Agents that do the work between the screens - scoped to one process, grounded in your SAP data, guarded where it matters, and measured so you can tell whether they earned their place.
Most agent programmes stall because they start too wide. An agent asked to “handle procurement” has no measurable definition of success; an agent asked to resolve one class of blocked delivery does. We start where the process is painful, repetitive and well-bounded, prove it, and only then let the flock grow.
The boid metaphor is the point: simple agents following clear local rules produce useful system-level behaviour. One large agent that tries to hold the whole process in its head produces something much harder to debug.
Use-case analysis, then a working prototype, then evaluation and guardrails, then production - with observability wired in from the start rather than added after the first incident. Every agent we ship reports into Boidra Platform, and failures in SAP AIF can route into automatic resolution.
If you are still deciding where Joule fits, start with our guide to SAP Joule, or see the full delivery lifecycle.
A report answers a question you already knew to ask. An agent handles the long tail - reconcile these two documents, find why this delivery is blocked, draft the correction and route it for approval. The value is in the steps between the screens, not in the screens themselves.
Both, and the answer depends on where the process lives. Joule is the right home for SAP-native conversations in cloud systems. A2A-native agents on BTP cover on-premise S/4HANA, non-SAP systems in the same flow, and anything Joule does not reach yet. Because we build on the open A2A protocol, the second grows into the first without a rebuild.
Scope, guardrails and evidence. Agents get a narrow tool set rather than open access, write actions carry confirmation steps and value limits, and every run is traced and scored so a regression shows up as a number rather than as a surprise in production.
We aim for one scoped, measurable process rather than a platform programme. That usually means weeks, not quarters - and a result you can judge honestly before deciding whether the flock grows.
Tell us what you’re trying to automate in SAP. We’ll tell you - honestly - whether an agent is the right tool, and how we’d start.
