Salesforce has changed more in the last eighteen months than it did in the five years before that. Product names moved. The pricing model moved. And the questions buyers ask moved right along with them.
A while back, the question everyone asked was whether AI agents could do anything genuinely useful inside a CRM. That one is mostly settled. The questions now are harder and a lot more practical. What does this cost when volume triples? What data does it need to be accurate? Who is accountable when it gets something wrong? And how do we prove it actually worked?
That shift matters, because it moves the hard part of a Salesforce program away from the demo and toward the plumbing. Agents are easy to show and difficult to run well. The teams seeing real returns this year usually are not the ones with the most AI features switched on. They are the ones who cleaned up their data, tightened their permissions, picked two or three workflows worth automating properly, and built a way to measure whether anything improved.
Here are the nine trends we think deserve your attention in 2026, why each one is happening, what it changes for your business, and where the trade-offs sit. Not every trend applies to every company, so each section is clear about who it hits hardest and what you can safely leave alone for now.
TL;DR
The Year Salesforce Stopped Demoing and Started Operating
Agentforce left the pilot stage, Data 360 absorbed Informatica, and a chunk of your bill turned variable. Salesforce trends in 2026 are less about new features and more about the plumbing underneath: clean data, tight permissions, and two or three workflows worth automating properly.
Where Salesforce Buyers Get Stuck This Year
Most teams can build an agent. Fewer can forecast consumption pricing at five times volume, prove their CRM data is accurate enough, secure years-old connected apps, or name who owns an agent after go-live. That gap is where budgets quietly disappear and projects get canceled.
Nine Trends, Scored Against Your Own Org
Inside: nine evidence-backed Salesforce trends, scenario tables mapping priorities to your size, data quality, and regulatory exposure, plus a seven-step framework for deciding whether to adopt Salesforce AI now. It also flags what to act on immediately and what you can safely postpone.
The short version: what actually changed between 2025 and 2026
If you only have two minutes, here is the summary before we get into the detail.
- AI agents left the pilot phase. The work is now about running them safely, not building them.
- A chunk of your Salesforce bill is now variable, because AI is billed by usage rather than by seat alone.
- Data quality stopped being a nice-to-have. It is the gate everything else waits behind.
- Governance became engineering work, not a policy paragraph.
- Your biggest security exposure is probably a third-party app you connected years ago and forgot about.
- Technical debt got more expensive, because agents act on top of whatever logic already exists.
- Admin, developer, and consultant roles are all moving toward judgment work.
Now the detail.
Trend 1: Agentforce moved from pilots into production
AI agents in Salesforce are no longer an experiment for most large customers. The work has shifted from building an agent to running one safely, and that is a very different kind of project with a very different budget shape.
First, what an “AI agent” actually means
The word gets thrown around loosely, so it is worth being precise. An AI agent is software that takes an instruction, decides which steps to take, and then acts inside your systems. It updates a record, checks an order status, drafts a reply, or triggers a workflow.
That is different from a chatbot, which mostly answers questions and hands you off to a human. It is also different from a Flow, which follows a fixed path you defined in advance. The autonomy is the whole point of an agent, and it is also where the risk lives. A Flow does exactly what you told it. An agent decides.
How big is this, really?
The adoption numbers are not marketing fluff. In its first quarter of fiscal 2027, Salesforce reported combined Agentforce and Data 360 annual recurring revenue of nearly $3.4 billion, with roughly $1.2 billion of that coming from Agentforce itself. More than half of those bookings came from existing customers expanding what they already had.
That last detail is the interesting one. Expansion revenue is a better signal than new logos, because it means a meaningful number of companies tried something, kept it, and then bought more. People do not usually expand a deployment that is quietly failing.
What changed on the platform side
The product matured alongside the demand. The Agentforce 360 release added a conversational builder, an agent scripting layer for tighter control over behavior, voice capabilities, and a way to ground agents in unstructured content such as PDFs, knowledge articles, and call transcripts.
