Overview of AI Agents Like OpenClaw Are Here. How Can You Use Them?
This episode of The Wall Street Journal's What's New Sunday (host Alex Osola) features tech reporter Isabel Busquets explaining the current state and near-future of agentic AI — systems that can act on users’ behalf. The conversation covers what "agents" mean in practice, where they're already used (coding and customer service), the promise of orchestration frameworks like OpenClaw, business economics and monetization questions, security and liability risks, and implications for legacy SaaS companies.
Key topics discussed
- Definition and scope of "AI agents" and "agentic" behavior
- Current commercial use cases (coding agents, customer-service agents)
- OpenClaw: open-source orchestration enabling full personal-assistant behavior
- Business economics: costs (tokens) vs. productivity gains
- Security, hallucination, and liability risks from agents acting autonomously
- Investor and market implications, including pressure on legacy SaaS companies
What an "AI agent" means
- An agent is an AI that can do something for you — when asked to perform actions (book reservations, make purchases, handle customer requests) it becomes "agentic."
- Industry lacks a single formal definition; the term is widely used and sometimes overhyped.
How companies are using agents now
- Coding agents: Many startups and incumbents provide agentic tools (OpenAI, Claude-related tools, Replit, Cursor, etc.). Engineers use them frequently; usage is metered via tokens.
- Customer service: Agents power automated responses for order tracking, issuing loyalty cards, handling routine calls — an evolution from non-AI phone trees.
OpenClaw and next-generation agents
- OpenClaw is an open-source orchestration framework (recently acquired by OpenAI) that enables broader "personal assistant" capabilities by connecting agents to accounts, logins, and data.
- With access to many systems, agents can perform multi-step workflows autonomously — but this creates substantial security risks.
Business economics and monetization
- Token costs vs. employee salary: companies track token consumption and compare it to engineer costs. Some engineers may use tokens that "cost more" than salary, but productivity multipliers (10x–50x) can justify the expense.
- ROI remains hard to quantify across enterprises; many companies view agent capability as a necessary investment even without clear short-term monetary returns.
- Potential measurable effect: rising revenue per employee as AI augments workforces; longer-term workforce "right-sizing" is possible but uncertain.
Risks and limitations
- Hallucinations: agents can produce incorrect results or actions (e.g., erroneous refunds, incorrect promises).
- Security: giving agents access to accounts and credentials can enable data breaches or unintended destructive actions (deleting files, exposing credit cards).
- Liability and trust: unclear who is liable for an agent’s actions (user, company, or operator of the agent). Agentic shopping raises new legal/financial questions.
- Human-in-loop: many companies retain human oversight for approval on agent actions to mitigate risks.
Implications for SaaS and investors
- Agentic AI increases competitive pressure on legacy SaaS providers (the so-called "SaaSpocalypse" concern), particularly where coding agents reduce dependence on expensive legacy tools.
- Established SaaS companies (e.g., Salesforce) are investing heavily in agents to stay relevant; speed and quality of agent development will be key differentiators.
Notable quotes
- Isabel Busquets: “An agent is an AI that can do something for you.”
- Reported paraphrase from NVIDIA CEO Jensen Huang at his developer conference: “Every company in the world today needs to have an OpenClaw strategy” — underscoring how vendors view agentic systems as foundational.
Actionable takeaways for businesses
- Pilot first, with constrained permissions: start with limited-access agents in low-risk workflows.
- Keep a human in the loop for high-risk decisions (refunds, financial transactions).
- Implement strict access controls, auditing, and monitoring for agent activity.
- Track token and usage costs explicitly and measure productivity gains against those costs.
- Reassess legal/contract terms and liability policies for agent-driven actions and purchases.
- Invest in agent orchestration carefully; balance capability gains against security exposure.
Conclusion
Agentic AI is moving from niche tooling to potentially transformational orchestration frameworks that can automate multi-step tasks across systems. The benefits (productivity improvements, simplified workflows) are attractive but offset by hard-to-quantify ROI, serious security and liability risks, and market disruption pressures on legacy software providers. Organizations should proceed with targeted pilots, robust controls, and clear metrics to assess value and safety.