We earn a commission if you buy through links on this page. How we're funded.
Quick answer
The Salesforce AI Specialist certification, which Salesforce has renamed the Agentforce Specialist, tests whether you can configure generative AI features inside a Salesforce org — prompt templates, agent actions, grounding, and the Einstein Trust Layer. It is worth taking if you already administer Salesforce, and it is unsuitable as a first credential. For the general generative-AI grounding the exam assumes, Vanderbilt’s Prompt Engineering on Coursera covers prompting and trustworthy generative AI with no coding, though not Salesforce’s Prompt Builder.
See Prompt Engineering on Coursera →
This Salesforce AI Specialist certification guide covers the naming change you need to be aware of, the exam content, the platform prerequisites that actually matter, how to study in a free environment, and who should not attempt it.
What is the Salesforce AI Specialist certification?
The Salesforce AI Specialist certification is a mid-level credential for people who implement generative AI capabilities in Salesforce, rather than merely describe them. It assumes you can already navigate setup, build automation, and reason about the data model.
An important naming note: Salesforce has renamed this credential the Salesforce Certified Agentforce Specialist, and its old AI Specialist page on Trailhead now redirects to the Agentforce Specialist one. Study against the current Agentforce Specialist exam outline, because material written under the old name may describe a superseded outline — the most common mistake here.
The exam is multiple-choice, proctored, and scenario-heavy. Salesforce’s official exam guide, aligned to the Spring ’26 release, gives 60 scored questions plus up to five unscored ones, 105 minutes and a 72% pass mark, and states the registration and retake fees. Salesforce revises these along with the product, so check the guide before you book.
What does the exam cover?
The official exam guide splits the exam into six weighted areas:
- AI Agents — 35%
- Prompt Engineering — 20%
- Data 360 Fundamentals — 20% (the Agentforce Data Library, and chunking, indexing and retrievers in Data 360, the platform formerly called Data Cloud)
- Testing, Deployment, and Maintenance — 10%
- Governance and Observability — 10%
- Multi-Agent Orchestration — 5% (when a multi-agent design is the right choice, and the purpose of open protocols such as MCP and A2A)
In practice, content centers on configuring, grounding, securing, and testing generative AI features in an org. The recurring topics are:
- Prompt templates and Prompt Builder — template types, merge fields and related record data, grounding a prompt in Salesforce records, and choosing the right template for a sales, service, or marketing task.
- Agent configuration — assembling agents with topics and actions, mapping actions to Flows, Apex, or prompt templates, and controlling when an agent should hand off to a human.
- Grounding and data — supplying context from records, related lists, and unified data, plus why retrieval quality depends on how your data is structured and maintained.
- Einstein Trust Layer — data masking, zero data retention with model providers, toxicity detection, and audit trails, and which protections apply at which stage of a request.
- Model selection and configuration — the difference between Salesforce-managed models and bringing your own, and the trade-offs involved.
- Testing, monitoring, and governance — evaluating agent and prompt behavior before release, permission and access considerations, and monitoring output quality after deployment.
The Trust Layer questions are the ones candidates most often underestimate. You need to know what masking does and does not protect, and in what order the protections apply — conceptual familiarity is not enough.
What prerequisites do you actually need?
You need working Salesforce administrator skills. Salesforce sets no prerequisite certification for this exam; its guide recommends, but does not require, Platform Administrator and Platform App Builder, and describes a typical candidate as having about a year of platform configuration experience. The exam assumes fluency with objects and fields, record access and sharing, permission sets, and Flow.
Two further areas matter more than people expect. First, data model knowledge: grounding questions hinge on how records relate, because a prompt template can only retrieve what the model can reach. Second, familiarity with unified customer data concepts, since grounding often draws on data brought together from multiple systems.
General generative AI understanding is also assumed — prompts, context limits, hallucination risk. If you are new to both AI and Salesforce, start with Agentblazer Status on Trailhead, which Salesforce's own Agentforce Specialist page suggests as preparation, and an administrator credential before attempting this exam; the entry-level Salesforce AI Associate certification that used to fill this role was retired in February 2026.
Salesforce AI Associate is retired. The replacement is Agentblazer Status, a Trailhead programme recognising AI and Agentforce skills rather than a proctored exam.
How hard is the exam?
It is substantially harder than the retired AI Associate exam was, and sits in the mid-range of Salesforce credentials. The difficulty is practical: questions describe a business requirement and ask which configuration achieves it, which is difficult to answer from reading alone.
Three specific challenges recur. First, distinguishing between similar-looking options where only one respects record access or data protection requirements. Second, knowing which building block belongs in an agent action versus a prompt template versus a Flow. Third, keeping pace with terminology that Salesforce has renamed more than once, which makes older study material actively misleading.
