AI News Digest, June 4: When an Autonomous AI Business Actually Turns a Profit

For two years the promise of agentic AI has come with a quiet asterisk. It demos well, but can it actually run something unsupervised without losing money? This week we got a real data point. An autonomous AI business run by Anthropic stopped bleeding cash and turned profitable. Meanwhile India put agentic AI inside the HRMS your payroll team already uses. NASSCOM moved to reskill 150,000 developers, and a data-security startup doubled to a $12 billion valuation in five months. The thread tying it together: AI agents are moving from “interesting” to “operational.” That shift lands squarely on HR and ops teams first.

An Autonomous AI Business Finally Made Money

Anthropic published Phase Two of Project Vend, its experiment in letting Claude run a small shop end to end. The verdict: the AI-managed store, which the model named “Vendings and Stuff,” is now profitable. The weeks of negative margins from Phase One are largely gone. (Source: Anthropic Research)

The setup changed in three concrete ways. First, the team upgraded to Claude Sonnet 4 and 4.5. Second, they opened two more locations, in New York and London, alongside San Francisco. Third, they added an AI “CEO” layer above the shopkeeper agent to set goals and catch obvious mistakes. (Source: The Decoder)

That last change mattered most. In Phase One the lone agent gave away free items and handed out discounts to anyone who asked. After the CEO layer went in, discounts dropped by roughly 80% and giveaways were cut in half. The fix was not a smarter model alone. It was structure around the model.

Why an autonomous AI business matters for your ops team

Here is the takeaway for anyone running HR or operations. The story of this autonomous AI business is not “AI can replace your team.” It is “a single agent left alone makes expensive judgment errors, and a second layer of oversight fixes most of them.” That is exactly the design question you face if you let an agent touch payroll runs, leave approvals, or vendor payments.

So before you hand an agent a real workflow, ask what your “CEO layer” is. Maybe it is a human approval gate on anything above a dollar threshold. Maybe it is a second agent that checks the first. Either way, the lesson from the most-watched autonomous AI business test so far is simple. Guardrails, not raw model quality, decide whether the thing loses you money. Teams already building AI agents for HR should design that oversight layer first, not last.

Agentic AI Moves Into the HRMS Itself

India’s greytHR launched NAVOS, an agentic AI assistant built directly into its HRMS. It turns plain-language intent into action across Payroll, Core HR, Leave and Attendance, Performance Management, and Recruitment. It also ships free across all paid plans, with no extra setup. (Source: greytHR)

This is the same agent pattern as the autonomous AI business story, just pointed at HR. greytHR serves more than 34,000 organizations and processes over $23 billion in payroll a year, so the reach is real. If your HR software vendor has not announced something similar, expect it soon. The practical question for you: which routine actions would you actually trust an agent to execute versus only draft? Think running payroll, or pulling a department report. Start there before you flip anything live. For the workflows you do automate, AI payroll automation is the obvious first candidate because the rules are well defined.

India Bets on Reskilling 150,000 Developers

NASSCOM launched “AI Code Sarathi,” a national program to upskill 150,000 Indian developers. The goal: move them from dabbling with AI tools to shipping production code. The training is free, hands-on, and aligned with the IndiaAI Mission. (Source: Analytics Insight)

For HR and talent leaders, this is a signal about where the skills bar is moving. The program treats AI as a teammate across the full development lifecycle, not a novelty. So if you hire engineers in India, “uses AI tools” is fast becoming table stakes rather than a differentiator. Update your screening rubric accordingly. Then look at how you close the AI skills gap in HR on your own team.

AI Data Security Becomes a $12 Billion Bet

Cyera raised $300 million at a $12 billion valuation, roughly doubling its worth in five months. The round was led by Evolution Equity Partners and pushes the AI data-security firm’s total raised past $2 billion. (Source: SiliconANGLE)

Why should an HR leader care about a security raise? Because the money is chasing one problem: companies are pouring sensitive data into AI systems faster than they can track it. HR data, salaries, performance notes, and personal records, is some of the most sensitive of all. As you roll out agents like NAVOS or your own copilots, one question gets sharper: where does employee data actually flow? Tracking that stops being an IT footnote. It becomes part of your rollout plan.

Quick Hits

• Microsoft built its own model family. At Build 2026, Microsoft’s Superintelligence team unveiled seven in-house MAI models. The lineup is led by MAI-Thinking-1, its first reasoning model, and signals a move to lean less on OpenAI. (Source: CNBC)
• Europe pushed for tech independence. On June 3 the European Commission proposed a tech-sovereignty package. It arrives just as the AI Act’s transparency rules take effect on August 2, 2026. From then, companies must disclose when people interact with AI. (Source: European Commission)
• OpenAI aimed Codex at office work. OpenAI shipped six role-specific Codex plugins for jobs like data analytics, sales, and finance. Together they bundle 62 business apps so agents can work across your existing tools. (Source: TechCrunch)

If this week has you rethinking where agents fit in your stack, that is the right instinct. The teams that win will not be the ones that adopt fastest. They will be the ones that wire in oversight before they hand over the keys. Asanify’s global HRMS and payroll platform is built around that approval-first model. Automation speeds your team up without quietly running off on its own.

Autonomous AI Business: FAQ

What does Anthropic’s Project Vend prove about an autonomous AI business?
It shows an AI agent can run a small shop profitably, but only after adding an oversight layer. A solo agent gave away too many discounts and free items in Phase One. Adding an AI “CEO” to set goals cut discounts by about 80% and made the business profitable.

Can an AI agent run HR payroll on its own yet?
Not fully on its own, and you should not let it. Tools like greytHR’s NAVOS can execute payroll and HR actions from plain-language requests. But the safe pattern keeps a human approval gate on high-stakes runs. Let agents draft and prepare, then approve before anything goes live.

Why does AI data security matter for HR teams?
HR systems hold salaries, performance records, and personal data, which is among the most sensitive information a company stores. As HR teams adopt AI agents, that data moves into more systems. Knowing where it flows is now part of any responsible AI rollout, not just an IT concern.

Not to be considered as tax, legal, financial or HR advice. Regulations change over time so please consult a lawyer, accountant  or Labour Law  expert for specific guidance.