"Who got help from AI and agent at work today?"

Every hand was raised.

I then asked, "Who among you received help exceeding 5 times from AI and agent at work today?"

This time, the raised hand still did not come down.

Sitting around the table are Zoom Chief Technology Officer, XD Huang, Figma Vice President of Engineering, Josh Clemm, Coursera Chief Technology Officer, Mustafa Furniturewala, ServiceNow Executive Officer, Glean Chief Technology Officer, T.R. “Vish” Vishwanath, Netlify Chief Executive Officer, Matt Biilmann, PagerDuty Head of Engineering, Manu Gurudatha, Yahoo Mail General Manager, Kyle Miller, Lime Chief Technology Officer, Jon Jenkins, Proofpoint AI Security Officer, Ravi Ithal, Temporal Co-founder and Chief Technology Officer, Maxim Fateev, Inworld Chief Executive Officer, Kylan Gibbs, Xero Head of Engineering, Diya Jolly, MinIO Chief Technology Officer, Ugur Tigli, Tribe AI Partner, Pooja Brown, Basis Co-founder, Creston Brooks, and Empirik Chief Executive Officer Kartik Chandrayana.

No one debates whether AI will destroy humanity. These are seasoned technology leaders who have been working on the front lines; what they truly care about are practical issues, such as: how can we ensure that AI and agents truly make an impact within the company?

Context

The word that kept recurring that night was: context.

The Huang of Zoom indicates that AI and agents will only truly become useful after understanding more information about the company, colleagues, meetings, and work.

“You just need the context,” XD joked, borrowing a phrase from the famous research paper “Attention Is All You Need” by AI.

The Vishwanath of Glean indicates agreement. He said that AI and agents often fail because they are fed with too little information, or worse, incorrect information. They will confidently provide what seem to be reasonable answers but are actually not correct. Nowadays, providing the correct contextual information for the AI system may be more important than creating smarter models.

This topic was mentioned repeatedly throughout the night.

The leader of ServiceNow stated that the internal AI product was deployed first. Initially, the company used AI to assist human customer service representatives, and then gradually allowed AI and agents to handle simpler customer requests on their own. Now, the customer satisfaction brought by the customer service team that collaborates with AI is even higher than that of the purely human team, as they solve problems more quickly and provide more personalized answers.

Netlify CEO Biilmann believes that writing code is rapidly becoming a commodified skill. The increasingly scarce skills are domain knowledge and software judgment. He says that companies may eventually reorganize their engineering teams around these capabilities, rather than just around "people who write code."

Recruitment

The recruitment sparked one of the hottest discussions of the evening.

Several executives described a strategy that one participant referred to as "dumbbell-shaped": on one end, they recruited engineers with extremely rich experience who were capable of questioning the results generated by AI; on the other end, they recruited young graduates who were naturally familiar with AI tools.

Figma Vice President of Engineering Clemm warns that engineers in the middle of their careers may face the greatest pressure, as organizational structures tend to become flatter and managers increasingly favor more senior employees.

The Gibbs indicates that startups do not have such leeway. His company would prefer to hire engineers with experience, as they are capable of solving complex technical problems that AI still cannot handle. He believes that if AI can already complete something in a day, then that is most likely not your competitive advantage.

Optional software

At some point, MinIO CTO Tigli showed me something that exemplified another theme of the evening quite well: how AI can change the software industry and make certain software types no longer necessary.

Earlier in the day, he asked Claude Code to create a presentation. It wasn't PowerPoint, nor Google Slides. What Claude generated was just a HTML file. After opening it on his laptop, he used AirDrop to transfer the same file to iPhone. It worked perfectly on both devices. No presentation software was used, no cloud services were involved; there was just one file.

Software disruption often happens in this way. It's not through a dramatic product launch, but rather when people quietly discover a simpler way to accomplish the same tasks. If AI can create presentations based on given prompts and package them into a portable HTML file, then how much more Microsoft PowerPoint or Google Slides would you really need?

Prediction

At the end of the dinner, I asked one last question: "In two years, what will we look back on with a smile and remember that we are still doing by hand?"

Several participants predicted that status meetings are likely to disappear, as AI and agents already know what everyone has completed. This also corresponds to the situation observed within OpenAI: some research teams no longer hold troubleshooting and Q&A sessions, as researchers increasingly rely on AI.

Subsequently, Lime, the Chief Technology Officer Jon Jenkins, presented one of the most eye-catching views of the evening.

He believes that software engineers are not necessarily the only ones who worry about whether AI will replace their jobs.

Within two years, perhaps the executives of C-suite – such as those sitting at the desks – will also begin to ask themselves the same unsettling questions.

Enterprise Software

AI Insider