Healthcare organisations are one of the few sectors where coordination failures carry a direct and measurable cost in human lives. Traditional management models fail here because they treat hospitals as linear systems, ignoring the dense networks of interdependent actors whose interactions produce unpredictable outcomes. This situation results in a structural fragmentation of authority among doctors, nurses, administrators, and trustees, without a unified mechanism for resolving conflicts between the various professional roles. When failures occur, frontline staff manage them with stopgap solutions that restore immediate functionality while leaving root causes intact and silently eroding productivity. In this context, delay is not only inefficient, indeed, in cases like sepsis, it is lethal. Agentic AI introduces a qualitatively different possibility: unlike previous paradigms that passively awaited instructions, agentic systems initiate, adapt, and coordinate multi-step workflows with minimal human supervision. Where fragmented authority prevents any single actor from coordinating across diverse professional roles, autonomous agents are not subject to the same constraints. Yet deployment remains surprisingly limited: only 3% of US healthcare organisations have moved from pilots into live clinical workflows, a gap attributable to fragmented data environments, insufficient governance, and unprepared workforces. The partnership between Tampa General Hospital and Palantir Technologies offers one of the first operational windows into this technological transition. Among the most significant results documented were approximately 700 lives saved, $40 million in savings, and an 83% reduction in patient placement time. However, these results belong to the conventional analytics phase and represent returns on data infrastructure investment, not on agentic AI. The genuinely agentic deployment that followed restructured administrative workflows but left clinical bottlenecks largely unresolved. What placed Tampa General among the 3% was the deliberate decision to build the infrastructure foundation before deploying the technology. Looking ahead, sustainable adoption of this technology can be linked to fundamental strategic decisions: an actionable data architecture; a multidisciplinary embedded team working across clinical, administrative, and technical boundaries; a targeted training programme to address institutional scepticism; and a financial oversight function ensuring deployments generate value proportionate to their cost. The institutions that will lead the next phase will not be those waiting for the technology to mature, but those building these foundations today.
Le organizzazioni sanitarie sono uno dei pochi settori in cui i fallimenti di coordinamento comportano un costo diretto e misurabile in vite umane. Qui i modelli di gestione tradizionali falliscono perché trattano gli ospedali come sistemi lineari, ignorando le fitte reti di attori interdipendenti le cui interazioni producono risultati imprevedibili. Questa situazione si concretizza in una frammentazione strutturale dell'autorità tra medici, infermieri, amministratori e rappresentanti della governance, in assenza di un meccanismo unificato per risolvere i conflitti tra i diversi ruoli professionali. Quando si verificano dei problemi, il personale di prima linea li gestisce con soluzioni tampone che ripristinano l'operatività immediata lasciando intatte le cause profonde ed erodendo silenziosamente la produttività istituzionale. In questo contesto, il ritardo non è soltanto inefficiente, in condizioni come la sepsi, è letale. L'intelligenza artificiale agentiva introduce una possibilità qualitativamente diversa: a differenza dei paradigmi precedenti che attendevano passivamente le istruzioni, i sistemi agentivi avviano, si adattano e coordinano flussi di lavoro articolati con una supervisione umana minima. Laddove la frammentazione dell'autorità impedisce a un singolo attore di coordinarsi attraverso i diversi ruoli professionali, gli agenti autonomi non sono soggetti agli stessi vincoli. Eppure il deployment rimane sorprendentemente limitato: solo il 3% delle organizzazioni sanitarie statunitensi è passato dai progetti pilota a flussi di lavoro clinici operativi, un divario attribuibile ad ambienti di dati frammentati, governance insufficiente e forza lavoro impreparata. La partnership tra Tampa General Hospital e Palantir Technologies offre una delle prime finestre operative su questa transizione tecnologica. Tra i risultati più significativi documentati figurano circa 700 vite salvate, 40 milioni di dollari di risparmi e una riduzione dell'83% dei tempi di assegnazione dei posti letto. Tuttavia, questi risultati appartengono alla fase di analisi convenzionale e rappresentano ritorni sull'investimento in infrastruttura dati, non sull'AI agentiva. Il deployment genuinamente agentivo che ne è seguito ha ristrutturato i flussi di lavoro amministrativi, lasciando però largamente irrisolti i colli di bottiglia clinici. Ciò che ha collocato Tampa General nel 3% è stata la decisione deliberata di costruire le fondamenta infrastrutturali prima di implementare la tecnologia. Guardando al futuro, un'adozione sostenibile di questa tecnologia è legata a specifiche decisioni strategiche: un'architettura dati strutturata e azionabile; un team multidisciplinare integrato che operi attraverso i confini clinici, amministrativi e tecnici; un programma di formazione mirato a superare lo scetticismo istituzionale; e una funzione di supervisione finanziaria che garantisca che i deployment generino valore concreto e proporzionato al loro costo. Le istituzioni che guideranno la prossima fase non saranno quelle che aspettano che la tecnologia maturi, ma quelle che costruiscono oggi queste fondamenta.
Agentic AI as an Organizational Layer: Decision-Making Processes and Financial Impacts in the Healthcare Sector
CAUTI, DAVIDE
2025/2026
Abstract
Healthcare organisations are one of the few sectors where coordination failures carry a direct and measurable cost in human lives. Traditional management models fail here because they treat hospitals as linear systems, ignoring the dense networks of interdependent actors whose interactions produce unpredictable outcomes. This situation results in a structural fragmentation of authority among doctors, nurses, administrators, and trustees, without a unified mechanism for resolving conflicts between the various professional roles. When failures occur, frontline staff manage them with stopgap solutions that restore immediate functionality while leaving root causes intact and silently eroding productivity. In this context, delay is not only inefficient, indeed, in cases like sepsis, it is lethal. Agentic AI introduces a qualitatively different possibility: unlike previous paradigms that passively awaited instructions, agentic systems initiate, adapt, and coordinate multi-step workflows with minimal human supervision. Where fragmented authority prevents any single actor from coordinating across diverse professional roles, autonomous agents are not subject to the same constraints. Yet deployment remains surprisingly limited: only 3% of US healthcare organisations have moved from pilots into live clinical workflows, a gap attributable to fragmented data environments, insufficient governance, and unprepared workforces. The partnership between Tampa General Hospital and Palantir Technologies offers one of the first operational windows into this technological transition. Among the most significant results documented were approximately 700 lives saved, $40 million in savings, and an 83% reduction in patient placement time. However, these results belong to the conventional analytics phase and represent returns on data infrastructure investment, not on agentic AI. The genuinely agentic deployment that followed restructured administrative workflows but left clinical bottlenecks largely unresolved. What placed Tampa General among the 3% was the deliberate decision to build the infrastructure foundation before deploying the technology. Looking ahead, sustainable adoption of this technology can be linked to fundamental strategic decisions: an actionable data architecture; a multidisciplinary embedded team working across clinical, administrative, and technical boundaries; a targeted training programme to address institutional scepticism; and a financial oversight function ensuring deployments generate value proportionate to their cost. The institutions that will lead the next phase will not be those waiting for the technology to mature, but those building these foundations today.| File | Dimensione | Formato | |
|---|---|---|---|
|
tesi_Cauti_Davide_.pdf
accesso aperto
Dimensione
2.47 MB
Formato
Adobe PDF
|
2.47 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/112497