Every Employee Gets an AI Teammate

At Meta, employees no longer just "use AI." Each person works alongside their own small team of named agents, and Mark Zuckerberg is leading the change with his own AI right-hand man.

Two of those agents are already in wide use inside the company.

Second Brain indexes and searches every project document. Its creator calls it an "AI chief of staff".

My Claw reads an employee's files and chat logs, and can act on their behalf. These personal agents also coordinate with each other autonomously.

Meta is backing the shift with incentives. Starting in 2026, "AI-driven impact" is a formal part of performance reviews, and people chief Janelle Gale said the goal is to "recognize people who are helping us get there faster".

Meta's recent acquisitions point the same way. Manus brings a general-purpose agent that can execute real work. Dreamer brings a personal agent OS that learns one user at a time. Moltbook brings a social network where agents talk to other agents.

Put the three together and you get the shape Meta is building toward: every person backed by a team of agents, and those agents talking to each other.

This Is Not Just Meta

Meta is the loudest example, but the pattern is showing up across every major industry.

Finance

Goldman Sachs put Devin, an autonomous AI software engineer, alongside its 12,000 human developers. CIO Marco Argenti called Devin "our new employee" that will start "doing stuff on behalf of our developers." The plan is to scale to thousands.

Morgan Stanley gives every financial advisor a suite of named teammates. Debrief joins client meetings, writes the notes, drafts the follow-up email, and saves to Salesforce. Next Best Action dispatches personalised client outreach. Adoption across wealth management sits at 98%.

Consulting

McKinsey runs around 25,000 personalised AI agents alongside its 40,000 humans. CEO Bob Sternfels wants every employee "enabled" by at least one agent within 18 months.

Deloitte's Zora AI is branded as "digital workers" that interact with existing systems "much like their human counterparts." Role-based agents cover finance, HR, supply chain, procurement, and sales.

PwC deployed 250+ AI agents and 12,000+ custom GPTs internally, orchestrated through its "agent OS" so agents hand work to each other across vendors.

Tech

Microsoft runs more than 500,000 agents internally, with the most used ones focused on research, coding, sales intelligence, customer triage, and HR. Every employee gets Copilot, and anyone can build custom agents in Copilot Studio.

Salesforce reports 86% of employees use Agentforce in Slack. Roughly 25,000 employees use agents "tailored by role and department" for everything from HR to procurement.

Moderna employees built 750 custom GPTs within two months, with 40% of users creating their own. Examples include a "Dose ID" GPT for evaluating vaccine doses. Each employee assembles their own team of role-specific agents.

The common thread is simple. The question leaders are asking has shifted. It is no longer "should we use AI?" It is "how do we empower every individual with their own AI teammates?"

Startups Are Productising the Pattern

It is not just incumbents building this internally. A new wave of startups is packaging the same idea for everyone else.

Viven is the clearest example. Founded by the Eightfold co-founders and funded with a $35M seed, its pitch is "a personalized AI for every employee".

Each employee gets their own specialised model, trained on their email, Slack, and docs. When you are out or unavailable, a coworker can query your digital twin and get the answer you would have given.

Dreamer comes at it from the consumer side. Founded by ex-Stripe CTO David Singleton and ex-Google and Meta exec Hugo Barra, Dreamer pitches itself as a personal agent OS. Every user gets a Sidekick that learns their preferences, builds custom agents on their behalf, and manages the firehose of email, docs, and information coming at them.

The signal from Dreamer is strong enough that Meta just hired the entire team into Meta Superintelligence Labs to scale personalized agents across its products.

Viven and Dreamer are not selling chatbots or copilots. They are selling teammates, one per person, by default. Viven targets the workplace. Dreamer targets the individual. Both bet on the same shape: every person gets their own AI.

Why "Teammate" Is the Right Frame

"Tool" is the wrong word for what these companies are building.

A tool is something you pick up when you need it, use for a task, then put down. A teammate is someone you work with every day. They know the context. They have a role. They can take action on your behalf.

Harvard Business Review made this point directly in a March 2026 piece: think of agents like team members. That means giving them a scoped role, clear access, memory of past work, and real expectations.

Gartner went further. Its Future of Work Trends for 2026 named "digitally replicating employees" as a major shift.

The mental model matters because it determines what you build. If you think "tool," you ship a chat box. If you think "teammate," you ship something very different.

How I Built This at SAP

I built AI Teammate as a side project at SAP, originally just to get through my own work faster. A few colleagues saw it and convinced me to share it. It grew from a personal tool into a shared platform by pull, not by plan.