The practical effect is that building an agent got noticeably faster. Governing one did not get easier by anything like the same margin. That gap is exactly where most 2026 projects run into trouble, and it is worth naming before you start rather than discovering it in month four.
What this changes about implementation
Scoping starts with a job, not a feature
“Deflect password reset cases” is a scope you can finish and measure. “Deploy Agentforce” is a budget line with no finishing condition. If your project charter reads like the second one, tighten it before anyone touches a sandbox.
Testing looks completely different
You are no longer only checking whether logic fires correctly. You are testing judgment across messy, real-world inputs. That means building a test set from actual historical cases, including the ugly ones, rather than from tidy happy-path examples someone wrote in a workshop.
Escalation design is a deliverable, not an afterthought
Deciding when an agent must stop and hand off to a person is usually worth more than adding another capability. Write those stop conditions down early, and treat them as part of the build rather than something you tune later.
Somebody owns the agent after go-live
Agents drift as your products, policies, and data change underneath them. An unowned agent gets quietly worse over months, and nobody notices until a customer complains. Name the owner before launch, not after.
Where to start if you are early
Go narrow rather than broad. Pick one high-volume, low-variance workflow, instrument it properly, and measure a genuine before-and-after. That is the approach we take on Agentforce consulting engagements because a small workflow with clean measurement will teach you far more about your own readiness than a wide pilot that touches everything and proves nothing.
Trend 2: Consumption pricing is now the hardest part of the budget
Salesforce AI is largely billed by usage, not by seat. That single change affects planning more than most of the product announcements combined, because your bill now moves with volume and with how many steps each interaction takes.
The ways Salesforce sells AI right now
Salesforce currently offers several buying models. Its Agentforce pricing page lists credits charged per action, a per-conversation option, and per-user licensing. The worked examples on that page make the underlying math clear: cost is driven by the number of actions in a typical interaction, multiplied by how many interactions you have. Salesforce also notes those examples exclude data platform credits and other consumption services.
That last exclusion is the line item buyers most often forget to model, and it is frequently the largest one. The data layer underneath your agents is metered separately from the agents themselves.
What you are buying | What drives the bill | Where teams get surprised |
Per-action credits | Actions inside each interaction, multiplied by interaction volume | Real workflows quietly use several actions per customer question |
Per-conversation pricing | Number of agent sessions, whatever happens inside them | Good value for long, complex sessions and poor value for short, high-volume ones |
Per-user licensing | Headcount with access | Predictable, but it does not cover the data platform underneath |
Data platform consumption | Records ingested, stored, processed, and queried | Often the biggest line item, and billed separately from agent usage |
Where the forecast usually breaks
The trap is not the headline rate. It is the gap between a demo interaction and a real one.
In a demo, a customer asks one question, and the agent takes two actions. In production, that same question triggers an identity lookup, a knowledge search, an order check, a record update, and a follow-up message. Five actions instead of two means your forecast was off by more than double before anyone did anything wrong. Multiply that across a contact center handling thousands of cases a week, and the miss stops being academic.
How to model it before you sign
- Estimate actions per interaction for your two or three real use cases, not for the demo version.
- Multiply by realistic monthly volume based on your actual case or lead numbers.
- Rerun the same model at two times and five times that volume.
- If the five-times number is unacceptable, redesign the workflow rather than hoping volume stays flat.
This is also why we push clients through an AI readiness review before a contract conversation rather than after one. Knowing which workflows are worth automating and roughly how many steps each needs turns a vague procurement negotiation into a specific one. It also gives you a defensible reason to buy less in year one, which is almost always the right call.
Trend 3: Data readiness is the gate everything else waits behind
The strongest predictor of whether Salesforce AI works in your business is the state of your data, not the quality of the model. This is not a new idea. What is new is that 2026 is the year it stopped being advice and became a budget line.
Why Salesforce bought Informatica
Salesforce made its position official when it closed the Informatica acquisition in November 2025, bringing data cataloging, data quality, governance, metadata management, and master data management onto the platform alongside Data 360 and MuleSoft. Marc Benioff summed up the reasoning bluntly, saying that without clean, connected, trusted data there is no intelligence, only hallucination.
