Function cockpit
Research & Development
Agents run research operations. Scientists own hypotheses and conclusions.
Phase: Phase 2 — Supervised execution
Connect scientific programs to resources, evidence and decision gates: retrieval, scheduling, data integrity, simulation and reproducibility are agent work; hypothesis selection, safety and scientific conclusions remain human.
Trust score
0
Touchless
0%
Human review
0%
Override rate
0%
Capacity released
20-35% research-operating capacity (no presumption that success rates improve)
Accountable human
Chief Scientific Officer
CoS agent: R&D Chief of Staff
How a unit moves: the conductor, the five desks, the agents, and the people
Flow of work
How work actually moves through Research & Development
Work enters on the left. R&D Chief of Staff routes it to one of five desks. A named agent does the work. It leaves as cleared, as evidence posted, or as a hold for a person. Every packet you see below is weighted by the real distribution of 178 live units.
Live units
178
Moving now
90
in flight at a station
Held for a person
32
53 past SLA
Agents on the floor
86
11 sub-functions
Control gates
40
rules a unit cannot pass
Interchanges
7
handoffs out of the function
Flow of work
R&D Chief of Staff and the desks reporting to it
Each packet is a unit of work. Desk weighting and the outcome mix are taken from the 178 units live in this function.
What starts the work
The triggers that put a unit on a line, ranked by live volume.
- 51
Runs transactional
Runs transactional. Starts when a request lands in the queue.
- 44
Runs transactional
Runs transactional. Starts when a threshold is crossed.
- 42
Runs transactional
Runs transactional. Starts when a scheduled window opens.
- 15
Runs transactional
Runs transactional. Starts when an upstream system posts an event.
- 13
Runs analytical
Runs analytical. Starts when the reporting period closes.
Where the work goes next
Shared services that run the transaction, and the functions this one hands to.
The five desks
What R&D Chief of Staff is actually conducting
Each desk is a layer of the operating model, not a team. The conductor decides which desk owns a unit, then which agent on that desk picks it up.
Takes a position
Strategy
Agents that take positions, argue them, and get held to the outcome.
Receives
A question the function has to answer with money or people behind it.
Emits
A written position with assumptions, alternatives priced, and break conditions.
tasks 24h
1,884
in flight
487
held
0
gates
0
- Data Science & Modeling Challenger AgentData Science & ModelingA2
- IP Strategy AgentIP Creation & CaptureA2
- Portfolio Bets AgentInnovation PortfolioA2
- Innovation Portfolio Challenger AgentInnovation PortfolioA2
- Prototype & Pilot Challenger AgentPrototype & PilotA2
- Discovery Research Challenger AgentDiscovery ResearchA2
+ 12 more on this desk
Writes the envelope
Policy
The written rules every other agent is bound by, with version and owner.
Receives
A position that has been accepted, plus every regulator and control change.
Emits
Versioned rules that bind every other desk, with an owner and an effective date.
tasks 24h
198
in flight
217
held
17
gates
40
- Protocol Compliance AgentExperiment DesignA2
- Safety Compliance AgentLab OperationsA2
- Regulatory Policy AgentRegulatory ScienceA2
- Model Governance AgentData Science & ModelingA2
- Design Policy AgentExperiment DesignA2
Watches and warns
Observe
Continuous sensing of the outside world and of the agents themselves.
Receives
Outside signals and the behavior of the agents themselves.
Emits
Alerts, drift flags, and the evidence the conductor uses to reroute work.
tasks 24h
2,287
in flight
0
held
0
gates
0
- Landscape Monitoring AgentTechnology ScoutingA3
- Modeling OrchestratorData Science & ModelingA3
- Collaboration OrchestratorExternal CollaborationA2
- Prototype OrchestratorPrototype & PilotA3
- Innovation Portfolio OrchestratorInnovation PortfolioA2
- Instrument Health AgentLab OperationsA4
+ 13 more on this desk
Does the work
Workflow
Doing the work: lines, stations, and live units moving through them.
Receives
Live units of work entering at the first station of a line.
Emits
Cleared work, posted evidence, and holds when a unit leaves the envelope.
tasks 24h
73,807
in flight
1615
held
6
gates
0
- Explainability AgentData Science & ModelingA3
- Control Design AgentExperiment DesignA3
- Disclosure Drafting AgentIP Creation & CaptureA3
- Benchmark AgentPortfolio AnalyticsA4
- Partner Scouting AgentExternal CollaborationA2
- Validation AgentData Science & ModelingA3
+ 29 more on this desk
Processes at scale
AI GBS
Repeatable transaction processing run once for the whole enterprise.
Receives
Repeatable transactions handed off from every function on the same spine.
Emits
Settled transactions, a cost per unit, and exceptions back to the owning desk.
tasks 24h
14,436
in flight
1307
held
9
gates
0
- Calibration AgentLab OperationsA4
- Protocol Drafting AgentExperiment DesignA3
- Pilot Site AgentPrototype & PilotA3
- Build Planning AgentPrototype & PilotA3
- Success Rate AgentPortfolio AnalyticsA4
- Requirement Mapping AgentRegulatory ScienceA2
+ 3 more on this desk
Human lanes
Where the conductor stops and asks a person
The conductor does not decide everything. Materiality routes a unit to a named human role, and the clock keeps running while it waits.
Function Leader
Critical
Blast radius or value beyond what any agent in this function may commit.
waiting
0
oldest
--
- Nothing waiting in this lane right now.
Process Owner
High
A policy exception inside the function, or a gate the agent is not rated for.
waiting
3
oldest
8h 27m
Approver
Medium
Routine four-eyes on a commitment the agent has already prepared.
waiting
13
oldest
9h 52m
No person required
Low
Inside the envelope. Cleared by the agent and sampled after the fact.
waiting
46
oldest
19h 53m
Walk one unit
Follow the single piece of work with the most at stake
Diagrams argue. A unit proves. Step through the actual route this unit is taking, station by station, and read the evidence written at each action.
Analysis at ratify
Nadia Bianchi · raised by Simon Duarte · $1.1m at stake · touches 375 downstream things if it goes wrong.
The route
Aggregate history
clearedProgram Health Agent handled this station and released the unit onward. Design dwell here is 13h 43m against an SLA of 21h 26m.
Owned by
Program Health Agent
Layer
workflow
Dwell vs SLA
13h 43m / 21h 26m
Evidence trail recorded so far
4 entries. Written when an action is taken, not when a station is passed.
- -318mProgram Health Agent Cleared aggregate historyevidence: Inputs, rule version and output hash written to the evidence log
- -239mSpend Attribution Agent Cleared model driversevidence: Inputs, rule version and output hash written to the evidence log
- -159mCycle Time Agent Cleared draft findingevidence: Inputs, rule version and output hash written to the evidence log
- -80mDirector, R&D Analytics Holding at ratifyevidence: Inputs, rule version and output hash written to the evidence log
Trace a different unit
The walk above always picks the unit in research development with the most at stake right now.
What this view is and is not
Every count, agent, station, gate and unit on this page is read from the live records for Research & Development. The moving packets in the diagram are a visual weighting of that same distribution, not a live message bus: they show where volume goes, not individual transactions in real time. The human lanes are the routing rule, not a staffing model. Chief Scientific Officer remains accountable for this function whatever the conductor clears.