A manager trying to cover a vacant shift or revive an unanswered job application does not need another dashboard showing that work has stalled. The immediate need is someone—or something—to complete the follow-up.

Fountain, a 12-year-old San Francisco–based software company that helps employers hire and manage hourly frontline workers, is making that the premise of Orchestra, an AI agent builder for firms with large, hourly workforces.

Teams describe an assignment or business objective, approve the agent's proposed design, and let it carry out recurring recruiting and staffing tasks using information already in Fountain.

Orchestra is a proprietary service available at launch at no charge to selected existing customers inside their Fountain accounts.

Fountain, whose broader platform serves more than 500 frontline employers, says general availability is planned for early 2027. At that point, customers will pay for the work agents do and be able to limit how much work they take on. Prospective buyers can contact the company through Fountain's website.

In a November 9 post about Fountain's pricing model, chief product and technology officer Salim Jernite said credits cost 50 cents, with charges tied to completed outcomes and no charge for failed attempts.

He also described budgets for individual agents and accounts, advance cost estimates for large or unusual runs, and automatic pauses when spending looks unusual. The launch release gives no Orchestra rate card, leaving buyers without a confirmed price for outcomes such as a filled shift.

Jernite said in an exclusive interview with VentureBeat that about six customers were using Orchestra when he spoke ahead of launch.

From a requested answer to an ongoing assignment

Orchestra builds on Cue, the AI tool Fountain launched in April 2026. Fountain says customers have asked Cue more than 80,000 questions since launch. With Orchestra, customers can set up agents to handle recurring work or act when an event occurs.

Agents can reconnect with past applicants, prepare new hires for their first day or respond when a position opens. Teams can run an agent on demand or on a schedule, or have it start when a specific event occurs.

Jernite described how the shift coverage example would work. A worker dropping a shift could trigger an agent to check the roster, factor in overtime and contact preferences, and reach out to eligible workers.

“It’s going to call them until it reaches one that actually accepts the shift,” he told VentureBeat. The agent would then assign the shift and text the manager.

The setup process includes a human approval step. Jernite described three ways to start. A user can specify what an agent should do, describe a problem such as applicants stalling at a particular stage, or ask Orchestra to review operations and suggest where agents could help.

In the second case, the system might propose texting or calling an applicant after two days of inactivity. Orchestra presents a design for review before creating it, and Jernite said it can show customers how a proposed agent performs in simulations. Agents can serve a team without being assigned separately to every employee.

Fountain already offers preconfigured agents for tasks such as interviewing candidates and surveying employees. Jernite said Orchestra grew out of customers’ need to adapt automation to their own systems and processes. He expects corporate recruiting and operations teams to build agents that handle work for managers in the field. He pointed to the staff who coordinate hiring, onboarding and shift coverage. “This is the layer that we’re optimizing with agents,” he said.

The distinction matters for organizations deciding whether to purchase a workforce application or build similar automation themselves. A general agent can generate messages and call tools, but it still needs access to the right employee records, application status, scheduling information and permissions. Fountain's proposition is that much of that operational context already resides in its product.

How Fountain builds agents and connects outside tools

Fountain built the software that manages agent instructions, tools and memory, along with the records it keeps for messaging and consent, Jernite said.

The system primarily uses Anthropic’s Claude models. Jernite said Orchestra uses Opus 5.5 to design agents, most of the resulting agents run on Sonnet 5.5, and smaller agents can use Haiku for jobs such as reading logs and maintaining memory.

“We adapt the model depending on the complexity of the tasks,” he said. Customer-built agents can also assign work to other agents so they can carry out several actions in parallel.

The builder can suggest connecting outside tools through the Model Context Protocol, or MCP, Jernite said. “You could create agents anywhere else. You can actually bring them to Orchestra,” he said.

He described a Microsoft 365 integration that lets Orchestra call Microsoft Copilot agents to handle parts of an assignment. Orchestra itself runs within Fountain, and Jernite’s account of importing agents left open whether customers can swap out the models that Fountain uses to build and run its own agents.

Checking whether the agent actually finished

Fountain says the agent sets a plan and an objective each time it runs. A separate check reviews the actions the agent took against that objective before marking the work complete. When the work falls short, the system flags it for a person to review and explains why.

“Knowing when it hasn’t completed the job and why is the hard part,” Jernite said in the announcement.

“Each agent that is created on Orchestra has a goal and is checked against that goal,” Jernite told VentureBeat. He said the goal specifies a successful outcome and sets limits on what the agent may do to achieve it. A separate AI agent checks whether the goal was met in full or in part and can recommend up to three reruns.

The check is a significant part of the product's pitch. An agent that reports success after sending a message has not necessarily secured coverage for an open shift. Evaluating completed work requires a definition of success tied to the underlying business record, not merely the agent's own explanation.

Fountain says it uses more than 180 agents internally across functions including engineering, finance and customer-facing operations. The launch release provides no independent evaluation or customer performance figures for Orchestra.

Permissions and outreach controls are part of the design

According to Fountain, agents inherit the permissions of the person running them. On each call to an applicant or worker, the system checks consent, allowed calling hours and daily contact limits. It identifies itself as AI and honors opt-outs.

The company also says outreach follows predefined objective criteria, discriminatory instructions are refused, and people retain hiring decisions. Agent activity is recorded in readable language.

Those are product claims, not proof that every configured workflow will behave correctly. A buyer evaluating Orchestra would still need to test whether the permission boundary holds in practice, how failed checks reach a person and what information remains in an audit record.

Workday and Microsoft offer different routes to similar automation

Orchestra sits between specialized recruiting automation and general-purpose agent-building software.

Workday's Paradox candidate-experience offering already supports conversational applications, candidate questions and interview scheduling, with connections to applicant-tracking systems. It is a direct comparison for employers focused on recruiting throughput. Workday's public product page routes buyers to sales rather than providing a dollar rate card.

Microsoft Copilot Studio offers a broader toolkit for building agents across business processes. Its published pricing includes a $200 monthly capacity pack with 25,000 Copilot Credits, alongside other purchasing options. Those credits measure Microsoft's service consumption, while Fountain proposes charging for completed work.

Offering

Main focus

Access and licensing

Pricing information

Comparison with Orchestra

Fountain Orchestra

Customer-built agents for frontline recruiting and workforce tasks

Proprietary service; selected Fountain customers now, broader rollout planned for early 2027

Initial access free for selected customers; later work-based billing. Final task rates unspecified

Built around Fountain records, permissions and completion checks

Workday Paradox

Candidate engagement and recruiting automation

Commercial product; sales/demo route

No dollar price listed on the cited product page

Direct overlap in candidate communications and interview coordination

Microsoft Copilot Studio

Agents for a wide range of workflows

Commercial platform; online access and purchasing

$200 monthly pack for 25,000 credits; other billing options available

Broader flexibility, with workforce-specific integration and evaluation left to the implementation

The choice depends partly on where the employer's operating data lives. A company already using Fountain may reduce integration work by building within that environment. A business standardized on Workday may prefer recruiting automation connected to its existing hiring system. A team pursuing several unrelated workflows may find a general builder more useful, while taking responsibility for connecting and validating each one.

The vendors define credits differently, and the available materials offer no comparable performance figures. For employers in the free rollout, shift coverage offers a concrete test before billing starts. They can track how often agents fill dropped shifts, how much work managers still do and whether the activity records explain what happened. Once Fountain publishes its rates, buyers can use that data to estimate the cost of each successful fill.