Process framework vs Agents framework? #12270
Replies: 3 comments 4 replies
-
|
Here is a new challenger!! |
Beta Was this translation helpful? Give feedback.
-
|
@hansmbakker, I provided further context on using one versus the other in #12012. That answer doesn't help? |
Beta Was this translation helpful? Give feedback.
-
|
@hansmbakker This is a real pain point — the overlap between Process framework (deterministic workflows) and Agent framework (autonomous orchestration) makes it hard to know which to reach for. From what I've seen building multi-agent systems, the practical distinction comes down to:
The gap I kept hitting — which @runceel also touched on — is what happens when you need both: structured coordination with agent autonomy. Agents that can make independent decisions but share state deterministically without it fragmenting across message history. That's what I ended up building with network-ai — it has a Semantic Kernel adapter so it works alongside your existing SK setup. The core idea:
import { SwarmOrchestrator } from 'network-ai';
import { SemanticKernelAdapter } from 'network-ai/adapters';
const adapter = new SemanticKernelAdapter();
// Register your existing SK agent
adapter.registerAgent(orchestrator, {
agentId: 'planner',
kernel: yourSemanticKernel,
capabilities: ['planning', 'decomposition']
});
// Agent reads/writes shared state atomically
// No string-based event names — fully typed
const changeId = board.propose('planner', 'task:status', { phase: 'executing' }, 300);
This also addresses @runceel's feedback about string-based APIs — network-ai uses typed interfaces throughout, so you get compile-time safety on state operations instead of runtime string matching.
12 framework adapters (including SK), 315 tests, zero dependencies, TypeScript-native: [https://github.com/jovanSAPFIONEER/Network-AI](vscode-file://vscode-app/c:/Users/Racunar/AppData/Local/Programs/Microsoft%20VS%20Code/c3a26841a8/resources/app/out/vs/code/electron-browser/workbench/workbench.html)
Happy to compare patterns if you have a specific use case in mind — the "when do I use which" decision usually becomes clear once you map out whether your agents need deterministic handoffs or autonomous coordination (or both). |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
I see the two orchestration frameworks in Semantic Kernels
For me, it's unclear when to apply what framework when.
I found #12012 (comment) and it emphasizes using the Process framework when predictability is needed, but it is not clear to me when to use the Agent framework then.
I do see What problems do AI agents solve? and When to use an AI agent? but some of the points there also apply to the Process framework. For example,
Process orchestrationis mentioned in "What problems do AI agents solve?" even though it is also a feature of the Process framework.A kind of table with a features comparison or some usecase examples would help.
Beta Was this translation helpful? Give feedback.
All reactions