Reference article
Coordination AI
An emerging approach to artificial intelligence focused on maintaining shared coordination state across people, agents, conversations and time.
Coordination AI is AI that maintains continuity between what people decide and what people and agents ultimately do. It helps groups preserve, update and act from a trustworthy shared coordination state across conversations, systems and time.
Terminology note. “Coordination AI” is an emerging term rather than a settled academic or industry classification. Afoot uses the term to describe a category of systems whose primary function is maintaining and improving coordination rather than merely generating, summarizing or executing work.
Contents
Definition
Coordination AI is AI that maintains continuity between what people decide and what people and agents ultimately do by helping a group preserve, update and act from a trustworthy shared coordination state.
In this context, coordination means the process by which multiple people or agents align their actions around shared work. It includes reaching decisions, establishing alignment, making commitments, assigning ownership, escalating issues, deferring unresolved matters and returning to them when conditions change.
The central problem is not simply whether information exists. A meeting may be recorded, a transcript may be searchable and a task may be entered into a project management system while the group still lacks a shared understanding of what actually holds. Coordination AI therefore treats communication as evidence from which coordination state can be proposed, confirmed and maintained.
The coordination layer
Coordination AI can be understood as infrastructure for a layer of organizational activity that sits between communication and execution. Between these layers sits coordination: the decisions, commitments, alignments, escalations and unresolved matters that must remain coherent as work proceeds.
Organizations already use extensive software to support communication: email, messaging, video meetings, documents and increasingly AI meeting assistants. They also use extensive software to manage execution: project management systems, CRMs, issue trackers, workflow automation and autonomous agents.
Between these layers sits coordination. Conversations produce decisions, commitments, alignments, escalations and unresolved matters that must remain coherent as work proceeds. Ownership changes. New information modifies earlier assumptions. Priorities shift. Work completed in one place affects work elsewhere.
Communication → Coordination → Execution
Historically, much of this continuity has been maintained manually. Managers remember what was agreed. Chiefs of Staff reconstruct context. Project leads chase commitments. Operations teams route information between groups. Participants repeat earlier discussions when the original context has been lost.
Coordination AI attempts to make this middle layer explicit and persistent. Rather than relying primarily on human memory to carry coordination state from one interaction to the next, the system maintains a shared record that can be updated as new evidence arrives.
Coordination as continuity
From this perspective, coordination is not a single meeting, handoff or workflow. It is the continuity that allows decisions and commitments made in one context to remain available, attributable and actionable in another. The coordination layer connects the moment in which shared intent is formed with the later work that depends on it.
This becomes increasingly significant as AI systems participate directly in execution. Agents can act quickly, but their actions are only as coherent as the state from which they operate. A persistent coordination layer can give humans and agents access to the same accepted representation of what has been decided, what remains open, who owns what and what has changed.
Human coordination functions
Before Coordination AI, much of organizational continuity has been maintained by people performing a recurring set of coordination functions. These functions may belong to managers, Chiefs of Staff, operations leaders, program managers, product leaders or other cross-functional roles rather than to people with dedicated coordination titles.
Coordination architecture
Coordination architects keep the organization legible. They maintain an understandable picture of how decisions, responsibilities, initiatives and dependencies relate to one another as the organization changes. Their function is to preserve the map: what connects to what, where authority sits, and how new information changes the existing structure.
Information pipeline ownership
Information pipeline owners make sure signals reach the right places. They connect information produced in one context with the people, systems, decisions or work occurring elsewhere. Their function is routing: determining who needs to know what, and reducing the likelihood that important context remains trapped in meetings, messages or organizational silos.
Intrapreneur orchestration
Intrapreneur orchestration leads prevent work from colliding. They identify overlapping initiatives, conflicting decisions, unclear ownership and dependencies that cannot be resolved within a single stream of work. Their function is orchestration: surfacing collisions and bringing the appropriate people together to resolve them.
These functions have traditionally depended heavily on human memory, organizational knowledge and active follow-up. Coordination AI can externalize part of this continuity work into persistent infrastructure: maintaining the map, routing relevant state and detecting potential collisions, while leaving judgment and resolution to people.
Record → Route → Surface for resolution
The objective is not to eliminate the human coordination roles. It is to make the underlying functions less dependent on a particular person’s memory and vigilance. A persistent coordination layer can keep the organization legible, move relevant state to where it matters and identify situations that require human judgment.