Seeing Meta, McKinsey, and Morgan Stanley land on the same shape was validating. One platform, a team of agents per employee, no blank canvas.

The first-login experience is the most important design decision. New employees do not land on an empty screen asking them to "create an agent." They land on a roster. Three teammates are already there: a Strategy Partner for thinking through decisions, a Learning Coach for picking up new skills, and a Research Analyst with web search and scraping built in. You start chatting on day one.

From there, everyone can spin up their own teammates and extend any of them along seven axes.

Role. Each teammate has a description written in plain English that defines its role. No YAML, no prompt engineering. The platform handles the rest.

Skills. Lightweight instructions that teach a teammate a specific workflow. A skill can also bundle a small piece of automation, so a non-engineer can hand a teammate a real capability without writing code. The teammate activates the skill only when the task calls for it.

Memory. Every teammate remembers what matters across conversations. Your preferences, your projects, your ongoing context. It updates itself as you work.

Knowledge Bases. Upload your documents and the teammate can read and retrieve from them when answering. Teammates can also update those documents as things change, so knowledge stays alive instead of rotting.

Tools. A catalog of built-in capabilities like web search, scraping, email, and image generation. Each teammate only gets the tools it needs for its role.

Teammate sharing. Any teammate can be shared with coworkers. A good one built by one person becomes leverage for a whole team. Other people can also talk to your teammate directly, like asking a colleague's AI teammate for context when that colleague is unavailable.

Scheduled Tasks. Teammates can run on their own on a schedule. A daily briefing in the morning, a weekly review on Fridays. You do not have to be at your desk for your teammate to be working.

Marketplace. A shared catalog where anyone can publish the tools, skills, and MCP servers they have built. Browse what colleagues have created, import what you need into your own catalog with one click, and your teammates can use it immediately. One person solves a problem once and the whole organisation benefits.

Generative UI. Inspired by recent research from Google, teammates can reply with interactive dashboards, forms, tables, and charts right inside the chat instead of plain text. Ask for a summary of the quarter and you get a dashboard with cards and status indicators. Ask it to help you plan something and you get a form you can fill in.

For those curious, the whole thing runs on SAP stack:

• SAP BTP Cloud Foundry. Hosts the FastAPI backend and React frontend.
• SAP HANA Cloud. Persistence for teammates, memory, conversations, and knowledge bases.
• SAP Generative AI Hub. One gateway in front of Anthropic, OpenAI, and Gemini, so a teammate can pick the right model for the job without wiring up three SDKs.
• XSUAA. Identity and auth, so every teammate and every action is tied to a real SAP user.
• SAP Job Scheduler. Drives the recurring teammate runs for scheduled tasks.
• SAP UI5 Web Components for React (Fiori). The frontend design system, and what makes generative UI feel native: the model emits specs, and the renderer maps them to the same Fiori primitives the rest of SAP uses.

Lessons from the Build

Four things stood out after building this.

Starting with a roster beats starting with a blank canvas. The single biggest factor in adoption was giving people working teammates on day one. When someone logs in and three teammates are already there, ready to chat, they start using it immediately. Every "create your first agent" screen adds friction that kills momentum. Most people do not want to configure an agent. They want to talk to one.

Generative UI changed the questions people ask. Once the teammate can reply with a dashboard or a form, people stop asking for "a summary" and start asking for things they would never have typed into a chat box. The interface shapes the expectation. Plain text makes people think "chatbot." Rich components make people think "tool I can work with."

Scheduled tasks changed the relationship. When a teammate sends you a morning briefing or a weekly review without being asked, it stops feeling like something you go to and starts feeling like something that works alongside you. The shift from pull to push is what makes it feel like a real teammate.

Meeting people where they already are matters. A real coworker emails you, pings you on Teams, or updates a shared document. They do not ask you to open a new application. An AI teammate should work the same way. When the teammate shows up in Outlook or Teams instead of its own window, people treat it like a colleague, not a tool. The UX lesson is simple: do not make people come to the AI. Bring the AI to them.

Where This Is Heading

At Meta, employee agents are already talking to each other in internal group chats. At McKinsey, there are more agents than there will soon be humans. At Microsoft, the stated vision is "human-led, agent-operated."

The next phase is coordination. Agents talking to other agents on your behalf, handing work off, checking each other's output, with the human as the conductor.

What a single employee looks like is changing. The human is the brain in the middle. The AI teammates are the hands and legs. The human sets direction, makes the calls, and holds the context. The teammates do the work in parallel, across systems and tools, at a scale one person never could.