Speaking twice at Dreamforce · Sept 15-17 →

Salesforce MVP Hall of Fame · Certified Partner since 2010

Agents are an architecture, not a chat box.

The first week is in the org, reading what already exists. Five planes decide whether a digital worker is production or a demo: where people meet it, what it is allowed to call, what runs the step, what it remembers, and how the write survives a timeout. 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

Salesforce and MuleSoft map

The five planes

How the work is organised.

Read down. Skip one of these in your org and you get a specific failure, not a general slowdown.

  1. 01

    Experience

    Slack · Experience Cloud · Portals · Voice · Generative UI

  2. 02

    Control and governance

    Identity on the hop · Policy · Rate limits · Audit

  3. 03

    Runtime and agents

    Agentforce · Code agents · SDKs · A2A · MCP tools

  4. 04

    Context and memory

    Data Cloud · Graph · Vectors · Decision history

  5. 05

    Durable capability

    Queues · Retries · IDP · Long-running jobs

Five decoupled planes for digital labor. People meet the work on the experience plane. Every call passes the control plane. Runtimes execute, memory grounds, and the durable plane owns the write.
Mermaid source
flowchart TD
  EXP["1 Experience: Slack, Experience Cloud, portals, voice"] --> CTL["2 Control: identity, policy, rate, audit"]
  CTL --> RUN["3 Runtime: Agentforce, code agents, SDKs"]
  CTL --> MEM["4 Memory: source of truth, working context, durable memory"]
  CTL --> DUR["5 Durable: queues, retries, long jobs, writes"]
  RUN --> MEM
  RUN --> DUR
  DUR --> SOR["Systems of record: Salesforce, ERP, files"]

How Mindcat helps

Diagnostic, first slice, then the operating model.

Three engagements that already exist. No new pricing product, no invented savings.

  1. 01

    Diagnostic

    Where the governance debt sits, which writes have no rollback, and which steps should never have been a model call. Readiness and governance are the existing services.

  2. 02

    First slice

    One named desk end to end: experience surface, gateway path, runtime, memory, and a write someone can reverse. Shadow mode before write authority.

  3. 03

    Operating model

    Specifications, guardrails, and a review cadence so the next agent is a configuration, not a project.

How an architecture engagement sequences. Read the estate, ship one governed slice on a named desk, then move the team onto specifications and fleet operations.
Mermaid source
flowchart TD
  A["Diagnostic: map governance debt and write blast radius"] --> B["First slice: one desk, one runtime, one governed write"]
  B --> C["Operating model: specs, shadow mode, fleet ops"]

Questions

What we actually say.

Is this a replacement for the five-layer AI stack?
No. The stack is what we install: models, agents, memory, identity, governance. The planes are how the work is organised around it. Same practice, two views.
Do we need all five planes on day one?
No. You need to know which one you are skipping and what that costs. Most stalled pilots skipped control or memory and called it a platform problem.
Is this Salesforce-only?
No. Salesforce and MuleSoft carry most of it in the orgs we work in, so that is the language we use in a diagnostic. The planes still hold when your runtime is a code agent and your system of record is a warehouse.
Where does the first slice usually land?
A desk with a named owner, a documented process, and a write that can be reversed. Claims, invoices, cases, and field service are the usual four.

The brief · one email

Name the desk. Name the write.

We will tell you which plane is missing and what it costs to skip it.