AI Agents: Context, Key to Their Utility

Seventeen tech leaders gathered in San Francisco shared their daily and varied use of AI agents. They all emphasized the importance of rich context to make these tools effective. The discussions revealed changes in customer support, recruitment, and software usage, along with predictions about the disappearance of certain tasks within two years.

Meetings and Functions in Question in Two Years

Several leaders present believe that follow-up meetings could largely disappear in two years, as AI agents will be capable of tracking task progress. Internally, some research teams at OpenAI have already stopped organizing office hours to solve problems, with researchers relying more on AI. Jon Jenkins, CTO of Lime, suggests that in two years, even high-level executives may wonder if AI can perform their jobs.

Widespread Daily Use of AI Agents

During the dinner, all participants indicated they had used at least one AI agent in their workday, and all had used more than five. Seventeen tech leaders were gathered, illustrating the widespread adoption of these tools among product and technical managers.

For AI Agents to Be Useful, They Must Know the Company

The central theme of the discussion was the importance of context. XD Huang, CTO of Zoom, believes that AI agents will only become truly useful if they understand the company, colleagues, meetings, and work. He summarized this point jokingly: “Context is all you need,” referencing the paper "Attention is All You Need." T.R. "Vish" Vishwanath, CTO of Glean, emphasized that agents often fail due to a lack or poor quality of information, producing plausible but incorrect responses. He believes that providing the right context at work could now be more critical than improving the models themselves.

Customer Support: AI-Enhanced Human Assistance Outperforms

Ashraf Karim from ServiceNow described a gradual deployment of AI: initially used to assist human support, it was later entrusted to AI agents to handle simple customer requests independently. According to him, human agents aided by AI now achieve higher customer satisfaction scores than those working without AI, thanks to faster problem resolution and more personalized responses.

Recruiting Differently: Critical Seniors and AI-Native Juniors

Several leaders discussed a so-called "barbell" recruitment strategy, combining highly experienced engineers capable of questioning AI-generated work and recent graduates comfortable with these tools. Josh Clemm, VP of Engineering at Figma, warned that mid-career engineers might face the most pressure, as managers increasingly favor senior profiles. Kylan Gibbs from Inworld clarified that startups cannot always apply this strategy and rely heavily on experienced engineers to solve technical problems that AI does not handle. He noted that if AI can accomplish a task in a day, that is probably not where the competitive advantage lies.

No PowerPoint or Slides: A Presentation Generated in HTML

Ugur Tigli, CTO of MinIO, asked Claude Code to generate a presentation in the form of a single HTML file, which worked seamlessly on both a laptop and an iPhone after being transferred via AirDrop. This demonstration shows that a presentation task can be accomplished without dedicated software or cloud services, illustrating how AI allows for bypassing certain established software usages.

Who Was at the Table

Present were: XD Huang (Zoom), Josh Clemm (Figma), Mustafa Furniturewala (Coursera), Ashraf Karim (ServiceNow), T.R. "Vish" Vishwanath (Glean), Matt Biilmann (Netlify), Manu Gurudatha (PagerDuty), Kyle Miller (Yahoo Mail), Jon Jenkins (Lime), Ravi Ithal (Proofpoint), Maxim Fateev (Temporal), Kylan Gibbs (Inworld), Diya Jolly (Xero), Ugur Tigli (MinIO), Pooja Brown (Tribe AI), Creston Brooks (Basis), and Kartik Chandrayana (Empirik). The meeting took place in San Francisco.