The hybrid organisation: people and agents on one team
Org charts have always assumed one kind of team member: a person.
A possible future we are exploring in the Nebula Lab looks different — hybrid teams, where AI agents carry part of the routine teamwork and people keep the decisions.
This is a look ahead, not a description of today. But the building blocks already exist.
A shared mind for the organisation
Datacore is a local-first, open-source »extended mind«: the organisation’s knowledge, decisions and context live in plain text files, on the organisation’s own side — no cloud dependency, no vendor lock-in. In its authors’ words, it is not a chatbot you talk to but a mind you think with.
Questions, corrections and insights no longer evaporate after the meeting: they become structured knowledge that is searchable and citable — for people and for agents alike.
Datacore — own your intelligence →
Agents as helpers in teamwork
On top of a shared memory, agents become useful teammates. They capture and connect notes, keep task lists and project registers current, and prepare drafts overnight — in the morning the team reviews a briefing of what was done and what is proposed.
- An agent prepares meeting minutes and follow-ups; the owner confirms them.
- An agent keeps registers and documentation aligned with the decisions taken.
- An agent drafts overnight; the morning review stays human.
The rule we apply across our AI portfolio holds here as well: the agent prepares, a person decides.
A hybrid team is not fewer people — it is people with more attention for the work only people can do.
The hybrid organisation — a view ahead
In such an organisation, onboarding means inheriting the shared memory: a new colleague — and their agent — read the same decisions, conventions and context from day one. Institutional knowledge stops living in inboxes and starts living in an auditable, versioned core.
Sovereignty is the design constraint: knowledge in open formats, on the organisation’s infrastructure, readable without any vendor. Agents will come and go; the memory stays.
Lab Brief: An AI that remembers the project →
The same shift, on two levels
What we describe inside Nebula is happening across the industry.
In August 2026 Salesforce recast its platform from a set of applications into reusable business capabilities that agents invoke over the open MCP protocol; a week later, with Claudeforce, its CRM moved directly inside Claude.
The turn is not the interesting part; the reason is. They have offered APIs for 27 years, yet still had to build a capability layer. An API exposes an endpoint — data, operation, the rest is the caller’s problem. A capability also carries the rules: identity, permissions, validations, business logic, audit trail.
While a person sat in front of the system, the screen covered that difference: the greyed-out button, the mandatory field, the warning on an impossible combination. Once the agent is the interface, that layer is gone — what was never written down beneath the screen does not run, or runs wrong.
August 2026, in brief
For us this is not a turn but confirmation of the direction set out in February. The same line, with more at stake: an agent may prepare a quote, but the price comes from the system — price list, campaign, network rules, approved discount.
Business rules do not belong in a prompt: it is not versioned, testable or auditable. If the discount came out of a prompt, »why did this customer get this price?« has no answer — only text somebody once edited. That satisfies neither a regulator nor your own audit.
How we build agents at Nebula
- The model reasons, the business system executes.
- Business rules do not live in a prompt — they are versioned, testable and centrally managed.
- Every action leaves an audit trail: who, which model, which data, what it proposed, who approved.
Strategic Brief: From records to execution →
A closing thought
The question is no longer whether agents will join the team.
Until now we have been asking: who owns the memory they will work from?
Now a second one: who owns the capabilities the agents call — and where do they run?
Memory says what an organisation knows. Capabilities say what it can do.
Hybrid organisations will be built on shared memory and owned capabilities. Own both and you stay the owner of your business — whichever model leads in three years.