ChatGPT
Claude
Google AI
Copilot
The headline looked simple enough. An AI hiring startup announced that its whole team was joining Salesforce, and trade press wrote it up as an acquisition. A day later the story changed. Salesforce's own position, reported by TechCrunch, was blunt: "A Salesforce spokesperson told TechCrunch that Moonhub has not, in fact, been acquired (by the company’s definition of the term). Rather, Moonhub is shutting down, and only some of the company’s team, not the whole team, as MoonHub implied in its post about the news, will be joining Salesforce." So the first thing to say to anyone asking for a Moonhub partner is that the product is not there to be implemented, and the company is not a Salesforce brand you can license. The question worth answering is the one underneath it: who do you hire if you want AI to help with recruiting on Salesforce?
That answer depends less on Moonhub than on what kind of organisation you are. This page is organised by buyer type for that reason. The history sits on What happened to Moonhub, and the design question on Moonhub, AI recruiting, and what Salesforce did not buy.
Three buyers, three different hires
Salesforce's current recruiting offer is not a Moonhub rebrand. It is Agentforce applied to the hiring process, described on the company's recruiting use-case page as four subagents. A Candidate Engagement Subagent handles applicant questions and pre-screening, a Talent Matching Subagent compares candidate skills with role requirements, an Interview Coordination Subagent deals with scheduling, and a Recruiter Insights Subagent summarises interactions for recruiters. The page is clear about the plumbing too: "Agentforce for recruiting is designed to connect with your existing HR and talent systems through MuleSoft." The same page adds that Data 360 supplies the unified data. In other words, you are hiring for an agent build on top of systems you already own.
Which hire follows depends on who you are.
| Buyer | What the Salesforce material points to | The hire to look for |
|---|---|---|
| In-house talent team with an applicant tracking system | Agentforce recruiting subagents sitting on top of your HR and talent systems via MuleSoft | An Agentforce builder who has integrated an applicant tracking system and can show how candidate records reach Data 360 |
| Staffing or recruiting firm | Salesforce's CRM for recruiters, built on Agentforce, Sales Cloud, Service Cloud, Data 360 and Slack, with Starter Suite named for smaller firms | A Sales Cloud and Service Cloud consultant who understands candidate and client as two audiences in one org |
| Public sector employer | A separate recruitment skill for government hiring | A partner with public sector delivery experience and a view on procurement and accessibility |
| Anyone still assuming Moonhub exists | Nothing to buy | A conversation about requirements before a partner conversation |
Pick the row first. A partner who cannot tell you which row you are in, or who answers every row with the same slide, is selling a generic Agentforce engagement.
Skills and credentials, in the order they matter
Start with the Salesforce Certified Agentforce Specialist credential, because every one of the four subagents is an Agentforce build and the partner needs a working grasp of agents, subagents and guardrails. Next comes the Salesforce Certified Data 360 Consultant credential, since matching and insights draw on unified candidate data, and that data has to be modelled and unified before an agent can read it. For the link to the applicant tracking system, look for the Salesforce Certified Platform Integration Architect credential or a MuleSoft reference. A staffing firm should add the Salesforce Certified Agentforce Sales Consultant credential, because the sales and candidate sides share one org. And because candidate files are sensitive, the Salesforce Certified Platform Sharing and Visibility Architect credential, or a convincing design walk-through, is worth asking about.
None of those proves anyone has built recruiting automation. What does is a demonstration of a screened-candidate record with the source data visible, the reason for the match recorded, and the point at which a person took over.
What a realistic first engagement looks like
For an in-house team, begin with one subagent rather than all four. Interview scheduling is the safest first target, because the output is a calendar entry and not a judgement about a person. It touches the least sensitive data, the benefit is visible to hiring managers within weeks, and it forces the integration with your applicant tracking system, which every later subagent will need. A reasonable first scope is one job family, one region, one applicant tracking system, and one sentence on who may see what.
Candidate Engagement comes second. It answers questions and conducts pre-screening, which means it asks applicants things and records what they say. Decide before building what a pre-screen may ask, what it may never ask, and how an applicant reaches a person. Talent Matching comes last, deliberately, because it is the one that ranks people.
For a staffing firm, the first engagement is usually the data model rather than any agent: candidate profiles, client accounts, jobs and submissions in a form that both Sales Cloud and an agent can rely on. Agentforce inherits whatever mess lives in the org.
What each subagent design must specify
Whatever the buyer type, ask the partner to fill in one row per subagent before any build starts. The table below is a design prompt of our own, not Salesforce documentation, and it is deliberately plain. A partner who needs a workshop to complete it has not built one before.
| Subagent | Data it should read | What it may write | Human checkpoint |
|---|---|---|---|
| Candidate Engagement | Job postings, an approved question list, the applicant's own replies | The applicant's answers and a status on the application | A recruiter reviews every pre-screen before an applicant is declined |
| Talent Matching | Role requirements and the skills data you have agreed to use | A match note that cites the evidence used | A named person confirms before an applicant is rejected on the match |
| Interview Coordination | Interviewer availability and the applicant's contact details | Calendar entries and confirmation messages | Recruiter can cancel or reschedule at any point |
| Recruiter Insights | Interaction history already held in the org | A summary on the candidate record | Recruiter can edit or delete the summary |
Notice what the table forces into the open. Talent Matching is the only row where the agent's output influences who proceeds, and it is also the only row that needs an explanation attached to each result. If your partner proposes to build all four rows in a single release, ask them why.
