For business leaders
What does it do for us, and will people use it?
The questions a review asks of Promenaut, answered from the platform itself, and everything built for them.
The questions, in the order they come.
A specialist worker with an identity, not a prompt.
- Versioned agent registryAgents are persistent database records with lifecycle state, model configuration and cost tier, not code branches.
- Specialised agent implementationsAgents are distinct implementations rather than configurations of one generic agent.
- Deterministic and hybrid agentsNot every agent is a model call.
The full answer is in How a Digital Workforce Works.
Yes: six kinds, each with a different role in the workforce.
- Kinds of agentBeyond the domain agents that do business work, the workforce has specialised kinds: platform agents that evaluate work as it runs, service agents that other teams call on demand, assistant agents that talk with people, infrastructure agents that keep the platform running on its own outside any workflow, such as routing a request or composing each person's pulse, and rendition agents that turn a finished output into another form, such as a narrated summary.
- Assistant agentsAssistant agents are specialised to talk with people.
- What needs you nowRather than one dashboard for everyone, people open on a digest of their own work: what is waiting on them, what is running, what needs watching and what has finished.
The full answer is in How a Digital Workforce Works.
One of eight verbs, worked out from how the agent is built, strongest evidence first.
- Intent is worked out from how an agent is builtThe platform works out what each agent does from what it declares and writes, rather than from a label: one of eight verbs (ingest, organise, analyse, decide, generate, query, act, monitor), read first from its knowledge-graph writes, then from the kinds and direction of its inputs and outputs, with its declared capabilities only as a tiebreaker.
- Capability registryAgents declare the capabilities they provide into a shared registry, categorised and owned by the module that supplies them, so what the workforce can do is a queryable catalogue rather than tribal knowledge.
- Shared I/O type systemAgent inputs and outputs bind to a shared type registry spanning seven categories: file, structured data, knowledge, metric, artifact, prompt and voice.
The full answer is in How a Digital Workforce Works.
By comparing what the agent says, what the platform works out and what the agent actually did.
- Contract truthfulness monitoringThe platform checks whether each agent I/O contract still tells the truth, comparing the knowledge entry and relationship types an agent declares it writes against what its last run wrote, and the knowledge it reads against its declared reads, and flags an agent whose declaration has drifted from its behaviour.
- Intent is worked out from how an agent is builtThe platform works out what each agent does from what it declares and writes, rather than from a label: one of eight verbs (ingest, organise, analyse, decide, generate, query, act, monitor), read first from its knowledge-graph writes, then from the kinds and direction of its inputs and outputs, with its declared capabilities only as a tiebreaker.
- Per-agent control planeEach agent has one control surface covering identity and entitlement, contract and intent, model and behaviour, capabilities, governance, lifecycle and versions, control and containment, and assurance and cost.
The full answer is in How a Digital Workforce Works.
A statement of what it takes in and what it gives out, in shared types.
- Declared I/O contractsAn agent declares what it consumes and produces.
- Shared I/O type systemAgent inputs and outputs bind to a shared type registry spanning seven categories: file, structured data, knowledge, metric, artifact, prompt and voice.
- A team whose agents do not actually connectA team is an ordered sequence, so the platform checks the declared inputs and outputs along it and reports two silent failures: a handoff where one agent produces nothing the next reads, and an agent whose output nothing downstream or in the deliverable consumes.
The full answer is in How a Digital Workforce Works.
Like a member of staff, through one control plane.
- Per-agent control planeEach agent has one control surface covering identity and entitlement, contract and intent, model and behaviour, capabilities, governance, lifecycle and versions, control and containment, and assurance and cost.
- Versioned, diffable agent configurationAn agent configuration is a versioned object with its own lifecycle: versions can be listed, compared against each other, and published deliberately rather than taking effect on save.
- Per-agent and per-version performancePerformance is summarised per agent and per configuration version, with medians across the fleet and the ability to recompute.
The full answer is in How a Digital Workforce Works.
Three: domain teams do the work, platform teams evaluate it, service teams are called on demand.
- Domain, platform and service teamsAgents compose into three kinds of team: domain teams that do the business work of a workflow, platform teams that evaluate and improve that work as it runs, and service teams that other teams call as a shared service.
- Rule-triggered service teamsA service team can be invoked automatically when a declared condition holds after a team completes, in a workflow whose configuration turns service triggers on, rather than only when another team asks for it.
