Saperly maps 10 emerging agent identities

AI agents are starting to get real-world identities. Not personalities — identities. Phone numbers, email inboxes, browser sessions, memory, payment rails, voice, CRM hooks. That’s the shift sitting underneath Saperly’s map of 10 emerging agent identities. The post matters because it shows the agent market breaking into parts you can actually buy, swap, and assemble. ### What is Saperly actually mapping? Saperly calls itself “the first phone carrier for AI agents,” and its SDK pitch is blunt: give any AI agent a phone number with compliance built in. That framing is useful because it treats an agent less like a chatbot and more like an actor that needs credentials and channels to operate in the world. The map extends that logic outward — if an agent can call, it probably also needs email, memory, web access, and a way to pay. (github.com) ### Why does “identity” mean more than login? For human software, identity usually means auth. For agents, identity is broader. It means the handles and permissions that let software persist across tasks and systems. Microsoft’s newer agent identity docs make the same point from the enterprise side — agents need their own governed identities so they can authenticate to tools, get scoped permis(github.com). So “identity” here is really operational personhood for software. (learn.microsoft.com) ### Why is the phone number one category? Because the phone network is still a giant interface to the economy. Sales, support, scheduling, verification, reminders, collections — a lot of business still runs through calls and texts. Saperly is betting agents will need native access to that layer, not just browser automation or chat wid(learn.microsoft.com)hony systems, outbound calling, and mobile or web surfaces. (elevenlabs.io) ### Why are email and browsers separate layers? Because they solve different bottlenecks. AgentMail gives agents real inboxes that can create, send, receive, and search messages over API. That’s not just “email integration” — it’s a dedicated communications identity. Browserbase tackles a different problem: most of the web still isn’t exposed cleanly through APIs, so a(elevenlabs.io)tions on messy interfaces. One gives the agent an address. The other gives it hands. (agentmail.to) ### Why is memory its own product? Because stateless agents are bad employees. Mem0 pitches itself as persistent memory across sessions and agents, and its public research argues that structured memory can cut latency and token cost sharply versus stuffing everything back into context. Basically, memory is turning into infrastructure instead of a prompt hack. That’s a big reason these maps mat(agentmail.to)ide “the model” getting peeled out into standalone services. (mem0.ai) ### What about voice and payments? Voice is becoming a full stack of its own. ElevenLabs now has an agents platform with workflow design, telephony integrations, analytics, and tool calling. Vapi makes a similar pitch for developers who want to build and deploy voice agents fast. Payments are earlier and messier, but the category keeps showing up because once an agent can complete work, the obvious next que(mem0.ai)buy, charge, or settle. (elevenlabs.io) ### So what changed? The important change is not that one startup launched one feature. It’s that the market now has recognizable slots. A broader 2026 agent stack map making the rounds shows the same fragmentation across orchestration, memory, tools, and infrastructure. Saperly’s version is narrower and more concrete — less “AI ecosystem,” more “what identities an age(elevenlabs.io)bout. (aiproductivity.ai) ### What’s the bottom line? Agents are no longer being sold as one magical box. They’re being built like companies — with a phone line, an inbox, a browser, a memory, a voice, and eventually a wallet. Saperly’s map is useful because it makes that decomposition visible. And once the parts are visible, startups can compete on each part separately.