Strip away the framing, and the strategic message is straightforward. Salesforce now wants to own the readiness layer, not just the activation layer. For customers, that signals where product investment is heading, and it also signals that Salesforce expects data preparation to be a paid, ongoing workstream rather than a one-time cleanup before go-live.
What “AI-ready data” actually means
The phrase gets used to sell a lot of things, so let us be specific. It means five conditions are true for the workflow you are automating and only for that workflow. You do not need a perfect org. You need a clean path through the part of the org your agent will touch.
- One reliable version of each customer, so the agent is not picking between three contact records with different phone numbers.
- Fields that mean the same thing everywhere, including in the systems you integrate from. Two teams using “Status” differently will produce confidently wrong answers.
- Knowledge content that is current and owned. Agents ground answers in what you give them, so a three-year-old policy article becomes a three-year-old answer delivered instantly and with total confidence.
- Permissions that reflect reality. An agent inherits access, so an over-permissioned service account becomes an over-informed agent.
- Lineage you can trace. When an answer is wrong, you need to find which record or document caused it, or you cannot fix the root cause.
If your data is genuinely messy
Be honest about it and delay customer-facing agents. Start with internal ones instead, where a wrong answer costs an employee two minutes rather than costing you a customer. Getting the underlying layer right through Data 360 work is unglamorous, and it is usually the difference between an agent that quietly saves money and one that quietly creates cleanup work for your service team.
Trend 4: AI governance became real work, not a policy document
Governance in 2026 means controls that live inside the system, not a paragraph in a document nobody reads. The reason is simple. When AI only wrote text, the worst outcome was saying something wrong. When AI takes actions, the worst outcome is doing something wrong, then doing it quickly, repeatedly, and across a lot of customer records.
What the research says
Most organizations know this and have not caught up. The McKinsey AI trust survey of roughly 500 organizations found that only about a third reached a meaningful maturity level in strategy, governance, and agentic AI controls, even as overall responsible AI maturity improved. Nearly two-thirds named security and risk concerns as the top barrier to fully scaling agentic AI, ahead of both regulatory uncertainty and technical limitations.
In other words, the brake is confidence in controls, not capability. Two other findings from that research are worth carrying into your own planning. Active mitigation lags behind risk awareness across almost every risk category, which means teams can usually name the risks they are not managing. And organizations with explicit ownership of responsible AI scored materially higher than those without a clearly accountable function.
That second point is the cheapest win available to you. Naming an owner costs nothing and moves the needle more than most tooling decisions.
A governance checklist that fits on one page
- Write down what each agent may never do. Refunds above a threshold, contract changes, and anything touching regulated data are common stop conditions.
- Set the human checkpoint by risk level, not by how nervous the team feels. Low-risk reversible actions can run unattended. Irreversible or customer-visible commitments should not.
- Log agent actions and review a sample every week at first. Sampling catches drift long before a complaint does.
- Re-test after every Salesforce release and after any change to your knowledge base or data model.
- Assign one accountable owner per agent, with a named backup. “The AI team” is not an owner.
Where regulation sits right now
Regulation is moving in parallel, though not in a straight line. Rules such as the EU AI Act phase in across several years, and the timelines have been revised more than once, so confirm the dates and obligations that currently apply to your jurisdiction and industry rather than working from a deck written last year. If you are in a regulated sector, this is the point where Salesforce data governance stops being optional and becomes a prerequisite for shipping anything customer-facing.
Trend 5: Connected apps are your biggest new security exposure
The most damaging Salesforce data losses of the past year did not come from Salesforce being breached. They came from third-party applications that customers had connected to Salesforce themselves. That is a different problem, and it needs a different defense.
What happened, in plain terms
The Google Threat Intelligence Group documented the pattern clearly. In August 2025, a threat actor tracked as UNC6395 targeted Salesforce customer instances using compromised OAuth tokens tied to a third-party sales application, then systematically exported large volumes of data from numerous corporate orgs. The assessed intent was credential harvesting, which means attackers were mining support cases and records for secrets that would unlock other systems.