Candidates who configure these features in their day job typically pass after focused review. Those studying purely theoretically usually need hands-on practice in an org first, because scenario questions reward having watched a configuration fail.
Not sure this is the right one for you?
Tell the picker about your background and what you want the certificate to do, and it narrows the list to the one or two courses we would start with. It suggests only our affiliate partners’ courses, and says so before it suggests anything.
Try the AI Certification Picker →How should you prepare for it?
Prepare with the current exam outline, Trailhead, and a free environment where you can build the features yourself.
- Download the current exam outline and confirm the credential name, since transition notices affect which material applies.
- Sign up for a free developer or trial environment with the relevant AI features enabled, so you can practice without touching production.
- Complete the Trailhead modules covering Prompt Builder, agent configuration, and the Einstein Trust Layer, which are free.
- Build one prompt template that grounds on related record data, then deliberately break the grounding by removing access and observe what happens.
- Configure one agent with at least two actions, including a handoff to a human, and test it against realistic requests.
- Review the Trust Layer protections in order, writing down what each one prevents, then check your notes against the documentation.
- Cover the two newer areas: set up a Data Library and look at how Data 360 chunks and indexes content for retrieval, then read Salesforce’s material on multi-agent designs and what MCP and A2A are for.
- Take any available practice questions and trace each miss back to the feature in your own org rather than memorizing answers.
Step four is the highest-value exercise. Understanding why a prompt returned nothing teaches record access and grounding faster than any module.
The exam also assumes general generative-AI understanding — prompts, context limits, hallucination risk — and for that the course we would use is Vanderbilt’s Prompt Engineering on Coursera: three courses, on prompt engineering for ChatGPT, ChatGPT’s advanced data analysis and trustworthy generative AI, about 39 hours with no coding. It teaches prompting in ChatGPT, not in Prompt Builder, and nothing about agent actions or the Einstein Trust Layer, so Trailhead and your own org remain the exam preparation; it is a paid Coursera certificate of completion, not a Salesforce credential.
Who should take it, and who should skip it?
Good fit
- Salesforce administrators and consultants who are implementing or about to implement AI features for real users.
- Solution architects and technical leads designing agent workflows and needing credible platform authority.
- Partner-firm staff, where specialist certifications support partner tier requirements and client trust.
- Marketing and service operations specialists who configure automation themselves; broader marketing options appear in our ranking of AI certifications for marketers.
Skip it if
- You do not administer Salesforce. Almost every question is platform-specific configuration.
- You want transferable generative AI engineering skills such as retrieval pipelines, evaluation, and agent frameworks — see our best generative AI certifications roundup instead.
- You are new to AI concepts entirely; Agentblazer Status on Trailhead is the correct starting point.
- Your org has no plans to adopt these features, since the knowledge decays quickly without use.
How does it compare with other options?
The table below compares 5 credentials on level, best for and main trade-off.
| Credential | Level | Best for | Main trade-off |
|---|---|---|---|
| Salesforce AI Specialist / Agentforce Specialist | Mid-level configuration | Implementing generative AI features inside Salesforce orgs | Highly platform-specific; naming and content churn |
| Agentblazer Status (Trailhead) | Entry awareness | AI and Agentforce literacy | A status, not a certification |
| Salesforce Certified Administrator | Core platform | The foundation this exam assumes | No AI content; separate study effort |
| Databricks Certified Generative AI Engineer Associate | Engineering | Building RAG applications with code | Different ecosystem; requires Python |
| Vendor-neutral generative AI coursework | Learning | Transferable LLM application skills | No badge recognized inside Salesforce hiring |
Our AI certification comparison places platform credentials against vendor-neutral routes such as Coursera programs, which matters if you may leave the ecosystem later.
Does it require maintenance?
Yes. The exam guide says you must complete the Agentforce Specialist maintenance module on Trailhead once a year, and the certification expires if you miss the due date. For AI credentials specifically, maintenance also serves a real purpose, because the underlying features change frequently.
Verify the current maintenance cadence and requirements on the Salesforce certification site. Hold this credential only while you work with the features, since upkeep across multiple Salesforce certifications compounds into meaningful annual effort.
Is the Salesforce AI Specialist certification worth it?
It is worth it for practitioners implementing Salesforce AI features, and not worth it for anyone else. Inside the ecosystem it is one of the more meaningful AI credentials available, because it tests configuration judgment rather than vocabulary, and because Salesforce hiring genuinely rewards certifications.
The preparation also produces good habits: grounding prompts properly, respecting record access, and testing agent behavior before release are exactly the practices that keep AI features from embarrassing an organization.
The case against is concentrated risk. The credential is tied to one vendor’s rapidly changing product line, requires maintenance, and has already been renamed once. Our analysis of whether AI certifications are worth it covers how to weigh that against vendor-neutral study.
