From answering questions to ongoing tasks
Fountain is a San Francisco-based company that has been operating for 12 years. Its platform helps employers hire and coordinate hourly workers, and is used by more than 500 employers. Orchestra builds on Cue, a tool launched in April 2026. Fountain says customers have asked Cue more than 80,000 questions since its launch.
In Orchestra, companies describe a task or business goal, review the proposed agent setup, and authorize the agent to handle recurring work or respond to events. An agent can contact former applicants, prepare new hires for their first day, or start working when a vacancy appears. It can be launched manually, on a schedule, or when a specified event occurs.
For example, if an employee declines a shift, an agent checks a list of workers, taking overtime limits and contact preferences into account, then reaches out to those who could be offered the shift. According to Salim Jernite, Fountain’s chief product and technology officer, the agent will call until someone accepts, then assign that employee to the shift and notify the manager.
A person must approve the setup. There are three ways to get started:
If an applicant is taking a long time to respond, the system might suggest sending them a message or calling after two days of inactivity. Before an agent is created, Orchestra displays its plan for review; Jernite says the system can also show simulated results. Created agents can help the entire team and do not need to be assigned to each employee individually.
Fountain already offers ready-made agents, including ones for conducting interviews and employee surveys. Orchestra was developed in response to customers who wanted to adapt automation to their own systems and processes. Jernite expects recruiting and operations teams to create agents for local managers, including agents for coordinating hiring, onboarding new employees, and covering shifts.
The value of this approach depends on where work-related data is stored. A general-purpose agent needs more than the ability to write messages and call tools: it needs access to employee records, application statuses, schedules, and the necessary permissions. Fountain is counting on much of that information already being in its product.
Models, tools, and integrations
Jernite says Fountain built its own software to manage agents’ instructions, tools, and memory, as well as to store records of messages and user consent.
The system mainly uses Anthropic’s Claude models. Opus 5.5 is used to design agents, most agents created run on Sonnet 5.5, and agents running Haiku can handle smaller tasks such as reading logs and maintaining memory. The model is selected according to the task’s complexity. Customer-created agents can also delegate work to other agents and perform several actions in parallel.
Orchestra may suggest connecting external tools through the Model Context Protocol (MCP). Jernite also described an integration with Microsoft 365, through which Orchestra can call Microsoft Copilot agents to handle parts of a task. Orchestra itself runs inside Fountain. Jernite’s description of connecting external agents does not make clear whether customers will be able to replace the models Fountain uses to build and run its own agents.
How results are checked
Fountain says that before each run, an agent sets out a plan and a goal. Once the task is complete, a separate check compares the agent’s actions with the goal. If the work is incomplete, the system hands the task to a person and explains why.
According to Jernite, the goal defines the expected result and limits on the agent’s actions. Another AI agent checks whether the goal was achieved fully or partially and can suggest up to three retries.
This check is meant to distinguish an actual result from a report of actions taken. Sending a message does not, by itself, mean an open shift has been filled: successful completion needs to be tied to a record of the result in the work system.
Fountain says it uses more than 180 agents internally across engineering, finance, and customer-facing teams. But the launch materials include no independent assessment of Orchestra or performance metrics from customers.
Permissions and communication rules
Fountain says agents inherit the permissions of the user who launches them. Before each call to an applicant or employee, the system checks consent to be contacted, permitted calling hours, and daily limits on the number of outreach attempts. The agent identifies itself as AI and respects requests for no further contact.
The company also says that outreach follows predefined criteria, discriminatory instructions are rejected, and hiring decisions remain with people. Agent actions are recorded in plain language.
These claims describe how the product is designed, but do not establish that every configured workflow will run without errors. Employers evaluating Orchestra will need to check whether the restrictions are followed in practice, how failed checks are handed off to a person, and what information appears in the action log.
Comparison with Workday and Microsoft
Orchestra sits between specialized hiring automation and general-purpose agent-building tools.
Workday Paradox focuses on candidate interactions: it supports messaging, answers to questions, and interview scheduling, and connects to applicant tracking systems. It is a direct option for employers looking to speed up hiring. Workday’s public page does not list prices in dollars; it directs visitors to contact sales.
Microsoft Copilot Studio is designed for building agents across different business processes. One published option is a $200-a-month capacity pack with 25,000 Copilot Credits; other payment options are available. These credits measure use of Microsoft’s service, while Fountain plans to charge for completed work.
| Platform | Purpose and availability | Price and features |
|---|---|---|
| Orchestra | Agents for hiring and hourly-worker tasks; currently available to select Fountain customers, with a broader launch planned for early 2027 | Free at launch for select customers, with payment for completed work later. Rates for specific tasks have not been disclosed. Uses Fountain data and permissions, with completion checks |
| Workday Paradox | Candidate communications and hiring automation | No price in dollars is published on the cited page. Supports candidate messaging and interview coordination |
| Microsoft Copilot Studio | Builds agents for different business processes; available online | The $200-a-month pack includes 25,000 credits; other payment options are available. Broad flexibility, but connections to HR systems and checks on results must be set up separately |
The choice depends on where a company’s core work information lives. Fountain users may find it easier to build agents in an environment that is already connected to their data; an organization using Workday may prefer automation built into its hiring system. A general-purpose tool may suit a team with varied needs, but the team will have to connect data sources and check agent performance itself.
The published materials do not allow a comparison of platform performance: there are no comparable metrics, and the systems measure credits and completed work differently. During the free launch, employers can test how often agents fill shifts that open up, how much work remains for managers, and whether the logs help explain what happened. Once prices are published, this data will make it possible to estimate the cost of successfully filling a shift.
In Dataist’s view, built-in access to workforce data and permissions could make Orchestra more convenient than a general-purpose builder for companies already using Fountain. But until the company publishes pricing and performance metrics, its economic value cannot be assessed. The free launch gives customers a chance to test the results before paying.
Daily AI news
Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.
Only what matters — every day
Follow on X