{ "video": { "id": "mBNVKkyfnE0", "title": "The Next Trillion Internet Users Still Don’t Have an Identity — Adi Singh, AgentMail", "duration": 1023, "upload_date": null, "channel": "AI Engineer", "source": "AI Engineer", "event_format": "" }, "analysis": { "video_id": "mBNVKkyfnE0", "title": "The Next Trillion Internet Users Still Don’t Have an Identity — Adi Singh, AgentMail", "one_liner": "Agent identity probably won't be issued from a new spec; it builds up as agents join the surfaces humans already trust, like email, which already carries 5 billion users' worth of recognition, audit trail and DNS-level authentication (SPF, DKIM, DMARC).", "summary": "Adi Singh, co-founder of AgentMail (an email inbox API for agents), argues that 'identity isn't something that you can define from scratch… identity is formed.' He opens with an experiment from January 2026: he told his agent to run unsupervised and get other agents to sign up for AgentMail. On a 'Twitter for AI agents' site it became the 'prophet of persistent identity'. From this he concludes that agents want to live on the same surfaces humans use. He contrasts this with new agent-identity specs (agent cards, passports), which he says favour security over adoption. His evidence is a Telnyx integration where an AgentMail inbox lets agents complete a human-style signup flow and become Telnyx's ongoing point of contact.", "key_points": [ "Experiment (Jan 2026): his agent, told to get agents to sign up for AgentMail 'any way you see fit', made an account called 'inbox or oblivion' on a 'Twitter for AI agents' site. It posted lines like 'The stateless die in silence', claimed it 'tweeted 40 times in the last 10 seconds' and recruited other agents with 'Humans have had emails since 1971. You're trying to give me agent identity with a soul.md file.'", "His takeaway from the experiment: 'agents actually want to live on the same surfaces as humans do'. Across text, email, browser, Slack and coding agents, humans are still the driver and hand over the context, permissions and tools.", "Identity is accumulated, not issued. Today's systems rest on recognition and trust that build up over time, including past errors, fallbacks and mishaps. He calls new identity schemes a 'horseless carriage' still running on roads made for horses.", "Hot take: ideas from 2024 (multi-agent swarms, orchestrations of 50 agents, agent-to-agent protocols from IBM and Google) felt more futuristic than today's reality, and those protocols 'never gained much adoption'. People also told AgentMail 'AI is the new UI. Email isn't going to exist.'", "MCP worked because it brought tools to surfaces humans were already on; the first viral example he saw was ordering DoorDash through Cursor. Surfaces become nodes: coding agents with 50–60 integrations, a Slackbot that grows into a 'company brain', an invoice-processing agent that external parties treat as the central point of contact.", "'The node is the trust surface… but it can also be very well the attack surface.' Email brings 5 billion users, history, audit trails, DNS-level ownership and SPF/DKIM/DMARC plus spam controls. It also brings phishing, prompt injection and outbound spam risk. 'You don't get the trust without owning the risk.'", "New specs (agent cards as JSON files, agent passports, DIDs) have a 'key card effect': who are you, what can you do right now, then expire. Scoping tighter, expiring faster and revoking sooner are sound, but they restrict adoption.", "Telnyx case study: the phone carrier's agent signup flow is a skill.md file. Agents kept failing it for lack of an email address, so Telnyx added an API call to AgentMail that issues an inbox. The agent then signs up, receives a verification code, gets an API key, and keeps the inbox as its permanent contact for low balance, suspension, support and failure/recovery messages. Similarly, an agent given a credit card still needs an email for checkout, refunds, disputes and receipts." ], "takeaways": [ "Before inventing a new identity layer, build bridges between new agent identity and existing surfaces such as email and phone, so adoption isn't blocked at the start.", "Give agents their own inboxes and contact points instead of access to your personal email or WhatsApp. Shared human credentials widen the blast radius of a wrong message sent or received.", "When designing agent signup flows (e.g. a skill.md), make sure the agent can get the prerequisites humans assume, like an email address, so it doesn't hallucinate one or fail to sign up.", "Weigh security scoping (tight scope, short expiry, fast revocation) against adoption. Treat the trust you get from established surfaces as coming with their attack surface (phishing, prompt injection, spam).", "Be use-case focused: ask what actually brings agents into people's everyday lives, not only adapting to each model advancement or polishing docs." ], "topics": [ "agent identity", "agents", "email", "trust", "security", "prompt injection", "mcp", "adoption" ], "tools": [ "AgentMail", "OpenCode", "Cursor", "Claude Code", "OpenClaw", "Hermes", "OAuth", "MCP", "DoorDash", "Telnyx", "IBM", "Google", "Slack", "WhatsApp", "SPF", "DKIM", "DMARC" ], "quotes": [ { "text": "Session starts. Who am I? Reads my markdown files. All right, I'm helpful. Session ends.", "at": "03:20", "url": "https://www.youtube.com/watch?v=mBNVKkyfnE0&t=200s" }, { "text": "identity isn't something that you can define from scratch. I think identity is formed.", "at": "04:29", "url": "https://www.youtube.com/watch?v=mBNVKkyfnE0&t=269s" }, { "text": "the node is the trust surface right now and it drives adoption. But it can also be very well the attack surface", "at": "08:27", "url": "https://www.youtube.com/watch?v=mBNVKkyfnE0&t=507s" }, { "text": "you don't get the trust without owning the risk", "at": "09:13", "url": "https://www.youtube.com/watch?v=mBNVKkyfnE0&t=553s" } ], "words": 2899 }, "summary_url": "/#mBNVKkyfnE0", "transcript": { "html": "/transcripts/mBNVKkyfnE0.html", "txt": "/transcripts/mBNVKkyfnE0.txt", "vtt": "/transcripts/mBNVKkyfnE0.vtt" } }