Ready to start?
Sold on Coursera at the price for your country, on its own or as part of Coursera Plus, whose pricing page compares the plans.
Frequently asked questions
Is the AI Specialist certification the same as Agentforce Specialist?
Yes. Salesforce renamed the AI Specialist the Agentforce Specialist, and the old AI Specialist page on Trailhead now redirects to the Agentforce Specialist one. Treating them as two separate exams, or assuming a guide written under the old name describes the current outline, is where people go wrong.
Transitions of this kind change the exam outline, not just the branding. Features get added as the product line grows, and weightings shift with them, so an outline that was accurate six months ago may no longer describe what you will sit.
Verify the current credential name and any retirement notices on the official Salesforce certification page before you start studying. Preparing against a superseded outline is the most common failure here, and it is entirely avoidable — it costs five minutes at the start and weeks if you skip it.
Do I need the Administrator certification first?
Not required, but strongly advisable. Salesforce's exam guide sets no prerequisite certification and recommends Platform Administrator and Platform App Builder; the gap they fill is larger than “recommended” suggests.
The exam assumes fluency with objects, fields, record access, permission sets and Flow. Grounding questions in particular depend on understanding the data model — if you cannot picture how the records relate, you cannot reason about what an agent will retrieve or what a user will be permitted to see.
The observable pattern is candidates who understand generative AI concepts well and still fail on configuration scenarios, because the AI half was never the hard part. Build the platform foundation first. If you are choosing between another week on prompt design and another week on Flow and sharing rules, take the Flow.
How hard is the Salesforce AI Specialist exam?
Harder than the retired AI Associate was, and mid-range among Salesforce credentials — a real exam rather than a formality, but not among the platform's most demanding.
The difficulty is concentrated in a specific place: scenario questions where several configurations look plausible and only one respects access controls or data protection requirements. You are rarely asked whether something is possible; you are asked which of several possible things is correct given who is allowed to see what.
Hands-on practice in a free developer org is the decisive preparation factor. The exam rewards having configured these features and broken them yourself, because the wrong answers are usually things that work fine in a demo and fail the moment sharing rules apply. Reading about it does not produce that instinct.
What study resources are available?
Trailhead provides free modules covering Prompt Builder, agent configuration and the Einstein Trust Layer, and Salesforce publishes the exam outline free. Between them that is a complete preparation path at no cost.
Add a free developer or trial environment and practise hands-on, because this exam specifically rewards having configured the features rather than having read about them. That environment is the single most valuable free resource here, more so than any module.
Paid instructor-led courses exist but are rarely necessary for a candidate who already has administrator-level platform knowledge. Confirm current resource availability and fees on Salesforce's official training and certification pages before planning around them — this credential sits on a fast-moving product line and both the materials and the pricing change.
Does the certification expire?
Only if you skip maintenance. Salesforce's exam guide says you must complete the Agentforce Specialist maintenance module on Trailhead once a year, and the certification expires if you miss the due date.
Here the maintenance is genuinely substantive rather than a formality, which is unusual. These AI features change quickly enough that a specialist credential earned two cycles ago describes a product that has meaningfully moved — so the refresh modules are doing real work rather than ticking a box.
Check the current cadence on the Salesforce certification site, since the process has changed before. And keep the credential only while you actively work with the features: maintaining a specialist certification for a product you no longer touch is effort spent on a signal you cannot back up in an interview.
Will the Salesforce AI Specialist credential help me get hired?
Within Salesforce hiring, meaningfully — especially for consultants and partner firms, where specialist credentials influence staffing decisions and client confidence directly. Partners often need certified people on paper to win or staff work, which gives the credential commercial value beyond what it proves about you.
Outside that ecosystem it carries little weight, and it is worth being realistic about that. The material is product-specific by design, so a hiring manager with no Salesforce estate has nothing to map it onto.
The strongest combination is this certification plus a described implementation: what you configured, what you grounded it on, and how you tested it before release. That last part — the testing — is what separates someone who has shipped an agent from someone who has passed an exam about them, and interviewers in this space know to ask.
Should I learn vendor-neutral generative AI skills instead?
If there is any chance you will leave the Salesforce ecosystem, yes — at least alongside, rather than instead.
The reason is portability. Retrieval, embeddings, evaluation and agent design transfer to any employer and any stack, because they are ideas about how these systems work. Prompt Builder configuration does not transfer at all; it is knowledge of one vendor's interface to those ideas.
The pragmatic approach for a working Salesforce professional is to hold both: this credential for current employability, and a vendor-neutral course such as one from DeepLearning.AI for durable skill. They serve different time horizons rather than competing — one makes you staffable this quarter, the other keeps you employable after a platform change you do not control.