Salesforce and OpenAI both shipped agent products on September 11, 2026, and Anthropic followed on September 15. All three now let AI agents run full workflows, not just chat. If you work in sales, service, IT, HR or run a small business, expect your job to move from doing the task to reviewing what an agent already did.
Takeaways
- Salesforce made six of its seven named Agentforce agents generally available on September 11, 2026, including the customer service agent Casey, while the outbound sales agent Hunter stays in pilot until November 2026.
- OpenAI’s Agents API entered public beta on September 11, 2026, letting a single developer ship a long-running agent without building custom queues and retry infrastructure by hand, which raises the baseline skill expected of technical hires.
- Anthropic expanded Claude for Small Business to 43 workflows across 27 new integrations on September 15, 2026, and said the product has been installed more than 900,000 times since its May 2026 launch.
- Hunter, the Salesforce agent most directly aimed at sales pipeline work, has not shipped to general availability yet, so treat ‘job-ready’ as a description of intent covering agents at different stages, not a single proven product.
- Ask your manager this week whether any generally available Agentforce agent is scheduled to touch your team this quarter, since that answer will tell you more about your job than the press release will.
| Agentforce agents | 7 named, 6 generally available |
|---|---|
| Hunter status | Pilot now, GA November 2026 |
| Claude SMB installs | 900,000+ since May 2026 |
| Claude SMB workflows | 43 workflows, 27 new integrations |
What did Salesforce launch this week?
Salesforce introduced seven named AI agents inside Agentforce on September 11, 2026, covering customer service, IT and HR, ecommerce, sales, lead generation and supply chain work (Salesforce, September 11, 2026). Six of them are generally available now, meaning any customer running Agentforce can turn them on today rather than test them in a pilot. The seventh, an outbound sales agent named Hunter, is still in pilot and is not scheduled for general availability until November 2026 (Salesforce, September 11, 2026). That gap changes how you should plan for this: generally available agents get budgeted and put in front of paying customers immediately, while a pilot agent can still change before it ships. If your company uses Salesforce, the question worth asking is which of the six live agents was built to do part of your job, not whether Hunter’s pilot succeeds.
Who is Casey and what does Hunter do?
Casey is Salesforce’s customer service agent, generally available now, handling FAQs, returns and account questions across voice, SMS, WhatsApp and web chat (Salesforce, September 11, 2026). Hunter is the outbound sales agent, still in pilot, built to manage pipeline work from research through outreach over days or weeks using what Salesforce calls a long-horizon runtime (Salesforce, September 11, 2026). Salesforce describes the full lineup as “job-ready,” a vendor label, not an independent benchmark. Naming an agent after a person is a deliberate choice: it invites you to think of it as a colleague, not a script. For entry-level sales and service roles, Casey is the one already live, and Hunter is the one to watch once its pilot ends. Work that used to justify hiring a junior rep or support agent now has a named AI option, live or arriving soon, that a manager can turn on without adding headcount.
What can now run without an engineer?
OpenAI opened its Agents API to public beta on September 11, 2026, letting any developer spin up a long-running, tool-using agent with a single API call instead of building the queues, retries, sandboxing and monitoring by hand (InfoWorld, September 11, 2026). That infrastructure work used to be a specialized skill on its own. Lowering the bar means a single technical hire, or even a non-specialist with API access, can ship an internal agent this month instead of pitching a multi-quarter project. If you work in IT or engineering, this lowers the bar for everyone: agent-building stops being a rare line on a resume and becomes an expected baseline, the way basic scripting did a decade earlier. For a plainer walkthrough of what “tool use” and “long context” mean for agents like this, see Long Context, Tool Use And Computer Use Explained Plainly.
What changes for small business owners?
