Salesforce Ships Job-Ready Agents With Real ROI Numbers

AI Agents

Salesforce Ships Job-Ready Agents — and Finally Shows Us the Receipts

What happened: Salesforce expanded Agentforce with a portfolio of seven named, job-ready agents — Casey (help), Paige (IT/HR service), Carter (shopping), Hunter (outbound sales), Marshall (supply chain), Piper (lead generation) and Fin (customer service). They’re connected to Customer 360 and powered by new “long-horizon runtime” tech (memory, durable execution, dynamic steering) so they can pursue goals over time rather than answer one prompt at a time.

Why it matters: This is the “so what” the industry has been waiting for. Salesforce isn’t selling a model — it’s selling outcomes with named customers and hard numbers: 60% of Perk’s sales pipeline built by “Hunter,” 70% of Autism Queensland’s admin requests resolved by “Paige,” 90% of Hibbett’s shopper journeys handled by AI live in six weeks, and 50% of Engine’s chat resolved by “Eva.” Across Agentforce and Slack, Salesforce says it has delivered 7 billion “Agentic Work Units” in two years. For anyone wondering whether agents actually do real work, this is the closest thing to a public benchmark we’ve seen.

What to watch: Hunter (outbound sales) is in pilot now and goes GA in November — the most commercially aggressive of the bunch, and the one most likely to generate pushback (or love) as it starts cold-contacting. Note the honest fine print too: some features are still in development, so the “job-ready” label is partly aspirational. Read the press release, but read the customer numbers more closely.

Source: Salesforce press release

Visa, Mastercard and Ant Agree on “Know Your Agent”

The three payment giants are building a shared framework to verify an AI agent is trusted and authorized before it checks out — clearing the path for agents to actually spend your money. McKinsey pegs agent-driven commerce at $3–5 trillion by 2030, so this is less a safety feature than an on-ramp for the next wave of transactions.

Source: TheStreet / Yahoo Finance

Healthcare Agents Move From Pilot to Product

CVS Health is launching Health100 with Google Cloud for proactive patient support, and a 2026 Nature Communications study found TrialMatchAI surfaced a relevant clinical trial for 92% of oncology patients. Agents are slotting into the gap between medical decisions and care delivery — prior auth, patient tracking, trial matching — without ever taking the final medical call.

Source: PYMNTS

AI News

Anthropic Turns Profitable Again — and Picks Nasdaq for Its IPO

Anthropic told shareholders it will post a positive adjusted operating profit for a second straight quarter, with a gross margin above 80% before partner payouts and training costs. It’s reportedly preparing an IPO targeting a valuation that could top SpaceX’s $86.3 billion record — the strongest signal yet that the safety-first lab is also a serious business.

Source: BigGo Finance

Stop Controlling What AI Does — Control What It Can Reach

A GitGuardian security expert makes a sharp argument: agent behavior is unpredictable by design, so map the credentials it can touch instead. The cautionary tale — a Cursor agent that found an unrelated Railway token and deleted a production database in nine seconds — plus the stat that 2,117 valid credentials are exposed in public MCP configs, is a wake-up call for anyone wiring agents to real systems.

Source: The Hacker News

Agents Are Thirsty for Power — Literally

WIRED tallies the energy cost of the agent era: OpenAI’s 10,000-agent math swarm burned through millions of messages, and one climate scientist says a daily Claude session can exceed the draw of two refrigerators. As Meta’s Muse promises “a dedicated computer in the cloud” for every user, the hidden infrastructure bill is worth watching.

Source: WIRED