Introducing GraymatterLab Cowork

The shared workspace for human–AI agent teaming.

Two years into the generative-AI era, most healthcare teams still use AI the way they used a search engine: ask a question, get an answer, do the work yourself. That’s a personal upgrade, not an organizational one – and healthcare work isn’t personal. It’s a relay.

A referral lands. Intake qualifies it. A benefits coordinator runs eligibility. A prior-auth specialist escalates to a clinical reviewer. A scheduler matches the patient to a chair, a nurse, and a drug delivery. Billing reconciles the claim months later. Seven steps, six handoffs – and every handoff is where context goes to die. A dropped handoff is a denial, a delay, or a patient who never got care.

The frontier models are extraordinary. That was never the problem. The problem is that the AI lives inside one person’s chat window and the work lives across a team.

Closing that gap is what GraymatterLab Cowork is for. Today we’re making it available: the shared workspace where your people and their AI agents do the daily work together, in one place – chat, shared Spaces, agents from the Gallery, and the Apps your team builds to run its work, all under your HIPAA Business Associate Agreement from the first minute.

Why Cowork matters for your team

The best general-purpose AI platforms are individual-productivity tools with sharing bolted on – a shared folder, a shared instruction set, a document workspace. Useful. But none of them models the thing healthcare actually runs on.

Healthcare work isn’t document-shaped. It’s case-shaped: role-aware, audit-closed shared state, under a healthcare BAA, on day one. The case opens on referral, it stays open across every handoff, and it closes when the episode closes – months later, when billing reconciles the claim. That’s how the work is structured.

We built Cowork for that structure. The ongoing case, not the document.

The Space is the primitive – and a Space is not a folder

A Space is the shared, open-ended workspace inside Cowork. It’s the unit of team work. And the most important word in that sentence is open-ended.

Healthcare work is an ongoing case – an open denial that keeps getting worked, an account you’ll service for years, a recall list that never ends. So a Space is whatever the team needs it to be:

• A patient case – opened on referral, alive until the episode of care closes. Intake, benefits, prior auth, scheduling, and billing all work in it.
• An account – one Space per payer, employer group, or referring provider, alive as long as the relationship is.
• A queue or campaign – today’s intake list, this week’s denials, this month’s recalls. Spaces that move with the work, not with a calendar.
• A team or a person – a manager stands up a Space for their team; an individual stands up a Space to track their own work.

The product doesn’t prescribe the structure. It provides the secure, compliant container, and the team decides what goes in it. Every Space carries what every teammate needs to pick up where the last person left off:

• Shared conversation history – the next person sees what was asked, what was tried, and what was decided. No re-briefing. No “let me forward you the thread.”
• Shared agents and skills – the agents wired into the Space are already there for everyone, with the case-specific context intact.
• Shared documents – the referral, the chart excerpt, the EOB, the policy doc, in one place, scoped to the case.
• Role-based access – a clinician sees labs the biller doesn’t; the biller sees coverage the scheduler doesn’t need. Enforced at the data layer, not the UI.
• A PHI-aware audit trail – every action by every member, human or agent, logged with PHI redaction in the logging layer and retained inside your tenant.

The whole Space lives inside your HIPAA BAA from minute one. No team plan to upgrade to. No “collaboration mode” that makes you strip PHI before you can share. The compliance posture is correct out of the box.

Why a Space is a harness, not a feature

Agents inside a Space are bounded by the Space. They draw from its pinned skills, operate against its shared documents, write into its activity log, and answer to its role model. Take the same agent out of one Space and drop it into another, and its behavior shifts – because its harness shifted. That’s the design, not a side effect.

There are three layers to that harness, each one making the next safe:

• The runtime harness – every agent runs on the GraymatterLab Agent Platform: a defined runtime with a system instruction, a tool list, session state, and an event stream.
• The HIPAA harness – the compliance perimeter. Vertex AI under a signed BAA, 18-category PHI redaction before anything reaches logging, tenant isolation at the database layer, clinical role-based access, and a seven-year audit trail. This is the layer the giants are most ambiguous about – and the one we treat as table stakes.
• The Space harness – the team-coworking layer: membership, pinned agents and skills, shared state, role-scoped views, handoff, and audit closure.

A platform with one harness is a tool. A platform with two is a regulated tool. A platform with all three is a coworker.

What it looks like on a Tuesday

An intake coordinator opens a Space for a new referral. She drops in the referral document and runs the Insurance Discovery agent – already pinned to the Space – and it returns coverage details with a flag: this payer requires prior auth for the ordered therapy. She hands the Space to the benefits team.

The benefits coordinator opens it and doesn’t ask what’s been done. He can see it. The deductible check agent has already run and displayed its output for review. He notes that prior authorization is required, sees that the prior auth agent has pulled the necessary information and populated the payer form, and hands off to the clinical reviewer – who sees the labs intake never needed. The reviewer checks the medical-necessity rationale. The prior-auth specialist reviews the assembled packet and approves it before it goes to the payer. The scheduler agent matches a chair and a nurse. And billing – months later, when the claim posts – opens the same Space, reads the entire chain, and reviews the claim the billing agent has already generated.

Every step, the context survived. Every step, the audit trail closed behind it. The compliance officer reviewing this case six months from now reads one PHI-aware trail of everything every teammate and every agent did.

That’s not a chatbot bolted onto a broken process. That’s the process, with agents in it – and a human on every decision that matters.

Get started

Cowork is in preview. If you run patient access, revenue cycle, prior auth, care coordination, or intake at a non-hospital healthcare organization – specialty pharmacy, infusion, DME, home health, treatment providers, or ambulatory groups – and you want an early look at what your team’s day could look like when their AI coworkers actually share the work with them, reach out. Not a demo of a chatbot. A look at your workflow, with agents in it.