The coordination problem
Work involving multiple people creates coordination overhead. Decisions may be discussed without being clearly closed. Ownership may be implied rather than accepted. Commitments may be mentioned once and disappear into private notes or memory. An issue may be deferred without a return point. Later conversations may modify earlier agreements without everyone learning that the shared state has changed.
Teams often compensate for this uncertainty manually. Leaders reconstruct context, ask for updates, repeat decisions, clarify ownership and schedule additional alignment meetings. Managers, Chiefs of Staff, operations leaders and project leads frequently become the human continuity layer: keeping the organization legible, routing signals to the right places and noticing when work is about to collide. As the number of participants, projects, communication channels and AI agents increases, this cognitive burden grows.
Coordination AI does not imply that these human roles disappear. It externalizes part of the continuity work into shared infrastructure so that people can spend more of their attention on judgment, direction, negotiation and exception handling rather than acting as organizational memory.
Coordination AI is intended to reduce this burden by making coordination itself a persistent system object rather than leaving it as an informal by-product of communication.
Coordination state
A key concept in Coordination AI is coordination state: the present-tense condition of what a group has coordinated and what remains active. For a team or initiative, this may include what is decided, aligned, committed, open, owned and currently in need of attention.
Coordination state is distinct from a coordination record. The record is the durable accumulated history from which the current state can be derived. The state is what is true or active now.
This distinction is important because coordination changes over time. A commitment can move from open to fulfilled or blocked. An escalation can be resolved. A later discussion can dispute an earlier alignment. A new decision can be related to an earlier decision. Coordination AI must therefore do more than extract information from a single conversation; it must maintain continuity across events.
Shared state and personal orientation
The full coordination state of a team can be large, while the portion relevant to any one person at a given moment is usually small. A Coordination AI system can therefore separate the durable shared record from personal orientation: what needs a particular person now, and what recently changed in work to which that person is connected.
How Coordination AI works
Implementations can vary, but a Coordination AI system generally combines semantic inference with persistent application state. AI interprets unstructured communication; the application preserves the resulting coordination objects, human corrections and relationships over time.
Capture → Extract → Confirm → Organize → Follow up → Re-enter
Capture
The system receives evidence of a coordination event, such as a meeting transcript, notes, messages, calendar context or related files. The input may be incomplete or messy; the objective is not to create a perfect transcript archive but to preserve what may need to carry forward.
Extract
A domain-specific inference process identifies proposed coordination artifacts and supporting evidence. Instead of treating every sentence as equally important, it distinguishes durable coordination outcomes from discussion, explanation and background context.
Confirm
Relevant participants or owners review what the AI inferred. They can accept the proposed record, correct it or reject it. The system thereby distinguishes machine inference from socially accepted coordination state.
Organize and connect
Confirmed artifacts are connected to people, initiatives, source events and related historical artifacts. New conversations can be compared with the accumulated record to identify continuity, possible conflicts, unresolved items and relevant changes.
Follow up and re-entry
Open coordination should not disappear simply because a conversation ended. Coordination AI can return an unresolved commitment, escalation, deferral or disputed decision to attention when a deadline, planning cycle or later event makes it relevant again.
Human confirmation and trust
Coordination AI differs from systems that treat AI extraction as authoritative. An AI model can infer that a person appears to have made a commitment, but that inference does not by itself create legitimate accountability.
Confirmation is the process by which the responsible humans accept, correct or reject an AI-proposed coordination artifact. It creates a boundary between what the machine inferred and what the group is willing to treat as part of its shared record.
This has both epistemic and social importance. Human conversation is ambiguous, and models can misinterpret it. More importantly, accountability requires agency. A system that silently converts probabilistic interpretations into obligations risks becoming a surveillance or management system rather than coordination infrastructure.
For this reason, a Coordination AI design may use different confirmation rules for different artifact types. A personal commitment can be confirmed by its owner, while a collective decision or alignment may require confirmation from the people who participated in it.
Coordination artifacts
A coordination artifact is a durable output of a coordination event that affects future coordination. Afoot’s current model uses five primary artifact types:
Decision
A conclusion or choice that the relevant participants intend to treat as settled unless it is later changed or disputed.
