Short answer first: the Salesforce admin role is not dying, but the half of it that was mostly clicking is. What replaces it is harder, more valuable, and more accountable. If you are an admin, or you hire them, plan for that shift now rather than when the org already has five agents nobody owns.
This comes from running a live Salesforce implementation and building our own CRM alongside it. We have watched where the hours actually go, and where they are starting to go instead.
What agents take off the plate
A large share of classic admin work is translation: someone describes what they want, the admin turns it into fields, page layouts, validation rules, list views, and simple flows. That translation step is exactly what generative tooling is good at. Describe the field, get the field. Describe the flow, get a draft.
It is not perfect, and it should be reviewed, but it moves the bottleneck. Building a screen is no longer the scarce skill. Knowing whether that screen should exist is.
What gets more important
When software can act, not just display, the questions change:
- What is this agent allowed to touch? Which objects, which fields, which actions, under whose permissions.
- What happens when it is wrong? Is the action reversible, is it logged, who gets alerted.
- Is the data it reasons over true? Duplicates, stale stages, and ungoverned fields become safety issues, not just reporting annoyances.
- How do we know it is working? An agree rate, a resolution rate, an escalation rate, watched over time.
None of those are configuration tasks. They are governance tasks, and they land on whoever understands the org best. Today that is usually the admin.
The admin used to configure what people could do. Now the job is deciding what software is allowed to do on its own.
Where demand is moving
Across the ecosystem the direction is consistent, even if nobody can honestly put a precise number on it:
Architects. People who can hold data model, security, integration, and process in their head at once and make trade-offs explicit. Agents multiply the cost of a bad architecture, so the people who prevent one are worth more.
Data 360 and data engineering. Unifying sources, tuning identity resolution, classifying sensitive data, and designing what flows into the unified profile. This is the foundation agents stand on, and few teams have the skill in house.
Integration. MuleSoft specifically, or API design generally. Agents are only as useful as the systems they can reach, and every action an agent takes outside Salesforce is an integration someone has to build and secure.
Analytics with judgement. Tableau and CRM analytics, but less about building charts and more about defining the metrics that tell you whether an agent is helping or quietly hurting.
Talk to us about your Salesforce team→
The skills we would build now
If we were coaching an admin today, this is the order:
- Data modelling and Data 360 fundamentals. Understand how records become unified profiles and where that goes wrong.
- Flow, properly. Fault paths, bulkification, entry criteria that do not fire on every save. Agents call flows as actions, so flaky flows become flaky agents.
- Security and sharing, deeply. Permission sets, field-level security, and what an integration or agent user can actually see. This is where most real incidents start.
- Read Apex. You do not need to write a trigger framework, but you need to review what an agent action is calling and spot when it does too much.
- Agent design. Scoping a topic, writing clear instructions, defining actions with tight inputs, and testing with real, messy cases before go-live.
- Talking about risk. Explaining to a sales director why an agent ships in recommend mode first is a skill. It is often the one that decides whether the project succeeds.
Certifications help you get noticed. A small, well-governed agent you designed and can explain helps you get hired.
For the people hiring
Do not read "agents can configure" as "we need fewer Salesforce people." Read it as "we need different ones." Fewer pure screen builders, more people who own outcomes. Give your existing admins the time to learn data and governance, because they already know where the bodies are buried in your org, and that context is the one thing no agent brings with it.
The tooling is getting easier. The responsibility is getting heavier. Hire, and train, for the responsibility.
Good to know
Are Salesforce admin jobs going away because of AI?
The role is changing, not disappearing. Routine configuration such as building simple fields, list views, and basic flows is increasingly assisted or generated. The work of deciding what agents may do, keeping data trustworthy, and owning outcomes grows.
Which Salesforce skills matter most for the AI era?
Data modelling and Data 360, Flow design with proper error handling, security and sharing, enough Apex to read and review it, integration basics (MuleSoft or APIs in general), and agent design: topics, actions, guardrails, and testing.
Should an admin become a developer or an architect?
Most should move toward architecture rather than pure code. The scarce skill is judgement across data, security, and process, and being able to explain trade-offs to the business. Learn to read Apex, but invest more in design.
We engineer ideas into working software.
Tarka · logic · Shila · stone