Bird raises $450M: your AI agent gets its own phone number

Bird.com closed $450 million in debt financing and shipped a platform that gives an AI agent its own phone plan. The company announced the raise on September 23 alongside a rebuilt Agentic Harness that lets agents send SMS, place calls, manage email, and provision an eSIM on Bird's own network. The financing is the headline. The eSIM is the story: AI agent communications is no longer a thing you glue together with Twilio credentials and hope.

What actually happened

The $450 million breaks into a $400 million term loan and a $50 million revolving credit facility, structured as a dividend recapitalization — liquidity for existing shareholders, including current and former employees holding equity. J.P. Morgan is administrative agent, with Capital One and Citi as co-arrangers, joined by SVB, MUFG, Flagstar and Huntington. Bloomberg confirmed the J.P. Morgan-backed structure.

This is debt, not equity, and that matters. Banks lent against cash flow: Bird reported $165 million EBITDA in 2025. Headcount went from over 1,000 at peak to 120 today — the company automated itself down and then borrowed against the result. CEO Robert Vis: "Half the apps on your phone rely on our infrastructure, and yet we're running this business at a fraction of the size we used to be."

The product side gives agents direct channel access across ChatGPT, Claude and Cursor, with country-specific delivery mechanics and inbox routing handled by Bird rather than your code.

Why agent communications matter for your business

Every agent you have built stops at the same wall. It can draft the follow-up text, but a human pastes it. It can decide a customer needs a call, but a human dials. The last ten feet are manual because giving software a phone number used to mean a compliance review nobody wanted to run.

That wall is coming down, and the risk moves with it. An agent with an eSIM and send authority is an agent that can text 400 customers at 2 a.m. because a loop condition was wrong.

So build the fence before the feature:

Give the agent a dedicated number, never your main line. Separate identity means separate logs, separate rate limits, and a killswitch that does not take your business phone down with it.

Rate-limit at the infrastructure layer, not in the prompt. A hard cap of N messages per hour enforced outside the model is the only limit that survives a bad instruction. Prompt-level restraint is a suggestion.

Log every outbound message with the decision that caused it. When a customer asks why they got a text, "the agent decided to" is not an answer. Store the trigger, the input and the timestamp.

Require approval above a threshold. Anything to more than a handful of recipients, anything with a dollar amount, anything at an odd hour — queue it for a human.

Agent communications is getting bank-grade funding because it is about to be ordinary. Decide your guardrails while it is still a project and not an incident.

Sources: PR Newswire, Bloomberg.