Note what was not the vulnerability. The core Salesforce platform held up fine. The trust relationship between Salesforce and a connected app was the way in. An OAuth token is essentially a long-lived key that lets one app act inside another on your behalf, and stolen tokens look like normal integration traffic until somebody notices the volume.
Why agents make this worse
Every integration you connect to an agent becomes a new path into your data. Open standards for wiring agents to external tools are making those connections dramatically easier to create, which is genuinely useful and genuinely risky at the same time. Easier to create also means easier to create carelessly. A connected app someone added quickly for one project in 2023 is exactly the kind of thing nobody thinks to revoke in 2026.
A connected-app hygiene checklist
- Inventory every connected app and integration user, then remove anything nobody can explain the purpose of.
- Apply least privilege to integration users. Most are set up with far broader object and field access than the integration actually uses.
- Rotate and revoke tokens on a schedule, not only after an incident.
- Scan case comments, notes, and description fields for pasted credentials. That is how a data theft turns into a wider compromise.
- Monitor for unusual bulk export activity. Large queries at odd hours are your clearest early warning.
Trend 6: Technical debt became the tax on every AI plan
Technical debt in Salesforce is the pile of shortcuts, duplicate automation, unused fields, and undocumented customizations that build up over years of fast delivery. It has always slowed teams down. In 2026 it does something worse. It makes AI unreliable, because agents act on top of whatever logic already exists.
What debt looks like in a real org
The problems are remarkably consistent across the orgs we assess. Overlapping automations update the same field in an unpredictable order, so you cannot give an agent a dependable rule about what happens after it acts. Duplicate and half-complete records feed agents contradictory context. Permission sets that grew organically over five years grant access nobody intended. Hundreds of unused fields add noise that makes it harder for people and AI alike to work out what the data model actually means.
None of that is exotic. It is just the accumulated result of shipping fast for a long time without a cleanup budget, which describes most Salesforce orgs over about four years old.
The uncomfortable twist with AI build tools
AI-assisted build tools make it faster to create new automation. For an experienced, well-governed team that is a real gain. For an org that already lacks governance, speed applied to an ungoverned environment simply produces more of the same problem, faster. If your admin can now generate five Flows in the time it used to take to build one, and nobody reviews them, you have not solved anything. You have accelerated the thing that was already hurting you.
What to fix first
You do not need to fix everything. You do need to know what is there, which automations are still live, and which parts of the org your agent will touch. That is why we usually recommend a Salesforce health check before an AI roadmap rather than after one. An inventory takes a couple of weeks. Unwinding a bad agent deployment takes considerably longer.
Trend 7: Integration is being redesigned around agents
Integration used to be mostly about moving data between systems on a schedule. Nightly syncs, batch loads, and a middleware layer nobody looked at unless it broke. Agents change the requirement completely, because an agent needs to read current information and take action in other systems while a customer is still on the line.
From nightly batch to real-time action
For integration-heavy businesses, that reframing has real consequences. Real-time needs replace batch tolerance for the specific data an agent depends on, which may mean rethinking a sync that has worked fine for years. Error handling becomes customer-visible because a failed API call now happens mid-conversation instead of quietly at 2am. And every new connection widens the security surface from the previous section, so integration design and security design stop being separate conversations.
Salesforce is consolidating this layer deliberately, pairing MuleSoft API management with the data integration, quality, and governance capabilities it picked up with Informatica. At the same time, open standards for connecting AI agents to external tools are becoming part of normal integration planning, which shifts the design question from “How do we move this data?” to “Which systems may this agent act in, and under what conditions?”
A sensible scope
Resist the urge to rebuild your whole integration estate for AI. Identify the two or three systems your priority agent genuinely needs, make those connections real-time and properly monitored, and leave everything else on its existing pattern. Most of the Salesforce integration services work we do in 2026 looks like that: targeted, reversible, and scoped to one use case rather than a platform-wide modernization nobody has the budget to finish.
