Named AI Agents Are Now Job Postings, Not Feature Flags

Two things landed within 24 hours this week that are easier to understand together than apart. On September 10, OpenAI released the Agents API in public beta, giving developers managed access to the agent harness and execution infrastructure used by Codex. The next day, Salesforce introduced seven named Agentforce AI agents - Casey, Paige, Carter, Hunter, Marshall, Piper, and Fin - each built for a specific business function in sales, service, commerce, IT/HR, supply chain, and customer experience.

One gives you plumbing. The other gives you personnel. The interesting thing is that both moves are pointing at the same conclusion: the agent era is no longer about capability. It is about accountability.

What Salesforce's named agents actually signal

Salesforce put seven named AI agents to work across its Agentforce platform, each built around a specific job rather than sold as a general-purpose assistant customers configure themselves. That framing is deliberate. The product move is serious: Salesforce wants buyers thinking about work, not menus. A named agent with a job description is easier to place in a budget meeting than a feature list with a dozen screenshots.

The roster breaks down like this:

Each agent ships with the skills, actions, and data models required for its job and can be tailored to how each company works, including giving the agent its own name.

The customer numbers Salesforce cited are specific enough to take seriously: Autism Queensland reports that Paige resolves 70% of administrative requests, and Hibbett says its Hibbett AI handles 90% of core shopper journeys after going live in six weeks. Asana's Piper agent is driving four times the conversation volume, with customers deploying Piper in 45 days on average.

They are Salesforce-reported customer results, not independent proof. Read them as signals, not guarantees.

The company reported 7 billion Agentic Work Units delivered across Agentforce and Slack, with 3.2 billion of those in the second quarter alone.

The part most coverage skipped: alongside the named agents, Salesforce introduced the Trusted Enterprise AI Harness, a six-pillar governance framework, and an AI Control Plane for managing agents across an entire organization, including ones not built on Salesforce. That last clause matters. An AI Control Plane that registers and monitors third-party agents - not just Agentforce ones - is infrastructure, not a sales deck. For enterprise buyers, the announcement signals that AI agent governance - not agent capability - is becoming the next competitive battleground.

What OpenAI's Agents API actually costs

The OpenAI Agents API public beta opened on September 10, 2026, and gives every developer the managed Codex harness as a plain API: OpenAI runs the agent loop on its own infrastructure - coordinating model calls, tool use, and context - while you supply the tools, pick the execution environment, and pay only for the tokens and tools your agents consume, with no additional fee for the API itself.

The "no additional fee" headline is accurate. But it is worth reading the fine print.

Containers are priced at $0.03, $0.12, $0.48, and $1.92 per 20-minute session per container, with eligible container sessions billed by the minute, with a 5-minute minimum per session. If you run a research agent on the flagship gpt-6-astra model, Astra charges $12.50 per million tokens for cache writes, which the Agents API usage fields "do not expose," so its real cost per call can be higher than the list rate implies.

Long-running agents can generate many model turns and tool calls, so per-session budgeting should be based on complete task traces instead of the price of a single request.

There are two defaults worth catching before you ship anything:

- Outbound network access is on unless you switch it off, although OpenAI's own report example sets it to "disabled."

- Subagents default to six at once, while the launch example sets three and the overview sets four.

Neither of those is a reason to avoid the API. They are reasons to read the docs before your agents touch production data. Data stays US-only, and Zero Data Retention is unsupported during the public beta - relevant for any team with EU data residency requirements.

The timing of the Assistants API sunset is worth noting: the Assistants API, OpenAI's original stateful agent primitive, sunset on August 26, 2026, roughly two weeks before the Agents API opened to all developers. The message is clear. OpenAI considers the Codex harness the canonical way to build stateful agents going forward.

The thing both launches have in common

Strip the product names and the same architectural choice appears in both announcements: the Agents API is designed for workflows in which an agent does more than produce a single isolated model response - it works for longer periods, uses tools, and can coordinate multiple helper agents. Salesforce's Hunter operates on a runtime designed to preserve context and continue a goal across days or weeks using memory, durable execution, and dynamic steering.

Longer-lived agents doing real business work changes what "oversight" means. A chatbot you review after the fact is manageable. An agent that has been working your CRM pipeline or your IT ticket queue for three days while you were in Zurich is a different relationship entirely.

The real signal from this week is: stop chasing more agents and start using a few well-defined ones with tight limits, human review, and clear proof. Both launches - one a platform API, one a named roster - reinforce that framing. The question is no longer whether agents can do the work. It is whether your team has the controls to know what they did.

Named AI agents in the enterprise: common questions

What is a named AI agent?

A named AI agent is a purpose-built software agent assigned a specific job title and pre-configured with the skills, data access, and guardrails that job requires. Rather than a blank model you configure from scratch, it ships ready for a defined role - customer service, IT support, sales qualification - and can be tailored to your company's processes.

How does the OpenAI Agents API differ from the old Assistants API?

Instead of stitching together the Responses API, a custom orchestration loop, and a sandbox provider, you now call a single managed endpoint that runs the same Codex harness OpenAI uses internally. Sessions, context compaction, multi-step recovery, and subagent delegation are handled server-side. The Assistants API sunset on August 26, 2026.

Does the OpenAI Agents API cost extra?

OpenAI says there is no separate Agents API fee: usage is billed through the models and tools used by each session. The cost that surprises teams is the hosted sandbox, which bills at $0.03-$1.92 per 20-minute session depending on memory tier - and there is currently no memory limit field on the Create Session object.

What is Agentforce Coworker?

Agentforce Coworker gives employees an agent that works alongside them across the surfaces where they work, grounded in the context of their business. With AI Skills, employees can teach Coworker how to complete a task once and then scale that know-how across the workforce and across interfaces; AI Skills reaches GA in October 2026.

Why does the Salesforce AI Control Plane matter?

Most teams deploying agents end up with agents from multiple vendors - Agentforce, OpenAI, internal builds - with no single place to see what they are doing. The AI Control Plane is designed to register and monitor agents regardless of where they were built. For any enterprise already running five or more agents across tools, that governance surface is more immediately useful than another agent capability.