects. For organisations that struggle to staff IoT expertise, that could shorten the path from idea to working system.
It is still early. The service launched in Japan as a technology preview, and details on pricing and approval controls remain limited. How useful it proves will depend on how reliably it turns plain-language requests into working, secure IoT systems, and how much control customers keep over what it is allowed to do.