Alignment
A shared understanding or position among participants that matters to how subsequent work should proceed.
Commitment
An undertaking accepted by a specific owner to do, deliver or resolve something.
Escalation
An issue that requires attention, authority or input beyond the person or context in which it arose.
Deferral
A matter intentionally left unresolved for later return rather than silently abandoned.
These objects differ from ordinary meeting notes because they have identity and state. They can be attributed, confirmed, connected to an initiative, linked to source evidence and revisited as circumstances change.
Comparison with adjacent software
Coordination AI overlaps with several existing categories but has a different primary object of concern. The distinction is not that adjacent systems are incapable of supporting coordination; it is that their core data model and optimization target are different.
Category
Primary object
Typical question
Difference from Coordination AI
Meeting assistants
Meeting content
What was said?
Coordination AI follows what must remain true, owned or unresolved after the meeting and across later events.
Task management
Tasks and execution
What work needs doing?
Coordination AI focuses on the social and informational layer that establishes why work exists, who accepted responsibility and what shared understanding supports it.
Project management
Plans, projects and progress
Where is the project?
Coordination AI is concerned with the evolving agreements and commitments that allow project state to remain coherent.
Knowledge management
Documents and information
What do we know?
Coordination AI distinguishes informational knowledge from coordination state: what the group has actually agreed to carry forward.
AI agents
Autonomous actions
What can the system do?
Coordination AI helps determine what humans and agents should be acting from together, including the current accepted state and its provenance.
Coordination between humans and AI agents
The growth of autonomous and semi-autonomous AI agents increases the importance of explicit coordination state. A human can often repair ambiguous context through conversation, intuition or social cues. An agent generally acts on the context it is given.
As AI reduces the cost and time required to execute work, coordination can become a larger constraint on organizational performance. Faster execution does not resolve ambiguity upstream. If different people or agents act from different versions of a decision, AI can convert unclear or stale coordination into incorrect execution more quickly.
Coordination AI therefore has a potential role as shared infrastructure between human judgment and agentic execution: preserving which decisions and commitments have actually been confirmed, what remains unresolved and what changed later. The objective is not to establish an objective or permanent “truth,” but to maintain the group’s current accepted coordination state and its provenance.
In this model, AI is not only an executor of work. It also helps maintain the conditions under which multiple actors can execute coherently. Humans and agents can operate from the same coordination record while humans retain authority over what becomes accepted state.
Afoot as a Coordination AI system
Afoot is an implementation of Coordination AI focused initially on the post-meeting coordination layer for bounded teams. It uses a domain-specific inference model to analyze captured meeting material and propose coordination artifacts, while the application maintains continuity across meetings.
Afoot’s current product model separates per-meeting inference from persistent coordination state. A meeting-level model proposes decisions, alignments, commitments, escalations and deferrals. Humans confirm those proposals. The application then stores the confirmed coordination record and connects later meetings to the accumulated history.
An organize-and-connect step compares new material with existing initiatives, open items and related artifacts. Its purpose is to help maintain continuity rather than reinterpret the organization’s history from scratch after every conversation.
Specialized coordination agents
Afoot implements coordination functions through a set of specialized AI agents rather than relying on a single general-purpose assistant. Different agents are responsible for different parts of the coordination lifecycle, including extracting coordination artifacts, connecting new events to accumulated state, coordinating confirmation, following up on open commitments and resolving items when they re-enter attention.
Some of these agents encode functions historically performed manually by managers, Chiefs of Staff and project leads. They help keep the coordination map current (coordination architects), route relevant state to the appropriate people and contexts (information pipeline owners), and surface overlaps or conflicts that require human resolution (intrapreneur orchestration leads). Other specialized agents perform additional functions within the coordination lifecycle.
Coordination ontology
The agents operate against Afoot’s coordination ontology: a structured model of the entities, states and relationships that matter to human coordination. The ontology defines concepts such as decisions, alignments, commitments, escalations, deferrals, people, initiatives, ownership, confirmation, provenance and lifecycle state, as well as the relationships between them.
This ontology is what allows Afoot to treat a meeting transcript as evidence rather than as the product itself. A language model provides semantic interpretation of unstructured conversation; the ontology provides the domain model through which that interpretation becomes structured coordination state.
