The Agentic AI Observatory
From task automation to system-level coordination.
Agentic AI matters not only because it can perform individual tasks, but because it can coordinate decisions and actions across people, systems and organisations. The Observatory examines when that coordination remains inside an existing workflow and when it begins to change the firm, the ecosystem and the distribution of power and value.
For leaders, this separates an efficiency opportunity from a potential change in how the company or market is organised.
Product announcements show what agents are claimed or demonstrated to do. The Observatory examines what those capabilities could change in practice and how far the available evidence supports that conclusion.
The portfolio
Agentic coordination operates under different constraints in procurement, commerce, travel, logistics, financial transactions, hiring and energy. Each domain tests a different way in which AI could change a workflow, a firm or an ecosystem.
Coordinates requirements, suppliers, approvals, authoritative records and payment.
When agents enter organisational buying, who controls the transaction and captures the value?
Commerce
Coordinates buyer intent, merchant offers, payment and fulfilment.
What happens when buyers, sellers, merchants, platforms and their agents begin transacting?
Travel
Coordinates an end-to-end journey across independently controlled bookings.
Can agents assemble, book, modify and recover a journey across multiple providers?
Logistics
Coordinates digital instructions, custody and physical flows.
Can coordination become consequential without control of physical custody?
Financial Transactions
Coordinates mandates, accounts, risk, payment and settlement.
When can an agent initiate or complete a financial action, and who bears the consequence?
Hiring
Coordinates candidates, evaluation, decisions and accountability.
Can practical influence move before formal hiring authority changes?
Energy
Coordinates distributed physical assets within markets and grid constraints.
Does agentic coordination decentralise control or strengthen existing coordinators?
Each domain answers the same questions differently, because the object being coordinated, the institution that authorizes an outcome and the market relationships around it are different in each one.
Compare the domains
The domains differ in physicality, reversibility, institutional authority, multi-party coordination and market structure. These differences help reveal where agents can coordinate work within existing systems and where they may begin to reorganize those systems.
Agentic Procurement
When agents enter organizational buying, who controls the transaction - and who captures the value?
Procurement makes the distinction between action and control visible in approvals, contracts, authoritative records and payment states.
Explore the Procurement ObservatorySelect a domain to read its record
What happens when buyers, sellers, merchants, platforms and their agents begin transacting?
Coordinates buyer intent, merchant offers, payment and fulfilment.
Buyer authority, merchant rules, identity and the transaction infrastructure
Do agents shift power from merchant interfaces toward whoever controls buyer intent, identity and transaction access?
Two sides of a market can each field agents, so the merchant interface stops being the only place where intent, identity and transaction access are held.
Can agents assemble, book, modify and recover an end-to-end journey across multiple providers?
Coordinates an end-to-end journey across independently controlled bookings.
Provider rules, traveler authority and cross-provider coordination
Do agents replace aggregators, or become a more powerful orchestration layer above them?
One journey is many separately owned bookings, so a failure in one provider has to be recovered across all of them by an actor none of them controls.
Where can agents redirect, release, expedite and recover physical flows - and who authorizes the result?
Coordinates digital instructions, custody and physical flows.
Custody, carrier authority, customs and physical infrastructure
Can digital coordination become consequential without control of physical custody and institutional release?
The object moves in the physical world under someone's custody, so a coordinating agent can be correct and still require a carrier, a customs authority or a warehouse to act.
When can an agent initiate, authorize or complete a financial action - and who bears the consequence?
Coordinates mandates, accounts, risk, payment and settlement.
Identity, delegated consent, custody, regulation and the settlement infrastructure
Does agentic progress expand autonomy, or primarily produce more precise machine-enforceable delegation?
Authority to move money is explicitly regulated and explicitly delegated, so the boundary between what an agent may prepare and what it may complete is written down rather than inferred.
Where can agents source, evaluate, select, offer and onboard people - and which decisions must remain institutionally attributable?
Coordinates candidates, evaluation, decisions and accountability.
Employment authority, fairness, contestability and accountability
Can workflow authority shift upstream through ranking and screening even when final hiring authority remains human?
The outcome is a decision about a person that has to remain contestable and attributable, so screening can concentrate influence long before any formal authority changes hands.
Can agents coordinate distributed energy assets well enough to change how energy is produced, consumed, traded and controlled?
Coordinates distributed physical assets within markets and grid constraints.
Asset control, grid authority, safety, regulation and settlement
Does agentic coordination create decentralized markets, improve incumbents, or strengthen central platforms and system operators?