Trend 8: Salesforce roles are shifting, consultants included
AI is not eliminating Salesforce admins, developers, or consultants. It is moving all three toward judgment work and away from execution work, and that is happening fast enough to be worth planning for rather than reacting to.
What Salesforce did to its own support team
Salesforce has been unusually open about the effect inside its own business. Marc Benioff said the company reduced support headcount from about 9,000 to roughly 5,000 as AI agents took on a share of customer conversations, with many staff redeployed into other functions rather than simply cut.
Whatever you make of the framing, the underlying pattern is instructive. The volume of routine, repetitive handling fell sharply. The need for people who design, supervise, and improve the system did not fall at all. If anything, it went up because somebody now has to watch the agents.
How admin and developer work is changing
Admins are being pulled upstream into data quality, permissions, automation governance, and agent ownership. That is closer to architectural work than the user-and-field administration the role started with, and it is a genuine step up in scope for people who want it. Developers are spending less time writing routine code and more time on system design, review, and deciding whether something should be built at all.
Consultants face the sharpest version of the change, because the parts of consulting that amounted to fast configuration are exactly the parts AI compresses hardest.
So will AI replace Salesforce consultants?
Not for the work that matters, but it will absolutely replace parts of it, and pretending otherwise would be dishonest.
Anything that amounts to translating a clear requirement into a standard configuration is getting cheaper and faster every quarter. Clients will rightly stop paying premium rates for it. What holds its value is the work AI cannot do for you: understanding why a process exists before automating it, deciding where an agent must stop, spotting the requirement that will create three years of technical debt, and being accountable for an outcome rather than a deliverable. If you are evaluating partners this year, that is the distinction worth testing for in the first meeting.
Trend 9: Delivery moved from big projects to continuous ownership
The one-and-done implementation was already a poor fit for a platform with three major releases a year. It fits even worse now that a meaningful share of your spend is usage-based and your AI features degrade quietly when the data behind them changes. Something that needs weekly attention cannot be handled by a project that ended in March.
Why so many AI projects get canceled
The scrutiny is deserved. A Gartner agentic AI forecast predicts more than 40% of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls. Gartner also warned about “agent washing,” where existing chatbots and automation tools get rebranded as agents without meaningful autonomy.
Look at what is missing from that list of causes: model capability. Nobody is failing because the AI is not clever enough. They are failing on scoping and governance, which are management problems with management solutions. That should be encouraging, because those are things you control.
When managed services earn their cost
The case for outside help gets stronger when you have multiple clouds, heavy integrations, regulated data, meaningful technical debt, or a single admin carrying a platform that outgrew one person some time ago. Where Salesforce managed services earn their keep is in the unglamorous continuity work: watching consumption before it becomes a budget problem, reviewing agent output samples, cleaning data on a schedule, and regression testing after each release.
One warning worth taking seriously. If a provider is only closing tickets, you are paying for a help desk and calling it optimization. Ask what they will proactively change in the org each quarter, and ask how they will report on it.
When you should keep it in-house
Plenty of companies should. If you have one or two clouds, a competent admin, a stable data model, and modest AI ambitions, you can run most of this yourself. What you need is protected time and a clear success measure, which are organizational problems rather than technical ones. Hiring a partner will not fix an admin who has no time, and it may just add a coordination layer on top of the same constraint.
Which trends matter most for your situation
Recommendations change a lot based on org maturity, data quality, complexity, and how regulated you are. The table below is a starting point for prioritizing, not a prescription, and it assumes you cannot do everything this year. Almost nobody can.
