---
title: "Synthesis: From ABPMS (2023) to APM (2026) — what changed"
type: synthesis
tags: [abpms, apm, manifesto, evolution, comparison]
sources: ["[[sources/2023-dumas-ai-augmented-bpms]]", "[[sources/2026-calvanese-agentic-bpm-manifesto]]"]
created: 2026-04-13
updated: 2026-04-13
---

# Synthesis: From ABPMS to APM — what changed

A side-by-side comparison of the **AI-Augmented BPMS Manifesto** ([[sources/2023-dumas-ai-augmented-bpms|Dumas et al. 2023, ACM TMIS]]) and its agent-centric successor, the **Agentic Process Management Manifesto** ([[sources/2026-calvanese-agentic-bpm-manifesto|Calvanese et al. 2026, arXiv]]). Authored 2026-04-13.

## 1. Core paradigm shift: system-augmentation → agent-centricity

| | **ABPMS (2023)** | **APM (2026)** |
|---|---|---|
| Subject of augmentation | The **system** (BPMS) is augmented with AI | The **agent** is the primary functional entity |
| Position | "AI-augmented information system" | "Agentic socio-technical system" |
| Agency | Implicit — agents used, not first-class | Explicit — first-class citizens (human / software / embodied / AI) |
| Driving question | "How do we augment a BPMS with AI?" | "How do we govern autonomous agents within a process frame?" |

The APM Manifesto states this explicitly: ABPMS *"leaves the agentic perspective implicit"*. APM makes it explicit. This is the central shift.

## 2. Characteristics → capabilities (3 + 1 mapping)

ABPMS's four **characteristics of the system** become APM's four **capabilities of the agent** — plus one new dimension:

| ABPMS characteristic | → | APM capability |
|---|---|---|
| **Adaptable** | → | [[concepts/self-modification]] (adaptation aspect) |
| **Proactive** | → | [[concepts/self-modification]] (evolution aspect) + [[concepts/perceive-reason-act]] |
| **Explainable** | → | [[concepts/explainability-apm]] (preserved, elevated to first-class) |
| **Context-sensitive** | → | [[concepts/framed-autonomy]] + [[concepts/process-awareness]] |
| *(no counterpart)* | + | **[[concepts/conversational-actionability]]** ← genuinely new |

**Conversational actionability** is the new dimension absent from ABPMS — agents must converse (with humans and other agents) and link conversation to enactment (Query / Recommend / Create / Execute).

APM also imposes an explicit **ordering** of capabilities (framing → explainability → actionability → self-modification) that ABPMS does not — because ABPMS describes the *system*, while APM constructs the *agent* layer by layer.

## 3. "Frame" — terminological re-framing

| | **ABPMS** | **APM** |
|---|---|---|
| Frame = | *Operational* specification (≈ BPMN/DECLARE) | *Normative* specification (deontic) |
| Content | How the agent should act | What the agent must / must not do |
| Example | "Send email → update DB → archive" | "Never delete a rejected submission without notifying the student" |

APM explicitly distinguishes **operational frame** vs **normative frame** — and identifies ABPMS's frames as *operational* in this terminology. A sharpening of the theoretical foundation.

## 4. New explicit architecture

ABPMS did not specify an explicit *agent* architecture. APM adds:

- **Macro level (APM System):** the "world", framing mechanism, goal alignment, protocols (MCP, ACP, RDF, KQML).
- **Micro level (APM Agent):** [[concepts/perceive-reason-act|Perceive–Reason–Act]] loop with three modules (perception, reasoning, action).
- **Foundational ontologies & frameworks:** DOLCE, GFO, [[frameworks/bdi-agents|BDI]], [[frameworks/fipa|FIPA]], [[frameworks/tropos-i-star|i*/Tropos]].
- **Agent categories:** human, software, embodied, AI.

## 5. Self-modification: new explicit distinction

APM introduces the clean split **adaptation vs evolution** absent from ABPMS:

- **Adaptation** — short-term, instance-specific, non-persistent.
- **Evolution** — long-term, shared across instances, alters the process model itself.

This is APM's mechanism to operationalise what ABPMS called "adaptable".

## 6. Research agenda — tightened and broadened

| | **ABPMS** | **APM** |
|---|---|---|
| Challenges | "A set of challenges" (not enumerated by name in abstract) | **19 named** (F1–F4, X1–X5, A1–A5, M1–M5) + **5 cross-cutting** (C1–C5) |
| New themes in APM | — | Prompt-injection security (C2), benchmark contamination (C3), EU AI Act / liability (C4), engineering methods (C5), agent-oriented process mining (C1) |
| Frameworks built on | Trustworthy AI (general) | LLM agents, MAS, MCP/ACP, deontic logic, normative MAS, "learning to defer" |

## 7. Author continuity — same community, next generation

Of the 14 ABPMS authors, **8 recur** in the APM Manifesto: [[entities/marlon-dumas|Dumas]], [[entities/fabiana-fournier|Fournier]], [[entities/lior-limonad|Limonad]], [[entities/andrea-marrella|Marrella]], [[entities/marco-montali|Montali]], [[entities/diego-calvanese|Calvanese]], [[entities/giuseppe-de-giacomo|De Giacomo]], [[entities/marcello-la-rosa|La Rosa]].

The 10 new APM authors come predominantly from the agent / MAS tradition ([[entities/sebastian-sardina|Sardiña]], [[entities/timotheus-kampik|Kampik]], [[entities/artem-polyvyanyy|Polyvyanyy]], [[entities/niek-tax|Tax]]) — mirroring the conceptual shift from system to agent perspective. **APM is not a competing paradigm — it is the next-generation continuation of the same intellectual collective.**

## 8. What did NOT change

- **Trustworthy AI as ethical anchor** — preserved.
- **Explainability as first-class** — preserved (in fact reinforced in APM with 4 desirable properties).
- **Process-aware information system as substrate** — preserved; APM builds *on* PAIS, not in place of it.
- **BPM lifecycle as reference frame** — preserved; APM re-interprets but does not reject it.

## 9. One-sentence summary

> ABPMS asked *"how do we add AI to a BPMS?"*; APM asks *"how do we govern autonomous agents acting under normative frames?"* — and answers it by elevating the agent to first-class status, reframing ABPMS's "frames" as *operational* (vs the new *normative* APM frames), reordering ABPMS's four characteristics into four ordered agent capabilities, and adding **conversational actionability** as a fifth dimension that ABPMS did not foreground.

## 10. Practical use of this synthesis

- **For new readers**: read ABPMS first (system view), then APM (agent view). The terminology only fully clicks after seeing both.
- **For comparing future BPM/AI vision papers**: use the ABPMS quartet and the APM capability stack as anchor frames.
- **For research scoping**: APM's named challenges (F1–F4, X1–X5, A1–A5, M1–M5, C1–C5) are the most concrete current research agenda — most ABPMS-era questions are subsumed.

## Connections
**Sources:** [[sources/2023-dumas-ai-augmented-bpms]] · [[sources/2026-calvanese-agentic-bpm-manifesto]]
**Concepts:** [[concepts/agentic-bpm]] · [[concepts/framed-autonomy]] · [[concepts/process-awareness]] · [[concepts/explainability-apm]] · [[concepts/conversational-actionability]] · [[concepts/self-modification]] · [[concepts/perceive-reason-act]]