When AI Agents Go to Work: What Happens to Us?
Alibaba is deploying AI agents to millions of merchants. OpenAI is building autonomous researchers. Here's what it means for you.
Something changed this month. AI stopped being a tool you use and started becoming a worker you employ.
Alibaba announced it's deploying a "digital workforce" to millions of merchants on Taobao and Tmall. OpenAI's chief scientist said they're building a fully automated AI researcher that can tackle multi-day problems without human help. These aren't research projects or pilot programs — they're production deployments at scale.
This shift from AI as assistant to AI as autonomous worker is happening faster than most people realise. And the consequences will reshape how we think about work, business, and society.
What's an AI Agent?
An AI agent is software that can take actions on your behalf, not just answer questions. Instead of "write me an email," you give it "handle customer inquiries" and it decides what to do, takes actions, and reports back. It can operate for hours or days without human input, making decisions, trying approaches, and correcting itself when things go wrong.
The Deployment That Changed Everything
In March 2026, Alibaba announced something that got less attention than it deserved: a "digital workforce" for millions of merchants on its e-commerce platforms.
What can these digital workers do? According to Alibaba:
- Customer service: Answering queries, resolving complaints, 24/7
- Pricing: Adjusting product prices in real-time based on demand
- Marketing: Distributing vouchers, optimising promotions
- Operations: Managing inventory, processing orders
Let that sink in: one to two years until human-digital collaboration is the "standard operating model." That's not a distant future — that's your next performance review cycle.
What OpenAI Is Building
While Alibaba deploys agents to merchants, OpenAI is building something more ambitious: a fully automated AI researcher.
In a March 2026 interview with MIT Technology Review, OpenAI's chief scientist Jakub Pachocki laid out the timeline:
- September 2026: "AI research intern" — handles small research problems autonomously for days
- 2028: Full AI researcher — tackles large, complex problems without human oversight
The goal is a system you can hand a problem to and come back when it's solved. Not a chatbot that answers questions, but a research lab in a data center.
The Key Difference: Autonomy
What makes agents different from previous AI?
Old AI (Chatbots)
You ask a question, it answers. One interaction. You drive the process. If you stop, everything stops.
New AI (Agents)
You give a goal, it figures out how to achieve it. Multiple steps, decisions, and corrections over hours or days. The AI drives the process. It keeps working when you're not there.
This shift — from "AI helps you work" to "AI does the work" — is what makes 2026 different from 2023.
What This Means for Workers
The Jobs in the Crosshairs
If your job involves routine decisions applied at scale, agents can do it. That's customer service, inventory management, basic accounting, scheduling, email triage, report generation, price optimisation, ad placement. These aren't rarefied skills — they're the bread and butter of millions of jobs.
But it's more nuanced than "AI takes jobs." The pattern emerging is:
1. The management layer is most exposed.
Amazon isn't cutting warehouse workers — it's cutting managers. Jack Dorsey explicitly said Block's 40% reduction was "driven by AI's growing capability, not financial difficulty." Middle management involves routing information and making routine decisions. That's exactly what agents do well.
2. Entry-level roles are disappearing.
The ILO/World Bank warned that AI could "close off pathways" to decent work for young people in developing countries. The jobs that used to teach you the ropes — clerical work, basic admin, customer service — are being automated before the next generation can enter them.
3. The productivity boost is real — but uneven.
If you have access to agents and know how to use them, you're suddenly 10x more productive. If you don't, you're falling behind. The productivity gains are concentrating among those already best-positioned to use them.
What This Means for Businesses
Alibaba's model reveals what agent deployment looks like at scale:
- 24/7 operation: Digital workers don't sleep, take breaks, or go on holiday
- Instant scaling: A merchant can go from 1 to 100 "workers" overnight
- Consistency: No variation in quality, no "bad days"
- Cost structure: Pay per task, not per salary
For small businesses, this is liberating. A solo entrepreneur can suddenly afford capabilities that required a team. But for workers, it's a direct substitution.
The Economics of Digital Workers
When a business can deploy an AI agent for $500/month that does work equivalent to a $50,000/year employee, the math changes. The business gets: no training costs, no benefits, no turnover, no management overhead. The worker loses: the job entirely.
What This Means for Society
Pachocki, the OpenAI chief scientist, was remarkably frank about the implications:
This is the core issue: AI concentrates capability. A small team with the right tools can now do what previously required a large company. That's good for efficiency. It's concerning for power dynamics.
The Long-Term Consequences
1. The "middle" disappears.
For decades, the middle class was built on middle-skill jobs — roles that required training but not elite education. Agents automate the middle. The question becomes: what fills it?
2. Speed of change accelerates.
When agents can do R&D, product development, and market testing autonomously, the pace of business change accelerates dramatically. Companies that take years to adapt may find they've already been left behind.
3. New forms of competition emerge.
A merchant on Taobao with AI agents can now compete with established businesses that needed teams to operate. Barriers to entry fall — but so do the jobs those barriers protected.
4. The meaning of "work" shifts.
When Pachocki says "nobody really edits code all the time anymore" and that his job is now "managing a group of Codex agents," he's describing a fundamental change. Work becomes supervision of AI workers rather than doing the work yourself.
What Should You Do?
The honest answer: no one knows for certain. But some patterns are emerging:
Learn to manage agents.
The skill of the next decade isn't "doing the work" — it's "directing the AI that does the work." Understanding how to prompt, guide, and evaluate agent output will be as fundamental as computer literacy was for the previous generation.
Focus on what agents can't do.
Agents are bad at: building genuine human relationships, exercising judgment in novel situations, taking responsibility for outcomes, understanding context that wasn't explicitly provided. These become more valuable, not less.
Don't assume your job is safe because "AI can't do X yet."
The gap between "AI can't do X" and "AI does X" keeps shrinking. In March 2026, we saw autonomous research announced. In September 2026, OpenAI expects AI research interns. By 2028, they expect fully autonomous researchers. The timeline is measured in months, not decades.
Pay attention to who owns the agents.
The real power isn't in the technology — it's in who controls it. When Alibaba deploys agents to millions of merchants, Alibaba decides what those agents do, how they evolve, what they prioritise. Digital workers aren't neutral tools; they're products shaped by companies with their own interests.
The Bottom Line
March 2026 may be remembered as the month agent deployment went from experiment to scale. Alibaba's digital workers aren't a demo — they're a product. OpenAI's autonomous researcher isn't a concept — it's a roadmap with dates attached.
The technology is no longer "coming soon." It's here. The question is whether we're ready for what comes next.
Key Takeaways
Agents ≠ Chatbots: Agents act autonomously for hours or days. Chatbots respond to prompts.
Deployment is real: Millions of merchants will have AI workers by end of March 2026.
Management layer most exposed: Information routing and routine decisions are exactly what agents do well.
Power concentrates: A data center can now do what required a large organisation.
The skill shift: From "doing work" to "managing AI that does work."