Your situation | Prioritize in 2026 | Reasonable to postpone |
Growing mid-market company, one or two clouds, reasonable data | One narrow agent use case with measurement, plus a consumption forecast before signing | Multi-agent orchestration, voice, and broad platform modernization |
Large enterprise, several clouds and regions | Agent governance, a clear ownership model, and one data layer decision across business units | Rolling agents out to every function at once |
Poor CRM data quality | Deduplication, field standardization, and knowledge cleanup for one workflow only | Any customer-facing agent until that workflow is clean |
Mature org with heavy technical debt | Health check, automation inventory, and permission review before adding AI | AI-assisted build tools that will accelerate existing sprawl |
Evaluating Agentforce for the first time | Define stop conditions, run a scoped pilot, model cost at two and five times volume | Enterprise-wide licensing commitments |
Extensive custom Apex development | Architecture review of what agents will invoke, plus safe action boundaries | Rewriting custom code purely to make it AI-friendly |
Integration-heavy business | Real-time access and monitoring for the two or three systems an agent truly needs | Full integration estate modernization |
Regulated organization in health, finance, or insurance | Human checkpoints, audit logging, data governance, and a connected-app review | Autonomous customer-facing actions on regulated data |
What to act on now, and what to just watch
| Act on now | Worth a scoped pilot | Monitor for now |
| Connected app and OAuth token review | One agent on a high-volume, low-variance workflow | Voice agents, unless you run a large contact center |
| Consumption forecasting before renewal | Internal employee-facing agents | Multi-agent orchestration across departments |
| Data cleanup for one target workflow | Grounding agents in existing knowledge content | Broad platform re-architecture around AI |
| Naming an accountable owner per agent | Measuring handle time or deflection honestly | Emerging agent interoperability standards |
| Automation and permission inventory | AI-assisted development inside a governed sandbox | Cutting roles based on projected AI savings |
A seven-step way to decide whether to adopt Salesforce AI now
If you take nothing else from this article, take this sequence. It is deliberately boring, and it is the difference between projects that survive to year two and projects that quietly vanish from the roadmap.
- Pick the workflow before the technology. Name a specific, repetitive, measurable process. If you cannot name one, you are not ready to buy.
- Measure the current state honestly. Handle time, deflection rate, error rate, cost per case, whatever fits. Without a baseline you will lose the budget argument in year two.
- Check the data behind that one workflow. Not the whole org. Just the records, fields, and knowledge content the agent will actually read.
- Decide the human checkpoint up front. What must a person approve, and what can run unattended? Write it down before anyone builds anything.
- Model cost at three volumes: today, two times, and five times.
- Build small, instrument everything, and review weekly. Sample real outputs. Aggregate dashboards will not tell you whether an agent is behaving sensibly.
- Decide who owns it on day 31. If that person does not exist yet, sort that out before go-live rather than after.
When to handle this internally, and when outside help is worth it
We are a Salesforce consulting partner, so we have an obvious interest here, and we would rather be straight with you about it.
Plenty of the work described above is well within reach of a capable internal team, and we say so to prospects regularly. The calculation changes when complexity compounds. Multiple clouds with overlapping data models, regulated data, heavy integrations, years of accumulated automation, or an admin team already at capacity are the conditions where outside help genuinely pays for itself. The same is true when a decision needs defending to a board or when a previous implementation left problems nobody internally has the standing to unwind.
Twenty years of delivery work has taught us that most failed Salesforce programs were not beaten by the technology. They were beaten by unclear scope, unowned data, and nobody being accountable after go-live. If you want a second opinion on where your org actually stands before committing to an AI roadmap, our Salesforce consulting services team is happy to walk through your current state and tell you plainly which of these trends apply to you and which you can ignore this year.
Where this is all heading
The through-line across all nine trends is that Salesforce work in 2026 rewards discipline over enthusiasm. The capability gap between vendors has narrowed considerably. The gap between organizations that prepared their data, defined their controls, and measured their results and those that did not is widening quickly, and it shows up as wasted spend rather than as a dramatic public failure.
So the useful question for the rest of this year is not which Salesforce features to adopt. It is which two or three business problems are worth solving properly, what would have to be true for an AI agent to solve them safely, and who owns the result once the project team moves on. Companies that answer those three questions well usually find the technology decisions get much simpler afterward.
If it would help to pressure-test your 2026 Salesforce plan against what we are seeing across other implementations, we are glad to have that conversation with no obligation attached.