If you already signed with someone who named Moonhub
Some buyers reach this page after a proposal has arrived. If a partner has already put Moonhub into a statement of work, do not argue about the press coverage. Ask for the clause to be replaced by a deliverable you can test, such as "the Interview Coordination Subagent schedules a first-round interview for one job family in one region, with a recruiter able to override." That wording survives any naming change at Salesforce, and it exposes a partner who cannot write it. Ask also whether any fee was based on a Moonhub integration, because there is no such integration to build, and a fee for it should be withdrawn.
Staffing firms deserve a separate note. Their candidates are also their product, which means the agent is touching revenue as well as risk. Make sure the partner understands your submission process, how a candidate may be represented to more than one client, and what your consent wording allows you to do with a profile. An in-house team can afford a slow first month; a staffing firm usually cannot afford a wrong submission.
Why screening needs more care than most agent work
Automated screening sits in a category regulators watch. The EU's AI Act, in its list of high-risk uses, includes "AI systems intended to be used for the recruitment or selection of natural persons, in particular to place targeted job advertisements, to analyse and filter job applications, and to evaluate candidates;" That is European law, and it may not govern your hiring. It is nonetheless a good statement of what scrutiny looks like, and it tells you how a cautious partner should behave. Whatever your jurisdiction, ask for the same four things before any agent filters or ranks applicants: a named human who owns each decision to reject, a log of what the agent saw and recommended, a way for a candidate to reach a person, and a test with realistic applicants that looks for systematic differences between groups.
This is also where a Moonhub reference misleads. A vendor that was a startup selling candidate evaluation tools is not evidence that your use is lawful or fair, and Salesforce's decision not to acquire it says nothing either way.
Red flags specific to this product
A statement of work that includes Moonhub as a licence or a connector is the first. A partner who presents candidate scoring as a configuration step, with no mention of review or bias testing, is the second. Be careful with a partner who wants every candidate record loaded into Data 360 on day one when scheduling would work from a narrower slice. Be careful, too, with anyone who treats Slack as a place to paste candidate details because the recruiter hero image on the Salesforce page shows Slack; the screenshot is a use case, not a data-handling policy. Finally, a fee tied to time-to-hire reductions invites the agent to optimise speed at the expense of fairness.
Ten questions for the first meeting
- Which of the four recruiting subagents will you build first, and why that one?
- How will candidate data from our applicant tracking system reach Data 360, and who maintains the MuleSoft connection?
- What may the pre-screen ask, what may it never ask, and who approves the list?
- Who is the named person accountable when an applicant is rejected after an agent has seen their file?
- How does a candidate reach a human, and how fast is that route tested?
- What do you log for each match, and can a recruiter see the reason?
- How will you test for different outcomes across groups before go-live, and what will you do if you find them?
- How is candidate data separated from client data in our sharing model?
- How do multiple languages work for our regions, and who checks the translations?
- Which parts of this plan would you refuse to automate?
Languages and regions
Salesforce's recruiting page says candidate engagement can run in several languages, which is attractive for a business hiring across regions and risky when nobody on your side can read the replies. If you intend to use more than one language, ask the partner who will review the questions and answers in each one, how an applicant is told they are talking to an agent, and what happens when the agent cannot understand a message. Treat each language as its own test, not a translation of the first. A pre-screen question that is neutral in English can carry assumptions in another language, and the only way to find out is to have a native-speaking recruiter read real transcripts during the pilot.
Checking references and evidence
Ask for a reference where the partner connected an applicant tracking system, not only built an agent in a sandbox. Ask that customer what the agent was not allowed to do. Ask how many candidates a recruiter had to correct in the first month, what the correction route looked like, and what changed in the design as a result.
Ask the partner to show their own test plan for an agent with personal data. If they have never written one, they have never shipped one. A logo slide cannot answer any of these questions, and a firm that says it has deployed AI hiring across industries without naming a single constraint has not met a real one.
Commercial models and the statement of work
Run a short discovery first, ending in a written scope for one subagent, a data map and a decision on review. Then buy the build as fixed scope for that single subagent, with options for the next. The statement of work should name the subagent, the job family and region, the systems connected, the Data 360 objects involved, the human review points, the log, the test with realistic applicants, and an acceptance condition that a person can override any recommendation. Leave out any promised reduction in time-to-hire or cost per hire. Licence questions are on Moonhub Salesforce licence cost, where Moonhub itself has no licence to price, and the project view is on Moonhub Salesforce implementation cost.
When you can skip a partner
If all you need is interview scheduling for a single team and your applicant tracking system already offers a Salesforce connection, an administrator with Agentforce Specialist training may manage it. If you are a very small staffing business, Salesforce's Starter Suite is positioned for you, and a template-led setup may suffice. Anything that ranks or filters applicants, anything that spans several systems, and anything touching a public sector procurement process deserves a partner with a record. Before choosing, read Should I migrate from Moonhub and Integrating Moonhub with Salesforce, and see Salesforce's recruiting page for the current four-subagent description.