Anthropic expanded Claude for Small Business to 43 workflows across 27 new integrations on September 15, 2026, covering tools like Shopify, Stripe and Intuit QuickBooks, and said the product has been installed more than 900,000 times since it launched in May 2026 (Anthropic, September 15, 2026). In practice, Claude can now draft a client proposal, follow up on an unpaid invoice, qualify an inbound lead or reconcile the books, with every action staged for approval before it sends an email or moves money. For a solo founder or a two-person team, this removes tasks that used to eat evenings: chasing payment and rewriting the same proposal five times over. The approval step is the part worth protecting. Anthropic built the product to keep a human in the loop before anything external happens, and owners who skip reviewing those drafts lose the safeguard it was designed with.
Five roles closest to agent supervision, ranked by exposure
Not every job touched by this week’s launches is affected equally. Based on what Salesforce, OpenAI and Anthropic each shipped, here is where the pressure shows up first, ranked from most to least exposed.
- Entry-level sales development and customer support. Casey already runs support conversations end to end, and Hunter is set to do the same for sales pipeline work once its pilot ends in November 2026 (Salesforce, September 11, 2026).
- Bookkeeping and invoice follow-up at small firms. Anthropic’s 43 workflows target these repetitive back-office tasks directly (Anthropic, September 15, 2026).
- Internal tool and integration engineering. The Agents API replaces infrastructure that junior and mid-level developers used to build by hand (InfoWorld, September 11, 2026).
- IT and HR ticket triage. Paige, Salesforce’s IT and HR agent, is already generally available and resolves employee requests through Slack and portal tools (Salesforce, September 11, 2026).
- Supply chain and ecommerce coordination. Three more Salesforce agents, including Marshall and Carter, are also live now, but these roles sit lower here because supply chain and ecommerce work usually needs more specialized judgment before a business hands it full autonomy (Salesforce, September 11, 2026).
What don’t these vendor claims prove?
Salesforce’s own “job-ready” framing covers all seven agents, but the outbound sales agent, Hunter, is still in pilot and is not scheduled for general availability until November 2026 (Salesforce, September 11, 2026). That means the agent most directly aimed at sales pipeline work, arguably the closest to justifying fewer junior sales hires, has not shipped to the general customer base yet. None of Salesforce, OpenAI or Anthropic have published independent, third-party data on how their agents perform once thousands of businesses use them daily. Anthropic’s approval-before-action design (Anthropic, September 15, 2026) is a genuine safeguard, but it also confirms Claude for Small Business still assumes a human reviews every draft, a narrower claim than full autonomy. OpenAI’s beta label (InfoWorld, September 11, 2026) means the API itself is still being tested at scale. Three vendor launches in one week show investment and intent, not proof these agents already do the job as well as the person they aim to reduce.
What should you do this week?
Ask your manager directly whether any of the six generally available Agentforce agents, or a workflow like them, is scheduled to touch your team this quarter. If you work in IT or software, spend an hour with the Agents API documentation even if you never ship anything, because familiarity with agent infrastructure is becoming a baseline expectation rather than a specialty. If you run a small business, turn on one Claude for Small Business workflow and read every staged draft before approving it for a full week, so you learn where it gets things right and where it does not. For a broader read on how this week’s changes connect to where your job is headed, see This Week’s AI Price Collapse Moves Your Job To Supervision and What Half Of 2026’s AI Layoffs Mean For Your Job.
Prompts you can use
Paste these straight in. Change the parts in square brackets and nothing else.
You are a career strategist who specializes in how AI agents change specific jobs. I work as [your job title] at [company or industry], and my main weekly tasks are: [list 4 to 6 tasks]. This week, Salesforce made six of its seven named Agentforce agents generally available across service, IT/HR, ecommerce and supply chain work (with the sales agent still in pilot), OpenAI opened a public beta for building custom AI agents, and Anthropic expanded Claude for Small Business to 43 automated workflows. Based on my task list, tell me: (1) which of my tasks most resembles what these agents are built to do, (2) which of my tasks would be hardest for an agent to do without a human, and (3) one skill I should build in the next 90 days to move toward supervising agents rather than competing with them. Ask me clarifying questions about my role before answering if you need more detail.
Fill in your real job title, industry and task list before sending this, and treat the output as a starting point for a conversation with your manager, not a verdict.