Physical assets clear against a real-time market under a system operator's safety authority, so coordination is bounded by grid rules rather than by the coordinating software.
Offer, order and customer relationship
Purchase, payment, fulfilment or return
Buyer authority, merchant rules, identity and the transaction infrastructure
Moderate
Two-sided market mediated by platforms
Do agents shift power from merchant interfaces toward whoever controls buyer intent, identity and transaction access?
End-to-end journey assembled from interdependent bookings
Booking, modification, cancellation and disruption recovery
Provider rules, traveler authority and cross-provider coordination
Moderate
Fragmented network of airlines, hotels, aggregators, insurers and payment providers
Do agents replace aggregators, or become a more powerful orchestration layer above them?
Shipment, inventory and custody state
Release, routing, dispatch, redirection and physical recovery
Custody, carrier authority, customs and physical infrastructure
Low to moderate
Multi-party physical network
Can digital coordination become consequential without control of physical custody and institutional release?
Money, account position and financial mandate
Transfer, trade, payment, allocation or settlement
Identity, delegated consent, custody, regulation and the settlement infrastructure
Low to moderate
Regulated transaction network
Does agentic progress expand autonomy, or primarily produce more precise machine-enforceable delegation?
Candidate, role and employment relationship
Shortlisting, rejection, offer, contracting and onboarding
Employment authority, fairness, contestability and accountability
Low after commitment
Employer-candidate-platform-institution system
Can workflow authority shift upstream through ranking and screening even when final hiring authority remains human?
Distributed physical asset, capacity and energy position
Dispatch, consumption adjustment, storage, trading and grid response
Asset control, grid authority, safety, regulation and settlement
Low to moderate
Regulated physical network with real-time markets
Does agentic coordination create decentralized markets, improve incumbents, or strengthen central platforms and system operators?
Reversibility describes the characteristic difficulty of undoing consequential actions. It is a research variable, not a rating of domain suitability.
Procurement holds evidence; the other six domains do not. These are the comparisons the portfolio is built to make, not comparisons it has made.
The shared framework
Task automation changes how work is performed. Agentic coordination can change how the system is organised. Three levels separate the two, and each asks a question whose answer may be no.
Task automation
Performs defined work inside existing rules, dependencies and decision rights.
Agentic coordination
Organises interdependent work across changing conditions, steps, systems or actors.
The testDoes the agent merely complete work, or determine how interdependent work is organised?
Workflow
How are agents reorganizing the work or transaction journey?
What would establish changeAgents coordinate stages, handoffs, systems or actors differently rather than only improving one task.
Executive implicationRedesign the workflow around the new form of coordination.
Firm
How does that coordination change the operating model?
What would establish changeCapabilities, configuration authority, accountability, systems or governance are organized differently.
Executive implicationDecide which capabilities the firm must own, configure, govern or safely source from providers.
Ecosystem
How does agentic coordination change competition?
What would establish changeNew actors enter, control points become contestable, relationships adapt or strategic positions change.
Executive implicationIdentify where to defend, reposition or create a new category.
ReshuffleA reshuffle occurs when these changes redistribute control, bargaining power, strategic position or value.
The Observatory distinguishes what is announced, demonstrated and observed from what remains an interpretation, conditional possibility or unresolved question. Evidence standing and methodology
Six lenses organise the underlying research. They are how the work is done rather than something a reader needs in order to follow the four questions above.
For leaders
The level at which coordination is changing determines the decision leaders need to make.
Task automation
Is the efficiency worth implementing?
Workflow coordination
What must be redesigned around the agent?
Firm reconfiguration
Which capabilities, accountabilities and dependencies must change?
Ecosystem adaptation
Which relationships and control points should we defend or build?
Reshuffle
How should we reposition as power and value move?
The Observatory identifies which decision the current evidence justifies, and which would be premature. A more consequential conclusion requires stronger evidence.
It also identifies the evidence that would show whether the system is moving further.
Apply the research
Use the research to assess your company, build an evidence-backed instrument for clients, or establish a continuing Observatory for a domain.
For leadership teams examining workflows, decision architecture, operating models and competitive position.
Determine where agentic coordination could change your workflows, decision rights, dependencies and ecosystem position, and which conclusions the current evidence supports.
For advisory firms that need a repeatable evidence-backed instrument for a market, category or strategic question.
Create an evidence-backed way to show your clients how agentic AI may change a market, category, ecosystem, operating model or strategic scenario.
For platforms, associations, investors, institutions and companies seeking a continuing research programme.
Commission a continuing evidence programme for your domain, market or ecosystem, published with its evidence standing attached.