- Batched team execution with resumeA team runs its work in tracked batches, recording how many jobs are outstanding, completed and failed, and a batch that stops part way can resume from where it reached rather than starting the team again.
The full answer is in How a Digital Workforce Works.
As the work happens: platform agents evaluate each stage, and people are asked only when confidence is not enough.
- Agents evaluate other agents' work as it runsPlatform teams evaluate a workflow's work as it runs.
- Dialled autonomyHow much the compliance coordinator decides on its own is configuration, not a fixed posture.
- Human in the loopWhen a workflow is not confident enough to proceed it can interview a person and use the answers.
The full answer is in How a Digital Workforce Works.
As a sequence of teams, checked before it starts and evaluated as it goes.
- Workflows as composed teamsWorkflows compose agents into teams with declared inputs and outputs.
- Workflows refuse before they spendA workflow is checked before it starts, not part way through.
- What this will cost, before you run itBefore a workflow is launched the platform resolves the agents it would run, sums their credit tiers, converts the total to money at the enterprise contracted rate, and reports whether it fits the annual pool and what proportion of the pool it would consume.
The full answer is in How a Digital Workforce Works.
Only the earlier work it declares it needs, not everything that has happened so far.
- Each agent is handed the earlier work it declaresWithin a workflow, an agent built on the platform base is handed earlier work by type: its contract lists the types it consumes and where to look for each, and the platform resolves them from the outputs of agents that declare the matching type, the run's files, its launch parameters and interview replies.
- The living requirements document goes to named rolesA workflow names which agents act as scribe, enricher, curator and validators of its requirements document.
- Declared I/O contractsAn agent declares what it consumes and produces.
The full answer is in How a Digital Workforce Works.
In the knowledge graph, where every entry carries who may see it.
- Knowledge graphAgents read and write a shared knowledge graph through a governed seam that carries clearance and tenancy checks, rather than querying it directly.
- Knowledge domainsThe graph is divided into named domains rather than being one undifferentiated store, so knowledge can be organised, granted and reasoned over per subject area.
- Personal knowledge workspacesA person can hold their own workspace in the knowledge graph, scoped to them rather than shared with their organisation, so early or unfinished thinking has somewhere to live before it becomes a shared concern.
The full answer is in How a Digital Workforce Works.
Your own material goes in; finished deliverables come out.
- Bring your own documentsFiles can be uploaded directly to a workflow and read by its agents as a declared input source, so work starts from the organisation's own material rather than from what the platform already knows.
- Generated deliverablesWorkflows can produce readable deliverables such as reports and decks through a shared composition engine, rather than leaving output as raw model text for someone to format.
- Automatic renditionsA deliverable can carry derived formats: audio narration, PDF, a slide deck, a slide PDF, a PowerPoint file and interactive HTML.
The full answer is in How a Digital Workforce Works.
Four teams turn an investment case into an evaluated verdict and report.
- Business Case EvaluationAuthors and evaluates an investment case end to end.
- Intent is worked out from how an agent is builtThe platform works out what each agent does from what it declares and writes, rather than from a label: one of eight verbs (ingest, organise, analyse, decide, generate, query, act, monitor), read first from its knowledge-graph writes, then from the kinds and direction of its inputs and outputs, with its declared capabilities only as a tiebreaker.
The full answer is in How a Digital Workforce Works.
Eight teams turn a set of requirements into stories, criteria, tests and a readiness verdict.
- Requirements IntelligenceParses, classifies, scores and traces requirements: quality and uniqueness scoring, gap, dependency and duplicate analysis, identification of the applications a requirement touches, traceability checking, and generation of user stories, acceptance criteria, test scenarios and a requirements report.
- Intent is worked out from how an agent is builtThe platform works out what each agent does from what it declares and writes, rather than from a label: one of eight verbs (ingest, organise, analyse, decide, generate, query, act, monitor), read first from its knowledge-graph writes, then from the kinds and direction of its inputs and outputs, with its declared capabilities only as a tiebreaker.
- A team whose agents do not actually connectA team is an ordered sequence, so the platform checks the declared inputs and outputs along it and reports two silent failures: a handoff where one agent produces nothing the next reads, and an agent whose output nothing downstream or in the deliverable consumes.
The full answer is in How a Digital Workforce Works.
By exception: people set the thresholds and handle what the agents are not sure of.
- Dialled autonomyHow much the compliance coordinator decides on its own is configuration, not a fixed posture.