You are an operations consultant helping a manager set up human oversight for a new AI agent. My team is about to turn on [name the agent or tool, e.g. an Agentforce service agent or a Claude for Small Business workflow] to handle [describe the task, e.g. inbound support tickets or invoice follow-ups]. We need a review process so a human checks the agent's output before it reaches a customer or affects money. Write a simple, numbered review workflow that covers: what the agent is allowed to do without review, what always needs human approval first, how often someone should sample-check the approved actions, and what to do when the agent gets something wrong. Keep it to steps a small team can follow day to day, not a corporate policy document. Ask me what tools my team already uses before finalizing it.
Replace the bracketed details with your actual agent and task before using this, since the review steps should match what the specific tool is allowed to do.
You are a small business advisor who is cautious about automation risk. I am considering turning on a Claude for Small Business workflow to handle [describe the task, e.g. invoice follow-up, lead qualification or proposal drafting]. Before I approve it, walk me through: what could go wrong if the agent's draft is slightly wrong and I approve it without reading closely, what a good approval habit looks like for someone who is busy, and one test I should run for two weeks before trusting the workflow with real customers or real money. Be specific and practical, not generic advice, and ask me about my business and customer base if that would change your answer.
This is written for someone new to Claude for Small Business’s 43 workflows; adjust the bracketed task to whichever workflow you’re evaluating.
Questions people actually ask
Will AI agents take entry-level sales and service jobs?
Salesforce’s Casey agent, live now, already runs customer service conversations end to end, and the outbound sales agent Hunter is scheduled to follow in November 2026. That does not guarantee job losses, but it does mean the tasks that used to justify junior hires now have a named AI option, live or on the way, that a manager can turn on.
What is the OpenAI Agents API in plain terms?
It is a public beta tool, launched September 11, 2026, that lets a developer create a long-running AI agent with one API call instead of building the queues, retries, sandboxing and monitoring by hand. It lowers the technical bar to ship an internal agent, which is why agent-building is becoming an expected skill rather than a specialty.
Is it safe to let Claude handle my small business invoices?
Anthropic’s Claude for Small Business stages every action, like sending an invoice reminder or paying a bill, for your approval before it happens, as of the September 15, 2026 expansion. That design keeps a human in the loop, but the safeguard only works if you read each draft before approving it.
What does ‘job-ready AI agent’ mean?
It is Salesforce’s own marketing term for seven named Agentforce agents announced September 11, 2026, six of which are live now while the seventh, Hunter, stays in pilot until November 2026. It is a vendor claim, not an independently verified accuracy rating, so treat it as a description of intent.
Should I learn to supervise AI agents instead of doing tasks myself?
These three launches point that direction: agents increasingly execute the task, while the paid role increasingly means reviewing and correcting what the agent produced. Learning to audit an agent’s output is becoming as valuable as knowing how to do the task by hand.
Sources
- Salesforcesalesforce.com
- OpenAIinfoworld.com
- Anthropicclaude.com
What happens next
Watch whether Hunter, Salesforce’s outbound sales agent, reaches general availability in November 2026 as scheduled, and whether Salesforce publishes any independent performance data once it does. Watch too whether OpenAI’s Agents API graduates out of public beta, which would indicate enterprises are trusting it with production workloads. If Anthropic’s approval-before-action design stays intact as Claude for Small Business scales past its current 900,000 installs, it could become a template other vendors copy for agent safety.
Take this further
Act as a blunt hiring manager who has read ten thousand resumes, not a career coach. I will paste my full resume and the job description I want. Rewrite the whole resume for that role, section by section, in this order: summary, experience, skills, education. Rules: every experience line leads with impact, not duty. Use bracketed placeholders like [8 percent] for any number I did not give you, and list at the end every placeholder I need to replace with a real figure. Keep it to one page of text. Plain formatting only, no tables or columns, so screening software can parse it. After the rewrite, tell me the three weakest claims that need evidence before I send this anywhere. My resume: [paste resume]. The role: [paste job description].