FAQs
1. How long does a typical Salesforce Agentforce consulting engagement take?
Timelines depend on the use case, existing Salesforce setup, integrations, testing requirements, and approval process. A focused engagement can move faster than a broad rollout, especially when the team begins with one defined workflow and prepared stakeholders.
2. Can Agentforce be implemented without replacing our current Salesforce setup?
Yes. Agentforce can usually be introduced within an existing Salesforce environment rather than requiring a rebuild. Consultants first review current objects, automation, permissions, integrations, and custom logic to identify what can stay, what needs adjustment, and what should be isolated.
3. Can we start with Agentforce before committing to a company-wide rollout?
Yes. Many organizations begin with a limited proof of concept or departmental deployment before expanding. This approach lets teams validate usability, technical fit, user adoption, and operational requirements without making every business unit dependent on the first implementation.
4. Will an Agentforce implementation interrupt our existing Salesforce users?
A properly planned implementation should minimize disruption to daily Salesforce work. Changes can be developed and tested away from live users, then introduced through controlled deployment steps, user communication, access planning, and post-launch checks tailored to the affected teams.
5. Can Salesforce consultants work alongside our internal admin and development team?
Yes. A consulting partner can supplement an internal Salesforce team rather than replace it. The engagement can divide responsibilities across discovery, architecture, configuration, testing, documentation, deployment, and knowledge transfer while keeping internal administrators involved in decisions and long-term ownership.
6. Can Agentforce be customized for our company terminology and brand voice?
Agent behavior can be shaped around business terminology, approved instructions, knowledge sources, and communication standards. The exact customization depends on the agent type and channel, so teams should define examples of acceptable responses before configuration and user acceptance testing begin.
7. Can we create separate Agentforce agents for different departments?
Yes, separate agents can be designed around different jobs, audiences, permissions, and information sources. Sales, service, IT, or employee-facing use cases may require different instructions and access boundaries instead of forcing every department into one broad agent configuration.
8. Can Agentforce work with custom Salesforce objects and existing business processes?
Often, yes, but custom objects and processes should be reviewed before they become part of an agent workflow. Consultants typically check field definitions, automation dependencies, Apex behavior, permissions, and integration touchpoints to confirm that agent actions will behave predictably.
9. Do we need Data 360 before starting an Agentforce consulting project?
Not every discovery or planning engagement requires Data 360 from day one. Requirements depend on how the agent will retrieve, unify, or ground information. A consultant can determine whether your chosen use case needs Data 360 during solution design.
10. Which Salesforce licenses should we have before an Agentforce project starts?
Licensing varies by Salesforce edition, Agentforce product, user type, and the capabilities being deployed. Before implementation, confirm current entitlements with Salesforce and map licenses to the proposed users and channels so the technical design does not assume unavailable functionality.
11. Can Agentforce support users in multiple languages?
Agentforce provides language support, but availability and maturity can vary by agent type, feature, and channel. Multilingual projects should confirm supported languages early, then test real customer or employee phrases rather than assuming every language behaves identically in production.
12. Can a consultant take over an Agentforce project started by another partner?
Yes, but a takeover should begin with a structured review of the existing configuration, documentation, permissions, integrations, test results, and unresolved issues. That creates a reliable baseline before new changes are made and reduces the risk of repeating earlier mistakes.
13. What training does our team need after Agentforce is implemented?
Training should match each person’s responsibility. End users need practical guidance on when to use the agent and when to escalate, while administrators need deeper instruction on configuration, permissions, testing, troubleshooting, documentation, and controlled updates after launch.
14. Can Agentforce be deployed in phases across business units or regions?
Yes. Phased deployment can separate business units, regions, channels, or use cases so each group can be tested before wider expansion. This also makes it easier to adapt instructions, access rules, training, and support processes to local operating needs.
15. What should we prepare before speaking with a Salesforce Agentforce consultant?
Bring one or two business processes you want to improve, examples of current user problems, basic Salesforce architecture details, known integrations, relevant stakeholders, and any AI experiments. That gives the consultant enough context to make discovery specific rather than generic.