Together, the specialized agents and coordination ontology form the core reasoning system of Afoot. The agents perform bounded coordination work; the ontology gives that work a common model of what exists, how it relates and what changes over time. The persistent application layer then stores the human-confirmed record so later agents, people and conversations can operate from the accumulated state rather than reconstructing context from scratch.
Afoot is designed around a bounded team rather than the entire enterprise as its initial unit of value. The intended outcome is a trustworthy shared record combined with low-burden personal orientation: people see what needs them and what changed on work they are connected to, while the team leader can maintain awareness without manually reconstructing the team’s state.
Afoot explicitly does not position its core product as task management, project management, employee monitoring, performance scoring, sentiment analysis or a generic meeting summarizer. Meetings are treated as evidence and provenance; the durable object is the coordination that must carry forward.
Limitations and design constraints
Coordination AI depends on interpretation of human communication and therefore inherits the ambiguity of that communication. Missing participants, incomplete transcripts, private conversations and context outside the system can all reduce the completeness of the record.
Persistent coordination systems also create governance questions. Teams must be able to distinguish proposals from confirmed facts, understand provenance, correct attribution and avoid converting transparency into individual surveillance.
A further limitation is that not all coordination should be formalized. Teams rely on tacit knowledge, judgment and temporary ambiguity. The objective of Coordination AI is therefore not to capture everything or eliminate human negotiation. It is to preserve the subset of coordination that future action depends on, while leaving humans in control of what becomes authoritative.
Frequently asked questions
What is Coordination AI?
Coordination AI is AI that maintains continuity between what people decide and what people and agents ultimately do by helping a group preserve, update and act from a trustworthy shared coordination state.
How is Coordination AI different from a meeting assistant?
Meeting assistants primarily capture or summarize individual meetings. Coordination AI treats meetings as source material and maintains state across them, following decisions, commitments and unresolved coordination over time.
How is Coordination AI different from task management?
Task management systems organize execution. Coordination AI focuses on the coordination layer beneath execution: what people agreed, who accepted ownership, what changed, and what needs to return to attention.
Why does Coordination AI require human confirmation?
AI inference can propose what appears to have been decided or committed, but an inferred statement should not become a trusted social fact without human agency. Confirmation lets the responsible people accept, correct or reject the proposed record.
What is coordination state?
Coordination state is the current condition of a team’s shared coordination: what is decided, aligned, committed, open, owned and in need of attention now. It is derived from a durable coordination record.
Terminology
Coordination AI
AI that maintains continuity between what people decide and what people and agents ultimately do by helping a group preserve, update and act from a trustworthy shared coordination state.
Coordination event
A meeting or conversation in which coordination occurs and which can serve as evidence for proposed artifacts.
Coordination artifact
A durable coordination-relevant output, such as a decision, alignment, commitment, escalation or deferral.
Coordination record
The accumulated durable record of coordination artifacts, their confirmation and lifecycle state, provenance, relationships and history.
Coordination state
The current condition derived from the record: what is decided, aligned, committed, open, owned and in need of attention.
Confirmation
Human acceptance, correction or rejection of an AI-proposed artifact before it enters the trusted record.
Re-entry
The return of unresolved or changed coordination to human attention when new evidence, a deadline or a planning boundary makes it relevant.
Coordination ontology
A structured model of the entities, states and relationships involved in coordination, used by Afoot’s agents to interpret events and maintain consistent coordination state over time.
Specialized coordination agent
An AI agent with a bounded responsibility within the coordination lifecycle.
See also
Multi-agent systems
Computer-supported cooperative work
Organizational coordination
Workflow management
Knowledge management
Decision support systems
Human-in-the-loop artificial intelligence
About this article
This page is maintained by Afoot as a reference definition for the term Coordination AI and the concepts used in Afoot’s product model. The term is emerging and its definition may evolve as the category develops.
Coordination AI
Artificial intelligence category
Purpose
Maintain shared coordination state
Unit
Groups of people and agents coordinating over time
Inputs
Conversations, transcripts, notes, messages and contextual records
Outputs
Confirmed coordination artifacts and current state
Core objects
Decisions, alignments, commitments, escalations, deferrals
Trust model
AI proposes; humans confirm
Example
In one sentence
Coordination AI maintains continuity between what people decide and what people and agents ultimately do.