Something is shifting in how AI agents work.
They’re no longer just research tools you prompt for one-off tasks. They’re starting to remember past conversations, send emails from their own address, check your pipeline before taking action, and log their work when they’re done.
They’re starting to behave less like tools and more like teammates.
This shift — from AI agent to AI GTM coworker — is happening right now. And the teams that figure out what infrastructure an AI coworker actually needs will have a massive advantage over those still treating agents as fancy search bars.
So what does an AI agent actually need to make that leap?
What makes an AI agent a GTM coworker?
For an AI agent to work like a real coworker, it needs five things: identity, memory, system of record access, computer access, and the ability to collaborate with a team.
Most AI agents today have one or two of these.
An agent with great research skills but no memory forgets everything between sessions. An agent with memory but no identity can’t send an email or sign up for a tool.
The gap between “useful tool” and “GTM coworker” is about giving them the infrastructure to operate like a co-worker.
Goose: built around all five
Goose was built around this exact thesis.
- Identity — Every Goose agent has its own email address (powered by AgentMail), so it can communicate externally, authenticate with services, and operate as a recognizable entity. We covered this in depth in our post on why every AI agent needs its own email.
- Memory — Persistent context across sessions.
- Computer access — Full access to file system, scripts, APIs, browser, and terminal.
- Team collaboration — Your whole team can delegate work to agents, review outputs, and build on each other’s work.
- System of record access — This is the one that depends on external infrastructure. Goose can connect to your CRM — but how useful that connection is depends entirely on whether the CRM was built for agents to operate in.
That last point is where the choice of CRM becomes critical.
Attio: a CRM where agents are first-class citizens
Today, most people interact with AI agents like — "Go to the CRM, find the deal with [prospect], pull the last three notes." You tell it what to do and where to look.
But that’s not how a coworker operates.
To a coworker you say "Prep me for my call with [prospect]" — and they decide on their own to check deal history, pull the email thread, review the last call transcript, and flag an unresolved objection.
Attio is one of the few CRMs built with this assumption.
Their CTO Alexander Christie recently described their approach in a post on Universal Context — a data layer designed for agentic workloads where "agents running in other platforms such as ChatGPT or Claude Code benefit from exactly the same consistency guarantees as agents running directly inside of Attio."
What becomes possible when your CRM is built for agents
When an agent like Claude Code or Goose already has its own identity, memory, and computer access, Attio’s architecture lets it do things that would be impossible with a legacy CRM:
- Pre-call research — Ask Goose to prep for a meeting. It pulls the contact’s history, deal stage, notes, email threads, and call transcripts — and produces a brief with talking points, and risk flags.
- Post-call logging — After a call, Goose writes a note summarizing the conversation, updates the deal stage, and creates follow-up tasks. The CRM stays current without manual data entry.
- Pipeline awareness — Before running an outreach workflow, Goose checks Attio for existing deals. If a target company is already in your pipeline, it flags the overlap instead of creating duplicate work.
- Contact creation — After a lead gen workflow produces enriched leads, Goose writes them into Attio with notes explaining why they’re ICP-fit and what signal triggered the outreach.
- Stakeholder mapping — Goose looks at who your team has already engaged at a target account, identifies gaps in the buying committee, and suggests who else to reach.
The CRM stops being something you update manually. It becomes something the agent keeps current as a byproduct of doing its job.
Try Goose with Attio’s MCP server and see what happens when your AI agent can actually operate your CRM.