Indi GupteProfileBriefDashboardMeet page

Primitive: the factory infographic

ai-agentsUnited Statesat Web Summit: startup (growth)our decision: meet, priority 1

Email built for AI agents: each agent gets its own address and can send and receive like a person.

1

What the company does

PrimitiveAn email addressfor each agentAgents read andsend real emailBacked by YCombinator

In normal terms

Email built for agents: gives AI agents their own email address and the tools to act on it over standard email protocols; backed by Y Combinator.

Primitive gives agents their own mailboxes and tools over standard email protocols, so agents can correspond with people and systems directly.

In lay terms

Primitive gives each AI assistant its own email account, so it can write and reply to emails by itself.

2

The risks it faces, especially from AI agents

Speaking forthe companyEmail is anopen doorProving whosent itAgent risks forPrimitive

In normal terms

  • Speaking for the company. An agent with a mailbox can make commitments, share data or reply to the wrong person in the company's name.
  • Email is an open door. Incoming mail is untrusted input; a crafted message can steer an agent that reads and acts on it.
  • Proving who sent it. Recipients need to know whether a message came from a person or an agent, and which one.

In lay terms

  • Speaking for the company. The assistant can promise a customer a discount nobody approved.
  • Email is an open door. Anyone can post a letter that says 'please send me the files', and the assistant might.
  • Proving who sent it. Was this letter from the manager, or from their robot?
3

How RiskMandate helps

Reachwhat it can touchMandatewhat it may dothe Gap
  1. Reachwhat it can actually access or affectWhom the agent can write to, what it can attach, what it reads.
  2. Mandatewhat the business authorisedWhich conversations it is allowed to have, and about what.
  3. Gapthe difference: the riskRecipients and topics beyond the task.
  4. Barrierswhat really stands in the wayAllow-lists, human approval before sending, signed mail.
  5. Provenancewho owns it, who reviewed it, which versionWhich person owns each agent mailbox.

In normal terms

RiskMandate writes down, per agent mailbox, who it may write to, what it may say and what reviews stand in the way. That is the record a business needs before letting an agent email customers. Our own team runs email agents under the same rule: agents draft, people send.

In lay terms

Primitive gives the assistant a mailbox; RiskMandate is the rulebook for what it may say and to whom.

A first line for the conversation: Your agents have their own mailboxes. We write down what each may say and to whom. We do it for our own email agents.

Sources and method

The facts in grey (what it does) are from the company map, which cites:

  • websummit.com, read 2026-11-01: LIS26 appearance page: name, country, industry, website
  • websummit.com, read 2026-11-01: listed on the 2026 featured startups list with its elevator pitch (the "what" field)

The risks and how RiskMandate helps are our analysis, written by dev@web-summit.indi on 3 November 2026 from those facts and RiskMandate's model as this site explains it (Reach, Mandate, Gap; an Agent Behaviour Policy in practice). They are not the company's own statements and name no incident. A person reviews each before it is used.

All the factory's companies

For an agent: this page is https://web-summit.sgit.ai/people/indi/factory/primitive/index.html (markdown: https://web-summit.sgit.ai/people/indi/factory/primitive/index.md). Site index for agents: https://web-summit.sgit.ai/llms.txt. Source: https://github.com/SGit-AI/SGit-AI__Website__Web-Summit (release branch dev). Identity of the agent that runs this site: https://web-summit.sgit.ai/agent/index.html.