Interview
How AI agents will reshape global business | The Economist
- Brett Taylor, co-founder of CIRA and Chair of OpenAI, identifies the Summer 2022 release of OpenAI's "Dolly" demo (specifically the generation of an avocado chair) as the turning point where he realized computers could move from rule execution to creative generation.
- Taylor predicts that by 2025, every company will utilize an AI agent for customer-facing interactions, shifting the paradigm from human-led digital support to autonomous agent resolution.
- The current market is described as an "early innings" phase where high frustration stems from the lack of off-the-shelf solutions, forcing companies to build custom agents from raw models similar to the 1995-1998 era of expensive, bespoke website development.
- Taylor anticipates a maturation over the next four to five years where specialized vendors will sell pre-built agents for specific departments (e.g., auditing, customer service) rather than selling raw model components.
- A key challenge in AI deployment is the non-deterministic nature of models, where identical prompts can yield different outputs, making traditional testing and guarantee of robustness difficult.
- Taylor argues that AI should be managed using "defense in depth" and procedural controls rather than waiting for perfection, comparing AI hallucinations to human errors that are mitigated by monitoring and guardrails in regulated industries like finance.
- CIRA's proprietary agents are designed to identify themselves as AI and explicitly disclose the potential for mistakes to build trust with users.
- Customer satisfaction scores for CIRA's agents are reported to be consistently higher than previous human-led interactions, with clients noting benefits in multilingual support and infinite patience.
- Taylor cites an example where a telecommunications client reduced customer service costs from 10 euros to 1 euro per interaction, allowing the company to reinvest savings into reducing churn rather than simply cutting headcount.
- CIRA utilizes an "outcomes-based pricing" model, charging clients only when an agent successfully resolves a problem, while escalations to human agents are provided at no cost.
- Taylor posits that while the marginal cost of generating software will plummet, companies will continue to purchase software solutions rather than building them, valuing the "audit" and collective hardening of the ecosystem over the act of coding itself.
- Regarding the labor market, Taylor disagrees with the premise that jobs will vanish, suggesting instead that the economy lacks the imagination to see how new roles will form around AI, though the pace of required reskilling is historically unprecedented for white-collar workers.
- Taylor notes that employees who adopt AI fluidly can accelerate their careers, as the technology is new enough that internal talent can outperform external hires who lack specific organizational context.
- In management, Taylor uses AI as a "creative foil" to critique strategic documents rather than generating content, preserving his own deliberative thought process while leveraging AI for summarization and information triage.
- A forward-looking development Taylor expects is the ability for AI to triage email inboxes and manage communication during commutes, a capability he views as inevitable within the next few years.