Which SDK do you default to?

None. Multi-turn tool use with reflection before writes tends to be Claude. Structured routing tends to be OpenAI. Long cyclic work with approval gates tends to be LangGraph. The step decides.

Can Agentforce call our Python agent?

Yes, as a peer over A2A through the gateway, with the user’s identity carried across. Not as a shared key pointed at an internal URL.

Do you replace existing automation?

Not on day one. Deterministic automation that works stays. We move the steps that need judgement, and say which ones do not.

Runtime plane

Not every step deserves a model.

The first week is in the org, reading what already exists. Cut the process into steps, mark each one deterministic or adaptive, then choose the runtime. Agentforce where the permissions already live, a code agent where the exception lives, and a plain function where nothing needs judgement. The honest no: if the work is a software factory with a named delivery date, we are the wrong partner.

Get the written assessment

Certified Partner since 2010 · MVP Hall of Fame · 200+ agents in production · UAE and US desks

Runtime questions

What it is

Four runtimes, one classification.

The runtime plane is where a step actually executes. Choosing it is a two-question test: is the step deterministic, and is a human waiting on it?

Why it matters

One runtime for everything is how programmes stall.

If you have only one seat to put work in, you will put the wrong work in it. These are the four we find.

Peers

Agentforce is a peer, not the ceiling.

Agent-to-agent delegation with the user’s identity carried across the wire. The gateway still sees every hop.

01
Declared skills
Each agent publishes what it can do and what scopes it needs. A registry, not a prompt that lists its friends.
02
A2A delegation
One agent asks another as a peer over a standard envelope, instead of a private wrapper per pair.
03
Identity travels
The user’s authority carries forward, so delegation cannot become privilege escalation.
04
Hop budgets
Traces, max hops, and timeouts stop cyclic delegation before it becomes an incident.

Two runtimes as peers. A code agent and an Agentforce agent discover each other through a registry and delegate over A2A, with on-behalf-of identity and hop budgets enforced at the gateway.

Mermaid source

flowchart TD
  U["User on a surface"] --> G["Gateway: identity, policy, audit"]
  G --> B["Broker or registry: declared skills"]
  B --> AF["Agentforce agent: org-native step"]
  B --> CA["Code agent: SDK or LangGraph"]
  AF <--> CA
  CA --> T["MCP tools"]
  AF --> R["Salesforce records"]

In an org

A manufacturer.

The case step is org-native. The exception is not. Both are needed in the same process.

01The standard caseAgentforce answers the installed-base question, checks entitlement, and drafts the response inside the sharing model.

02The exceptionA part is discontinued and the spec sits in a supplier document. A code agent reads, reasons, and proposes a substitution with citations.

03The hand-offThe code agent returns a structured recommendation to the Agentforce agent over A2A. The service rep sees one thread.

04The writeThe substitution posts through the durable path, because the ERP has to accept it and a person has to be able to reverse it.

How Mindcat helps

Where we would start in your org.

We classify the steps before anyone picks a framework. That conversation is short and it saves a quarter.

Questions

What we actually say.

Which SDK do you default to?: None. Multi-turn tool use with reflection before writes tends to be Claude. Structured routing tends to be OpenAI. Long cyclic work with approval gates tends to be LangGraph. The step decides.
Can Agentforce call our Python agent?: Yes, as a peer over A2A through the gateway, with the user’s identity carried across. Not as a shared key pointed at an internal URL.
Do you replace existing automation?: Not on day one. Deterministic automation that works stays. We move the steps that need judgement, and say which ones do not.

The brief · one email

Send one process, step by step.

We will mark which steps are deterministic and which deserve a model.