- Run quality never suppresses governanceWhether a piece of work needs human review is decided by declared configuration alone.
- Agent handbrakeA platform administrator can pause an agent from its control plane, which suspends it for every organisation so it is refused at its next dispatch, and resume it later, with the agent's containment state and last transition shown alongside.
The full answer is in How a Digital Workforce Works.
What the platform does for you, layer by layer.
Strategy and portfolio
- Business Case Evaluation
- OKR Intelligence
- AI Maturity Assessment
- Business Maturity Self-Assessment
- AI Maturity Export
Requirements and delivery
- Requirements Intelligence
- Requirements Intake
- Delivery Design
- Release Notes
- Delivery Intelligence
Client and market
- Client Intelligence
- Market Intelligence
- Meeting Intelligence
- Mail Intelligence
Modernisation
- Legacy System Migration
- SWIFT Message Migration
- Data Pipeline Conversion
Operations and regulation
- Regulatory Documentation
- Operational Flow Analysis
Knowledge and reference
- Domain Discovery
- Knowledge Operations
- Reference Architecture Intelligence
- Platform Guide
Home and conversation
- Notifications
- A follow-up keeps the subject
- What needs you now
- Bring your own documents
- Live workflow updates
- Operators get a reduced workspace
- A workflow card shows how the run really ended
- One homepage, with the chat on it
- A chosen workflow waits on its card
- What you can run, and what it needs
- Past conversations, findable
- A basic or an advanced workspace
- Search or jump to anything
- The chat says it is an AI
- One menu, organised around the work
- What the deployment ships, module by module
- Every run in My Work opens
- Run history across workflows
- A conversation says what changes, for a screen reader
Deliverables
- One definition of what an output is
- Artifacts and collections
- Dashboards read from knowledge graphs as the viewer
- The set of forms is configuration, not code
- Render an existing output again, safely
- Automatic renditions
- Generated deliverables
- Formats appear as they finish
- Ready-made dashboards a module offers each organisation
Module screens and cards
- Module-contributed interface
- A module-specific deliverable looks like itself
- A distinctive card without shipping code
- A module adds next steps to a finished workflow card
- A module supplies its own launch-time field
Design system and accessibility
- Token-based design system
- Each page announces its own name
- Accessibility and reflow are checked on each route
- Enforced interface consistency
- Light and dark themes
- Destructive actions confirm with a second press
- The interface works on a narrow screen
Adoption insight
- Where people fall out of adoption
- Which teams took it up, and where seats sit idle
- Which agents people actually reach for
- The headline adoption measures
Sources
- Account source of truth
- Bring the material you already have
- Meeting notes arrive from Granola on their own
- Client email is synced from named mailboxes
Kinds and roles
- Specialised agent implementations
- How an agent can be reached
- Standalone agent execution
- Interactive agent sessions
- Assistant agents
- Kinds of agent
- Try an agent on different models and prompts, and compare the results
Documents and media
- Voice primitive
- Audience-targeted documentation generation
- Screenshots and diagrams are read by a vision model
Registry and versions
- Capability registry
Fleet and control plane
- Agents allocated to a team
Teams and workflows
- Workflows as composed teams
- A template resolution chain is built and not yet wired in
- Curated workflow catalogue
- Workflow Lab: the anatomy of a workflow
Conversation to work
- Conversation resolves to declared capabilities
- Platform-wide and module-specific capabilities
- A declared capability answers honestly before it is built
- Cheap matching first, model call last
Scheduling
- Scheduled workflows
People in the loop
- Human in the loop
- The platform asks like a specialist
Graph and search
- Knowledge can be searched by meaning
Domains and scopes
- Knowledge domains
- Knowledge layers
- Personal knowledge workspaces
- Five domain scopes
- Knowledge graphs can be created, loaded, checked for health and archived from one page
Ingestion and provenance
- Source artefacts become typed knowledge
Identity and access
- Modules switched on for an organisation or for one team
Agent building blocks
- Reports are composed as a typed model
Builds and releases
- Release notes written from the tickets in a release
Weighing us against the tools you already know? See how we compare, with sources →
The full answers, written for you.
How a Digital Workforce WorksAbout 25 minutes
The Platform for an Agentic Workforce in Financial MarketsAbout 25 minutes
Bring a piece of work.
Choose a first workflow with us and see what a digital workforce does with it.
Promenaut
